Hundreds of billions of dollars are committed. Construction crews can’t break ground. That gap — between what the tech giants have pledged to build and what is actually being built — is the defining infrastructure story of 2026.
AWS, Google, Microsoft, and Meta have collectively committed to more than $650 billion in AI infrastructure spending in 2025 and 2026 alone (Bloomberg). Zoom out to the full North American picture through 2030 and the forecast hits $710 billion. And yet, somewhere between 30% and 50% of all US data centre projects planned for 2026 are facing delays or outright cancellation. Do the maths and you get a potential 7 GW capacity shortfall — the equivalent of 30 to 70 large AI training facilities that will not be delivered on schedule this year.
Two forces are driving this and most coverage treats them as separate stories. The first is organised, legally sophisticated community resistance at a scale the hyperscalers genuinely didn’t see coming. The second is a transformer and switchgear shortage severe enough to delay projects even after they win their permitting battles. This article is one entry point into the data centre community revolt reshaping AI infrastructure timelines across the country — the macro numbers that make what’s happening in individual states legible as signal rather than noise.
The $710 billion figure is the North American data centre capital expenditure forecast through 2030, cited in a New York Times analysis from March 26, 2026. The $650 billion figure is Bloomberg’s verified tally of what the Big Four are committing to in 2025 and 2026 specifically. These are signed commitments and announced capital allocation plans. They are not trend-line projections.
What they don’t reflect is how much of that capital is actually being converted into energised infrastructure.
Nixon Peabody‘s May 7, 2026 “Data Center Site Selection Strategy Update” documents the scale of the problem. More than 140 local community groups have mobilised across the US since early 2025, mounting legal challenges and moratorium campaigns that have placed $60 billion or more in a blocked or delayed state. In early 2026 alone, 26 new AI-related legal cases were filed — and the pace is accelerating, not stabilising. Data Center Watch recorded $156 billion blocked across full-year 2025.
Nixon Peabody frames this shift through what they call the “power-plus-permission” model: grid access is necessary but no longer sufficient. Regulatory readiness, legislative durability, and community approval are now co-equal requirements alongside power access — replacing the old “energy-first” standard where a permit was expected to follow grid access as a matter of course.
The collision is structural. A March 2026 Gallup survey found 71% of Americans oppose the construction of AI data centres in their local area — higher than opposition to nuclear power plants, which sits at 53%.
The headline figure: approximately 30–50% of all US data centre builds planned for 2026 face delay or outright cancellation, according to Sightline Climate analysis corroborated by Bloomberg. Of the roughly 12 GW of US data centre capacity initially slated to come online in 2026, only approximately 5 GW is under active construction. That arithmetic gap is the 7 GW capacity crisis — which we cover in the next section.
Full cancellations are less common than extended permitting timelines or rezoning reversals, but they are happening. The clearest documented case is the Natelli Apex project in Wake County, North Carolina. Developer Michael Natelli withdrew a planned data centre after sustained community pressure — abandoned outright due to opposition, not permitting denial.
North Carolina has become the national epicentre of the resistance movement. A community-mapped survey documents 21 data centre projects statewide at various stages of opposition. In Stokes County, a narrowly approved $10 billion project on a 1,844-acre rezoning faces a lawsuit from community members and environmental groups alleging commissioners improperly approved it.
Virginia carries the largest documented financial exposure. Data Center Alley, home to 600+ operational facilities, has seen local opposition cited as a contributing factor in the cancellation of at least 25 proposed data centre projects, according to AFS Law. The QTS Realty Trust Digital Gateway project near Gainesville — which at full buildout would have been the largest data centre campus in the world — was upheld for cancellation by the Virginia Court of Appeals on March 31, 2026, with a Virginia Supreme Court appeal pending. Courts, developers, and investors across the country are watching the ruling as a precedent.
The 7 GW capacity crisis is the direct consequence of the delay and cancellation wave. Of approximately 12 GW of US data centre capacity announced for 2026, only approximately 5 GW is under active construction — a gap of 7 GW. Source: Sightline Climate’s pipeline analysis, reported by Tech Insider on April 17, 2026 and corroborated by Bloomberg.
To put that in terms that mean something: at typical hyperscale densities of 100–300 MW per campus, 7 GW is equivalent to roughly 30 to 70 large AI training facilities missing from the 2026 schedule — each representing $1–4 billion in deferred capex.
If you want to keep tabs on the permitting battles driving these numbers, datacentertracker.org provides near-real-time tracking of community opposition actions by county.
Community opposition is the primary driver of data centre delays. But the transformer and switchgear shortage is a separate, compounding factor — and the two interact in ways that most coverage misses.
Lead times for large power transformers have gone from 12–18 months pre-2020 to 36–48 months as of 2026, approaching five years for the largest high-voltage units (Bloomberg). The supply crunch pushed Chinese transformer imports from fewer than 1,500 units per year in 2022 to more than 8,000 in 2025, as developers tried to route around domestic manufacturing backlogs.
Here’s how the two problems interact: community opposition extends permitting timelines, which shifts construction into periods of deeper supply constraint. A project that wins its rezoning battle in mid-2026 may then face a 36–48 month queue for the transformer needed to energise the facility. Winning the permit does not translate quickly to operational capacity.
Community opposition is national and bipartisan. Data Center Watch analyst Miquel Vila put it plainly: “There’s no safe space for data centres. Opposition is happening in very different communities.” Republican senators and progressive Democrats have both spoken out. “Stop the Steal” activists and DSA organisers have jointly mobilised in Michigan.
North Carolina is the epicentre. Elon University polling found 44% of state residents oppose data centre development, versus 24% in support.
Virginia has seen at least 25 project cancellations attributed in part to local opposition (AFS Law). The QTS Digital Gateway Court of Appeals ruling represents a national precedent for how far courts will go in upholding community opposition.
Maine passed the nation’s first state-level freeze — a moratorium on new facilities over 20 MW — which Governor Janet Mills then vetoed on April 24, 2026 (Reuters). Mills’s stated reason was not opposition to the principle: she’d have signed it had it exempted a specific project in Jay. A vetoed bill is not a failed signal. Legislative passage demonstrates appetite, and subsequent bills are likely.
Michigan has recorded at least 27 communities with moratoriums or restrictions, with Sterling Heights unanimously approving a one-year moratorium in February 2026.
It’s worth understanding the Home Rule vs. Dillon’s Rule distinction here, because it explains why opposition translates into policy very differently depending on which state you’re in. In Home Rule states, municipalities can enact moratoriums independently — any local government can act, which raises investor risk considerably. In Dillon’s Rule states, local authority derives from what the state legislature grants, making local moratoriums harder to enact. Virginia is Dillon’s Rule, which is why Loudoun County stated it lacks legal authority to impose a moratorium. Full treatment of this is in the North Carolina and Virginia cluster articles.
The jurisdictional variation above is the symptom. The underlying cause is a structural reset in how data centre projects get approved — and it happened fast.
The power-plus-permission model has replaced the old energy-first standard. Grid access is still necessary, but it’s no longer sufficient. The structural trigger was Loudoun County, Virginia — the world’s highest concentration of data centres — eliminating by-right approval in March 2025 and making public hearings mandatory for decisions that previously required only staff approval. Fairfax, Prince William, and Fauquier counties are following suit.
Nixon Peabody introduces a second key concept: “social licence to operate.” This is the community buy-in that projects must earn beyond regulatory permits. Social licence encompasses transparency commitments, early stakeholder engagement, and Community Benefit Agreements (CBAs) — the primary formal mechanism through which developers codify what they owe to host communities. If you’re still relying on pre-development NDAs, audit those practices now. NDA secrecy is the documented trigger most consistently converting passive concern into organised legal action. Residents who discover a multi-billion-dollar project was hidden from them for a year tend to mobilise fast.
Nixon Peabody’s 26 new early-2026 legal cases make the consequence clear for developers who haven’t adapted: legal challenges at the permitting stage, not after construction. For a complete overview of every dimension of the $710B collision — from legal mechanics to geographic epicentres to the CTO risk framework — see the full data centre community revolt resource.
The 7 GW shortfall translates directly to compute that cannot be provisioned — for AI inference, training workloads, or enterprise cloud services. And the regions facing the highest concentration of delayed or cancelled projects are not secondary markets. Northern Virginia — Data Center Alley — is home to AWS us-east-1 and Azure East US, among the most heavily used cloud availability zones in the world. North Carolina carries the second-highest documented opposition pressure.
When you’re evaluating cloud vendor roadmaps, the relevant question is not whether delays are happening — they are. The question is which regions face the highest concentration of delayed or cancelled projects, and whether a given hyperscaler’s stated capacity commitments are actually backed by physical construction progress.
For teams monitoring permitting developments that could affect regional cloud capacity: datacentertracker.org provides near-real-time tracking of community opposition actions by county. The full framework for translating these macro numbers into procurement decisions — what to ask your hyperscaler, how to build delay probabilities into capacity planning, and which regions carry the highest regulatory risk — is the subject of what the 7 GW shortfall means for cloud capacity planning.
140+ local community groups have mobilised across the US since early 2025 to block, delay, or litigate against data centre construction. Four grievances are driving the movement: water consumption (17 billion gallons in 2023), electricity cost-shifting to residential ratepayers, persistent cooling noise, and NDA secrecy that keeps residents in the dark until construction is imminent. The AI infrastructure boom of 2024–2025 brought industrial-scale projects to communities with no prior experience of data centres, faster than consultation processes could handle.
Full cancellations are less common than extended permitting timelines or rezoning reversals — the aggregate is 30–50% of 2026 builds facing delay or cancellation (Sightline Climate / Bloomberg). In investment terms: $156 billion blocked across full-year 2025 (Data Center Watch); $60 billion or more currently blocked (Nixon Peabody, May 2026). Documented full cancellations include Natelli Apex in Wake County, NC.
Of approximately 12 GW of US data centre capacity announced for 2026, only approximately 5 GW is under active construction — leaving a 7 GW gap equivalent to 30–70 large AI training facilities missing from the 2026 delivery schedule. Source: Sightline Climate pipeline analysis, reported by Tech Insider, April 17, 2026.
Four documented grievances: water consumption (millions of gallons daily per facility), electricity grid strain and cost-shifting to residential ratepayers, persistent noise from cooling systems, and NDA secrecy. NDA practices are the documented trigger most consistently converting passive concern into organised legal action — residents who discover a multi-billion-dollar project was hidden from them for a year tend to mobilise quickly.
Delayed or cancelled data centres represent compute that cannot be provisioned. The highest-pressure regions — Northern Virginia (AWS us-east-1, Azure East US) and North Carolina — host the most heavily used cloud availability zones. Regional capacity constraints in these markets are the most direct exposure for enterprise cloud customers.
$710 billion is the North American data centre capital expenditure forecast through 2030 (NYT, March 26, 2026). $650 billion is Bloomberg’s verified annual spend figure for the Big Four hyperscalers (AWS, Google, Microsoft, Meta) specifically for 2025–2026. Both are legitimate figures — they’re just describing different scopes.
Nixon Peabody’s term for the new site selection standard replacing “energy-first.” Grid access is necessary but no longer sufficient; regulatory readiness, community buy-in, and legislative durability are now co-equal requirements. A developer who secures grid interconnection and acquires land cannot assume a permit follows — they now face a political durability assessment and community engagement process before breaking ground.
datacentertracker.org provides near-real-time tracking of community opposition actions by county. Nixon Peabody’s May 7, 2026 “Data Center Site Selection Strategy Update” covers the legal landscape. Data Center Watch (10a Labs) publishes annual blocked-investment tallies.
In Home Rule states, municipalities can enact moratoriums independently — any local government can act, which raises investor risk. In Dillon’s Rule states, local authority derives from the state legislature, making moratoriums harder to enact locally. Virginia is Dillon’s Rule; Loudoun County has stated it lacks authority to impose a moratorium. Full treatment is in the North Carolina and Virginia cluster articles.
The two delay vectors compound each other. Opposition extends permitting timelines, shifting construction into periods of deeper supply constraint. A project that wins its rezoning battle in 2026 may still face a 36–48 month queue for a transformer — transformer lead times have extended from 12–18 months pre-2020 to 36–48 months currently, approaching five years for the largest units. Winning the permit does not translate quickly to operational capacity.
Maine passed the nation’s first state-level data centre freeze — a moratorium on new facilities over 20 MW — before Governor Janet Mills vetoed it on April 24, 2026 (Reuters). Mills called a moratorium “appropriate” and would have signed it had it exempted a specific project in Jay. Legislative passage signals regulatory appetite; subsequent bills are likely.
Ranking in AI Responses — What the Data Actually ShowsIf you’re optimising for position one on the assumption that gets you into Google’s AI Overviews, the data says you’re wrong about four times out of five. Two independent studies — one peer-reviewed, one from a benchmarks firm — arrive at the same finding: the sources cited in AI responses and the pages ranking in traditional organic search are largely different populations.
Two numbers anchor everything in this article: a Jaccard similarity of 0.17 between AI Overview citations and top-10 organic results (arxiv 2604.27790v1), and a 17% AIO-to-top-10 overlap figure from Conductor AEO/GEO Benchmarks. This is an honest evidence synthesis — confidence levels are stated explicitly and you’ll find calibrated uncertainty here, not advocacy. This article is the evidence base for the AI search zero-click crisis — part of a broader series covering the full scope of the shift.
The research base on AI citation predictors is thin relative to the volume of practitioner claims. Before using any finding to drive strategy, it helps to know what kind of claim you’re actually looking at.
Three sources with disclosed methodologies anchor the evidence hierarchy. The arxiv study (2604.27790v1) analysed 11,500 queries from the ORCAS dataset — real-user queries from Bing search logs — measuring source overlap between SERP results and AI Overview citations, with data collected in December 2025. The Ahrefs study sampled 300,000 keywords using Google Search Console data to measure CTR per keyword position. The Conductor AEO/GEO Benchmarks are large-scale but not independently peer-reviewed.
💡 Rank-Biased Overlap (RBO) is a ranking similarity metric that weights top positions more heavily than Jaccard — useful when citation order matters, not just which sources appear.
What the Ahrefs study actually measured: CTR differential at position one with and without an AI Overview present. The 0.073 → 0.016 CTR collapse is the result. What it does not measure: whether those ranked pages were cited in the AI Overview — which is a different question entirely.
Most other practitioner guidance on “what gets you cited” is pattern-matching on observed outcomes, not causal research. Where that applies in this article, it will be stated.
Jaccard similarity measures set overlap on a scale from 0 to 1. A score of 0 means two sets share nothing. A score of 1 means they’re identical.
Here’s a concrete example. Google returns 10 URLs in organic search for a query. An AI Overview cites 10 sources for the same query. If Jaccard similarity is 0.17, roughly two sources appear in both lists. The other eight in the AI Overview come from somewhere else entirely.
The arxiv study found average Jaccard similarity of 0.17 for real-user ORCAS queries. The authors state it plainly: “On average, only 18% of the sources returned by either the AIO or traditional SERP will be retrieved by both search engines.” Conductor’s finding from a different methodology lands in the same place — only 17% of AIO-cited URLs come from pages ranking in the traditional top 10.
Ahrefs’ longitudinal analysis of 863,000 keywords adds the trend line: AIO citations from Google’s top-10 pages dropped from 76% in mid-2024 to 38% by early 2026. For AI assistants like ChatGPT, Gemini, and Copilot, only 12% of cited links rank in the top 10. The trend is moving away from overlap, not towards it.
The practical implication: optimising exclusively for traditional SERP ranking addresses approximately 17% of the AI Overview citation pool. The architectural mechanism that explains these citation patterns is covered in detail in the AI Overviews mechanism article.
Not all citation predictors are equally supported by evidence. Here is the hierarchy with confidence levels based on data quality, not marketing appeal.
Entity clarity — Strongly evidenced (arxiv 2604.27790v1). Generative search engines are less likely to cite sources that lack consistent entity signals — even high-traffic, well-ranked sites. Inconsistencies reduce AI citation accuracy by 30–40%. Content with 15 or more connected named entities shows 4.8x higher selection probability.
Passage-level content structure — Strongly evidenced (multiple GEO studies). AI retrieval systems extract specific passages, not entire pages. Q&A format sections show a +25% citation impact; clear H2/H3 hierarchy shows +40% citation likelihood. The mechanism is straightforward — the system extracts a passage that answers a query, it doesn’t summarise a document.
Schema markup — Moderately evidenced (AIVO, Lumar, arxiv). Pages with three or more schema types are 13% more likely to be cited. Organisation, Product, Person, and FAQ schemas are most consistently associated with AI citation. The evidence is correlational, not causal.
Topical authority over domain authority — Moderately evidenced (ZipTie/Conductor). Domain Authority explains less than 4% of AI citation variance (r²=0.032). Topical authority — depth of coverage within a subject area — shows a correlation of r=0.41. Pages ranking sixth to tenth with strong topical authority are cited 2.3x more often than pages ranking first with weak topical authority.
Google-Extended blocking reduces citations — Strongly evidenced (arxiv). Blocking the Google-Extended crawler statistically reduces AIO citation rate. Playwire data shows publishers who blocked AI crawlers experienced a 23.1% monthly traffic decline in early 2026 with no reduction in AI citations.
First-party data, author markup — Practitioner hypothesis. The Princeton/Georgia Tech GEO study found adding statistics improved citation visibility by 41% on Perplexity. Real effect sizes, but not yet verified across other platforms.
What the data does not support: a clean checklist that guarantees AI citation. Citation is probabilistic and query-dependent. What GEO and AEO practice should look like based on this evidence is covered in the AEO and GEO disciplines article.
The measurement range here is enormous, and the endpoints reflect genuinely different things.
AIVO (tryaivo.com) reports AI search visitors convert 23x better than traditional organic visitors. Pew Research‘s tracking of 900 US adults across 68,879 searches found users clicked AI Overview links less than 1% of the time. These figures aren’t contradictory — AIVO’s 23x is self-reported attribution from a self-selected, high-intent population; Pew measured observed behaviour across a general population performing ordinary searches. Neither is wrong. They answer different questions.
The Ahrefs data provides the most reliable signal on click volume: position-one CTR collapsed from 0.073 to 0.016 when an AI Overview is present — a 78% reduction based on 300,000 keywords.
For B2B tech businesses, Semrush‘s 10M+ keyword study found AI Overview saturation highest in Computers & Electronics at 17.92%. At that saturation level, whether you’re cited or not has real stakes regardless of per-citation click rate.
The honest summary: the evidence supports “AI citations matter” without supporting “AI citations produce X conversions for my specific business.” How agent search ranking diverges from human search citation is explored in the AI Search for Agents article.
Attribution decay is the mechanism that makes AI search damage invisible in standard analytics. GA4 strips attribution from visits that arrive via AI-answered queries because no referral click occurs. The RADM (Revenue Attribution Decay Model) framework from DigitalApplied estimates 35–52% of branded-query attribution is lost before the click stage even begins.
The framework identifies three decay stages. Pre-click decay: zero-click SERPs answer the query before any click is possible — the content contributed to a decision, GA4 records nothing. Click-path decay: a prospect asks ChatGPT about vendors, gets an answer naming your product, then types your URL directly — GA4 records a direct visit, the AI citation is invisible. Post-click decay: users who previously googled your brand name now just ask the AI directly.
RADM’s sample worksheet shows an Organic + AI Citations channel with a reported ROAS of 1.2x and a decay-adjusted ROAS of 5.4x — model outputs, not guarantees, but the direction is consistent. Google-Agent adds a further gap: it ignores robots.txt and doesn’t appear in analytics as an identifiable source.
Before you cut content investment on the basis of declining GA4 traffic, the AI search zero-click crisis context explains why that number may be misleading you.
Only 16% of brands systematically track AI search performance. Without monitoring you can’t distinguish between two completely different problems: “our content is not being cited” (a content problem) and “we are being cited but can’t measure the pipeline contribution” (an attribution problem). Both require different responses.
The rotating-prompt harness. Identify 20–50 queries relevant to your topic clusters. Run them against Google AI Overviews, ChatGPT, Perplexity, and Gemini weekly — citation patterns can shift materially within a month. Record citation frequency, citation position, and citation share by platform.
Primary tools. Profound tracks citation frequency by prompt and query volume with platform-level breakdowns. Peec AI provides AI visibility tracking across platforms as a category alternative.
Platform prioritisation. Google AI Overviews and AI Mode are the highest-priority citation surfaces by query volume. ChatGPT search is second: 1.2 billion referral clicks in Q4 2025. Perplexity is smaller but cited for high-intent traffic. Citation logic differs across platforms — Google favours brand-managed websites; ChatGPT favours encyclopedic depth; Perplexity is community-weighted — with only 10–15% overlap in citations across platforms. For most B2B businesses, Google AI Overviews is the right place to start.
Ranking optimisation (traditional SEO) and citation optimisation (GEO/AEO) are partially overlapping but fundamentally different disciplines. A content team treating them as the same thing is addressing 17% of the AI citation pool and leaving the rest to factors it isn’t optimising for.
💡 GEO (Generative Engine Optimisation) and AEO (Answer Engine Optimisation) are the emerging disciplines focused on making content retrievable and citable by AI systems — distinct from traditional SEO.
The evidence-supported investments in priority order: entity clarity (Knowledge Graph resolution), passage-level structure (declarative opening sentences, self-contained sections), structured data (Organisation, Product, Person, FAQ schemas), and topical coverage completeness — the r=0.41 correlation is the strongest single quantitative predictor in the data.
The attribution problem is as urgent as the citation problem. Run RADM Step 1 — baseline funnel segmentation — before making any content investment decisions based on GA4 traffic figures alone.
Set up a rotating-prompt harness for 20 target queries. Check your robots.txt for Google-Extended blocking. Implement schema markup on your top-performing content. Then build the citation frequency dataset that lets you test whether changes are working — because the published research cannot answer that for your specific business. That is the evidence base for how to respond to the zero-click shift.
No. The arxiv ORCAS study found Jaccard similarity of 0.17 between AIO citations and top-10 organic results. Conductor shows only 17% of AIO citations come from URLs in the traditional top 10. Ahrefs confirms the share dropped from 76% in mid-2024 to 38% in early 2026. Position 1 is not a reliable predictor.
A score of 0.17 means roughly 17% of items are shared between two sets. For any given query, approximately four out of five sources cited in an AI Overview do not appear in the top-10 organic results. The two lists operate on different selection logic.
No. The arxiv study found blocking significantly reduces AIO citation rate even though Google may still access content through other crawlers. Playwire data shows publishers who blocked AI crawlers experienced a 23.1% monthly traffic decline in early 2026 with no reduction in AI citations.
They measure different things. AIVO’s 23x is self-reported by visitors who said they arrived from AI search — a self-selected, intent-filtered population. Pew Research measured observed click behaviour across 68,879 ordinary searches by 900 US adults. Different populations, different funnel stages. Neither is wrong.
RADM (Revenue Attribution Decay Model) is a seven-step framework from DigitalApplied that quantifies how much pipeline credit is lost when AI search answers queries before a trackable click occurs. It identifies pre-click, click-path, and post-click decay stages. Its sample worksheet shows an Organic + AI Citations channel with a reported ROAS of 1.2x and a decay-adjusted ROAS of 5.4x — which explains why content contribution is systematically undercounted in standard dashboards.
Google AI Overviews and AI Mode are highest-priority by query volume. ChatGPT search is second (1.2 billion referral clicks in Q4 2025). Perplexity is smaller but high-intent. Citation logic differs fundamentally across platforms with only 10–15% overlap. Monitoring all three using Profound or Peec AI is more reliable than guessing.
Organisation, Product, Person, and FAQ schemas are most consistently named in GEO/AEO practitioner research. Pages with three or more schema types are 13% more likely to be cited. The arxiv entity-resolution finding provides the strongest mechanistic rationale: if the AI can’t identify the entity behind content, citation becomes less probable.
E-E-A-T appears to function as a binary eligibility filter: 96% of AI Overview citations come from sources with strong E-E-A-T signals. Among eligible pages, E-E-A-T does not appear to be a gradient boost. Its direct effect on AI citation beyond the gatekeeper function is not empirically verified.
Search Google for your target queries with AI Overviews active and check whether your domain appears in the citation sources. For systematic monitoring, Profound and Peec AI automate this across multiple queries and platforms.
No. Google-Agent is a user-triggered fetcher launched in March 2026 that ignores robots.txt and does not appear as an identifiable referral source in GA4 or CRM attribution pipelines. It is one component of the attribution decay problem the RADM framework is designed to surface.
The AI Search Zero-Click Crisis — What Every Content-Led Business Needs to KnowYour website traffic is falling. Your Google rankings haven’t changed. If that sounds familiar, something structural has shifted in how search works — and it is not coming back.
It’s called zero-click search. That’s when Google answers a query inside its own interface before the user has any reason to visit your site. 60% of all Google searches end without a click — a figure from SparkToro and Similarweb that rises to 83% when an AI Overview is present and 93% in Google AI Mode. Those three numbers are quietly demolishing the assumption that good content leads to traffic. This article introduces the crisis, explains the key concepts, and maps out the seven articles in this series.
There are two kinds of zero-click. The first is navigational — the user was always going to type a URL. The second is AI-driven — the query gets answered inside Google before anyone clicks anywhere. The second kind is the one growing fast: from 58.5% in 2022 to 64.8% by 2026. 60% Zero-Click — The Numbers That Rewrite Every Traffic Assumption pulls apart all three figures and explains exactly what each one measures.
When an AI Overview appears above organic results, position-1 click-through rate drops by 58% according to Ahrefs — without your ranking changing at all. AI Mode goes further: it decomposes a query into up to 16 parallel sub-queries across Google’s web index, Knowledge Graph, Shopping Graph, Maps, and YouTube, which is why its zero-click rate hits 93%. AI Overviews Kill 61% of Clicks — The Mechanism Behind the Drop covers the full architecture of how this works under the hood.
B2B technology queries trigger AI Overviews up to 70% of the time. Informational queries — “what is”, “how does” — are the most exposed category, with 99.9% triggering an AI Overview according to Ahrefs. HubSpot’s 70–80% organic traffic decline between 2024 and 2025 has become the standard warning sign for content-led B2B SaaS businesses. If that pattern matches your metrics, you need to read Publisher Revenue Crisis — When Traffic Dies, What Survives, which covers case studies in detail.
Three disciplines now coexist and you need to understand all of them. SEO optimises for traditional SERP rankings — still necessary, but no longer sufficient on its own. GEO (Generative Engine Optimisation) optimises for AI citation in Google AI Overviews, ChatGPT, and Perplexity. AEO (Answer Engine Optimisation) targets direct-answer boxes and featured snippets. The reason all three matter: only 17% of AI Overview citations come from pages in the top 10 organically, which means your SEO work alone is not getting you into AI results. AEO and GEO — Two New Disciplines, One New Optimisation Target covers where each discipline starts and how they differ.
AI Overviews answer queries on-platform. Referral traffic falls. CPM revenue follows. OpenAI’s crawl-to-referral ratio sits at 1,700:1 — crawlers index your content without sending visitors back. The IAB Tech Lab puts the annual publisher ad revenue already displaced at $2 billion. And if you’re thinking about blocking AI crawlers, there’s a catch: blocking Google-Extended correlates with reduced AI Overview citations even when your content is still technically accessible. There’s no clean exit. Publisher Revenue Crisis — When Traffic Dies, What Survives covers the full revenue picture. For the other side of this equation — how Google’s search monetisation strategy is evolving while publisher revenue collapses — Article 4 analyses the structural paradox.
In April 2026, Sundar Pichai said that “Search would be an agent manager” — AI agents completing tasks on behalf of users rather than retrieving information for them to act on. This is not a distant scenario. The Universal Commerce Protocol, launched in January 2026, already enables checkout within Google’s surfaces without a user ever visiting a merchant’s site. Google’s 2026 capital expenditure is $175–185 billion — they are not experimenting. AI Search for Agents — When the Searcher Isn’t Human covers what the agent vision means for your infrastructure.
An Arxiv study 2604.27790v1 covering 11,500 queries found less than 20% overlap between organic rankings and AI Overview citations. Ranking and AI citation are different activities with different predictors. What actually predicts AI citation: entity clarity in Google’s Knowledge Graph, passage-level content structure, and topical depth. Ranking in AI Responses — What the Data Actually Shows covers the full evidence on what you need to change.
Pick the article that matches what you need to decide:
It’s when a query is answered within the search interface and no external website is visited. 60% of Google searches end this way. For content businesses the number that really matters is 83% — that’s what happens when an AI Overview is generated.
Not even close. Informational queries are the most exposed — 99.9% trigger an AI Overview according to Ahrefs. E-commerce sits at 3.2%, because Google pulled back after AI responses weren’t converting.
AI Overviews appear within standard search results and produce an 83% zero-click rate. AI Mode is a separate conversational interface — US-only since March 2026 — that produces a 93% zero-click rate. Same direction, just further along.
SEO improves your SERP ranking. GEO improves the probability of your content being cited in an AI-generated response. Only 17% of AI Overview citations come from pages in the top 10 organically — so your ranking work and your citation work need to be treated as separate problems.
AEO targets direct-answer boxes and featured snippets. GEO handles complex AI summaries. Both share core tactics — structured data, declarative content, E-E-A-T — but they have different primary targets.
An AI Overview reduces position-1 click-through rate by 58% according to Ahrefs, without changing the ranking itself. Stop monitoring position alone and start watching impressions and CTR in Google Search Console.
He is the CEO of Google and Alphabet. His April 2026 framing of Search as an “agent manager” — a system that completes tasks rather than retrieves information — is the clearest public signal that the zero-click shift is structural and intentional, not a side effect.
Google captures value inside its own surfaces — sponsored AI Overviews, Shopping Graph ads, and commerce via UCP — making its revenue increasingly independent of whether anyone visits a publisher’s site.
Probably not without thinking it through carefully. The arxiv study found that blocking Google-Extended correlates with reduced AI Overview citations even when your content is still technically accessible. There’s no clean exit. Article 5 covers the dilemma properly.
Standalone options include Profound, Peec AI, Scrunch AI, and Otterly AI. If you’re already using a major SEO platform, Semrush AI Visibility Toolkit and Ahrefs Brand Radar are the integrated options worth looking at.
Treat it as structural. Google’s $175–185 billion 2026 capex, Pichai’s agent manager framing, and the trajectory from 58.5% in 2022 to 93% in AI Mode all point in one direction. Regulatory proceedings may eventually impose some constraints, but the downside of not adapting far exceeds the downside of adapting prematurely.
AI systems extract at the passage level — paragraphs and sentences, not whole pages. The practical implication: structure each section of your content as a self-contained answer, not as part of a continuous essay.
AI Search for Agents — When the Searcher Isn’t HumanIn April 2026, Sundar Pichai said something worth reading slowly: “Search would be an agent manager.” Not a faster search engine — a platform that orchestrates AI agents to complete tasks on behalf of users, rather than returning links for humans to follow.
This is the second-order disruption that the AI search zero-click crisis set in motion. The first-order disruption was AI answering queries for humans. The second-order: the searcher is no longer human. AI agents are doing the searching, the evaluating, and the transacting — and when the searcher is a machine, optimisation changes with it.
Three infrastructure shifts define what that actually means. WebMCP (Web Model Context Protocol) turns websites from passive sources into callable tools. Google-Agent is a user-triggered fetcher already sending traffic to your site since March 2026 — invisible in your analytics. And the Universal Commerce Protocol (UCP) lets AI agents complete checkout within Google’s surfaces without a user ever visiting your website.
When a human searches, they click, dwell, and bounce. When an AI agent searches, none of that happens. It calls functions, extracts data, completes actions, and reports back. There is no click to measure.
So the things that matter for optimisation are completely different. Structured data completeness and entity graph resolution matter more than keyword density. API surface and callable endpoints matter more than page design. Machine-readable pricing matters more than persuasive copy.
This creates a two-audience problem. GEO and AEO remain necessary — human users still exist and still search. But the infrastructure layer now needs to serve a second audience with entirely different requirements.
Three dimensions organise the challenge: discoverability (can the agent find your platform?), accessibility (can it extract what it needs?), and actionability (can it complete the task without human intervention?). Most organisations are still working through the first two. Actionability is where agentic infrastructure actually begins.
WebMCP (Web Model Context Protocol) is a proposed web standard co-developed by Google and Microsoft engineers through the W3C. It was previewed in February 2026 and is currently available in Chrome Canary behind a feature flag — not in Chrome stable as of May 2026.
Here is the practical shift. Instead of scraping your page and inferring structure, an AI agent can call a defined function on your site and get a structured result back — the same way a developer calls an API. Your website goes from something an agent interprets to something an agent uses directly.
In practice, a WebMCP-enabled pricing page exposes a capability an agent can query with specific parameters and get a machine-readable response. Same goes for demo requests, documentation search, and feature lookups. The B2B scenarios the standard is targeting include industrial quoting, vendor qualification, and wholesale ordering.
WebMCP is not retrofit-friendly. Design what you would expose now, and implement when Chrome stable ships. That is on a 12-month horizon. For how GEO and AEO disciplines apply when the searcher is an AI agent, the entity and structured data work required for human-facing AI search and agent-facing discovery converge at the same foundation.
Google-Agent was added to Google’s official fetcher list on March 20, 2026. It is not a crawler — it is user-triggered. When a human directs an AI agent to complete a task, Google-Agent may visit your site as part of executing it.
There are a couple of things about it that will surprise you. First, it ignores robots.txt. Google classifies it as “user-triggered,” which puts it outside the scope of the robots.txt protocol — so conventional blocking does not work. Second, it does not appear in your analytics. Google-Agent does not execute JavaScript, produces no standard session signals, and does not match bot-detection patterns. Your site is already receiving this traffic. It is not in your dashboards.
What to do about it: audit your server logs for the Google-Agent user-agent string — not GA4. Cloudflare logs and Microsoft Clarity’s Bot Activity report provide ongoing monitoring. And keep this in mind: Google-Agent does not render JavaScript and does not retry on timeout, so critical content needs to be in raw HTML with sub-second response times. For how ranking in AI responses differs for agent-facing content — including what the empirical data shows about the content characteristics that predict citation when an agent is the retriever — the evidence picture is meaningfully different from traditional SERP ranking.
The Universal Commerce Protocol (UCP) is an open standard launched at the National Retail Federation conference in January 2026, co-developed by Google and Shopify, with Etsy, Wayfair, and Target as founding partners. By April 2026, Amazon, Meta, Microsoft, Salesforce, and Stripe had joined.
What it actually enables: AI agents can discover products, negotiate fulfilment conditions, and complete checkout within Google’s surfaces — without the user visiting the merchant’s website. The Shopping Graph is the product data layer. UCP is the commerce equivalent of WebMCP: where WebMCP lets agents call website functions, UCP lets agents complete purchases.
It is live with B2C commerce partners. Not yet live for B2B SaaS as of May 2026 — so plan for it, do not implement it yet.
The planning implication is worth sitting with. Pricing written as narrative copy is invisible to agent-mediated discovery. An agent evaluating SaaS vendors queries a structured data source and returns a ranked result. No queryable pricing data means no appearance in that comparison. The same gap applies to demo flows — an agent expects a structured endpoint, not a form.
Pichai named 2027 as “an important inflection point” and estimated only 0.1% of the world is living this future today. 2026 is the diffusion year. He also confirmed that Antigravity — Google’s internal agent orchestration platform, formerly called Jet Ski — had already been deployed to the Search team. Not a consumer announcement; an internal infrastructure signal pointing in one direction.
The vision produces a three-layer stack you need to be building toward.
The Knowledge Graph is the entity layer. If you are not a resolved entity there, AI agents cannot surface you in response to category queries — you do not enter the candidate pool, regardless of content quality.
The Shopping Graph is the product layer. Merchant Center accuracy and attribute completeness are now agent-discoverability prerequisites, not just advertising inputs.
WebMCP and UCP are the capability layer — turning your web presence from a passive source agents scrape into an active resource agents invoke.
All three layers: discoverable, accessible, actionable. Miss layer one and you are invisible, full stop.
The agentic search conversation has been dominated by B2C e-commerce. The B2B implications are the gap — and they are structural. BrightEdge data shows AI Overviews appear on 82% of B2B technology queries. The pressure is already here.
Pricing page architecture. Most B2B SaaS pricing pages are optimised for human persuasion — deliberate friction, no public enterprise pricing, “contact us” CTAs. When an agent evaluates vendors, a page that requires human navigation is not evaluated. It is skipped.
Documentation discoverability. HealthTech and FinTech platforms carry extensive technical documentation. Under agentic search, this becomes an agent-callable resource — but only if it is structured at the passage level. Q&A format and declarative section headings with self-contained answers perform best. Dense prose does not. For documentation teams, this is a structural rewrite.
Entity graph as prerequisite. For an agent to surface a B2B SaaS platform in response to a category query, the platform must be a resolved Knowledge Graph entity. No resolution means no candidate pool entry — binary. Organisation schema, consistent brand descriptions across LinkedIn, G2, Crunchbase, and Wikipedia. Months of work, and you need to start now.
FinTech regulatory overlay. Agent-readable pricing data must also satisfy compliance requirements around financial product disclosure. That is legal and infrastructure coordination from the architecture phase.
Three tiers, no false urgency.
Tier 1: Do this now
Audit your server logs for the Google-Agent user-agent string — not GA4. Set up Cloudflare log monitoring or Microsoft Clarity’s Bot Activity report as ongoing metrics.
Audit your Knowledge Graph entity status. No Knowledge Panel means starting with Organisation schema, a consistent brand presence across LinkedIn, G2, Crunchbase, and Wikipedia, and sameAs links.
Review your pricing pages and documentation for passage-level extractability. Each H2 should work as a standalone, declarative answer. That is immediate GEO and AEO value, and it is the groundwork for agent-readiness at the same time.
Tier 2: Design work (~12-month horizon)
Enable the WebMCP feature flag in Chrome Canary and get across the API. Identify which functionality your platform would expose as callable tools. Design the interface. Do not ship yet.
Tier 3: Plan for it (12-18 months)
Read the UCP specification at ucp.dev. Design machine-readable, parameter-queryable pricing structures. Build the agent-callable demo endpoint alongside your existing human-facing form — both audiences, one infrastructure.
Pichai named 2027 as the inflection year. The preparation window is 2026.
WebMCP (Web Model Context Protocol) is a web standard that lets websites expose functions as callable tools for AI agents. Instead of scraping your page, an agent calls a defined function and gets structured data back. It is available in Chrome Canary now; stable availability is expected within 12 months.
No. Googlebot is Google’s traditional scheduled web crawler. Google-Agent, added March 20, 2026, is user-triggered — it visits when a human directs an AI agent to complete a task. It ignores robots.txt. You cannot block it through conventional means.
Not in production. WebMCP is only in Chrome Canary and still evolving. Understand the API, identify what callable functionality you would expose, and design the architecture now to avoid a costly retrofit later.
Not yet live for B2B. UCP launched with B2C partners in January 2026. Start designing machine-readable pricing and capability data structures now so you are ready when the B2B layer activates. The spec is public at ucp.dev.
Google-Agent does not execute JavaScript — which is how GA4 collects data. The only reliable detection method is server log analysis, filtering for the Google-Agent user-agent string.
Through the Knowledge Graph. If your company is not a resolved entity there, AI agents cannot reliably surface it. Building Knowledge Graph presence requires consistent Organisation schema, accurate brand descriptions, and third-party presence on G2, Crunchbase, LinkedIn, and Wikipedia. It takes months.
Antigravity (formerly Jet Ski) is Google’s internal agent orchestration platform, deployed to the Search team before Pichai’s April 2026 interview. It is the internal model for what consumer Search becomes — a platform that manages agents rather than returns links.
AI Overviews synthesise information for a human who decides what to click. Agentic search uses AI agents to execute tasks on a human’s behalf — the agent searches, evaluates, and acts. The optimisation requirements differ because the agent is not deciding what to click; it is completing a task.
Blocking Google-Agent in robots.txt has no effect — it ignores robots.txt by design. Audit your baseline traffic via server log analysis and understand what it is accessing before making any policy decisions.
Organic traffic — a human clicking a result and landing on your page — becomes a smaller proportion of total search-driven value. The agent executes the task without a click. Attribution models built on click-based traffic will increasingly misrepresent your web presence’s actual value.
The Shopping Graph is Google’s product and merchant database. UCP uses it as its product data source. Shopping Graph accuracy and completeness is a prerequisite for UCP participation, not an afterthought.
This article is part of the broader shift from human to machine search interfaces documented across this cluster. For the optimisation disciplines that underpin agent-ready content — structured data, entity building, and passage-level extractability — see the companion piece on AEO and GEO. For how ranking in AI responses differs when the searcher is an agent rather than a human, see Ranking in AI Responses.
Publisher Revenue Crisis — When Traffic Dies, What SurvivesHere’s the paradox defining 2026 for anyone who publishes on the internet: AI bot traffic is surging — Playwire reports a 300% year-on-year increase in AI crawler requests — yet publisher ad revenue has collapsed. Bots don’t click links, don’t load articles, and don’t trigger the pageview events that CPM advertising depends on. And increasingly, neither do the humans, because AI Overviews answer the question before the click ever happens.
Chartbeat data from the Reuters Institute 2026 report shows Google search referrals to 2,500+ news sites dropped 33% globally in twelve months. The IAB Tech Lab puts sector-wide advertising revenue loss at $2 billion annually. Those are not projections. That’s what’s happening right now.
This article is part of our comprehensive guide to the AI search zero-click crisis, where we examine what the structural shift in search means for every content-led business. Here, the focus is narrower: what has happened, who got hit hardest, why blocking AI crawlers doesn’t work the way you’d expect, and what business models are actually surviving. For the underlying data, the zero-click statistics behind this revenue collapse has the causal chain in detail.
AI Overviews now appear in 51.5% of real-user Google queries. When an AI Overview answers the question on-platform, the referral doesn’t happen. Ahrefs found that position-one organic click-through rate fell from 1.41% to 0.64% when an AI Overview is present — a 54% effective click reduction for the top-ranking result.
The named cases make the aggregate numbers concrete:
Meanwhile, AI referral traffic remains negligible. All AI platforms combined account for less than 1% of publisher page views. The bots crawl at scale. They send almost nothing back.
The collapse isn’t hitting everyone the same way. Small publishers have lost approximately 60% of Google search referrals; large publishers are down approximately 22%. The difference isn’t really about traffic — it’s about business model architecture.
Large publishers have diversification levers that small operations simply don’t have. Subscription revenue, direct advertising relationships, branded events, and established brand authority all provide insulation that CPM-based advertising can’t.
💡 CPM advertising (cost per mille) pays publishers a fixed rate per thousand ad impressions — meaning revenue is directly proportional to page views. When referral traffic disappears, CPM revenue disappears with it.
Small publishers — niche content sites, travel blogs, independent journalism — depend almost entirely on CPM advertising funded by Google search referrals. And the queries they dominated are precisely the ones where AI Overviews are most prevalent: informational queries, long-tail questions, question-format searches.
The Planet D is the extreme version of this story. A travel blog entirely dependent on questions like “best things to do in Lisbon” or “is it safe to travel to Thailand” — exactly the kind of queries AI Overviews now answer directly. Once those referrals stopped, there was no revenue floor. It ceased publication.
The New York Times is the counter-example. Subscription revenue means traffic decline doesn’t translate directly to revenue decline. NYT’s willingness to litigate (NYT v. OpenAI) signals something too: large publishers with leverage are demanding compensation rather than quietly accepting extraction.
Publishers have reached for robots.txt. 79% of top news sites now block at least one AI training bot this way. The problem is that blocking doesn’t work as expected — and in some cases makes things measurably worse.
One technical detail matters here: Google-Extended isn’t a separate crawler. It’s a control token signalling whether content can be used for Gemini AI training. The actual crawling happens through standard Googlebot, so blocking Google-Extended won’t even show up in your server logs.
The arxiv finding (arXiv 2604.27790v1) is the key result: websites that block Google-Extended are significantly less likely to be cited in AI Overviews, even though Google still has full content access. The Playwire/Rutgers-Wharton finding compounds this: publishers who blocked all AI crawlers via robots.txt saw total traffic decline 23% with no corresponding reduction in content use by AI systems.
So the trap has three exits, and all of them are bad:
There is no clean answer to the “block or allow” question. The News/Media Alliance (NMA) formal demand letter to Common Crawl in April 2026 is the industry’s attempt to escalate beyond individual robots.txt decisions — but it’s a precursor to potential litigation, not an enforceable demand.
Cloudflare calculated the crawl-to-referral ratio by dividing total HTML requests from AI user agents by total HTML referrals sent back. The June 2025 results:
That 1,700:1 figure is the value extraction dynamic in a single number. Publishers bear the infrastructure cost of serving crawler requests. They receive almost nothing in return.
Common Crawl is the upstream layer. OpenAI trained GPT-3 on Common Crawl archives; Google used the C4 subset for what became Bard. Publisher content that wasn’t gated in the pre-LLM era was included without consent or compensation. That ship has sailed, but it explains why publishers are furious.
The April 2026 European Parliament research briefing identified five specific gaps in EU law — no transparency on AIO source selection, no right to fair remuneration for AI-generated summary use, no enforceable remedy as of May 2026. That last point is the operative one. For context on how Google’s own revenue fared through all of this, how Google’s revenue grew while publishers’ collapsed covers the monetisation model in detail. This dynamic — where the platform extracting value continues to grow while content creators bear the losses — is central to understanding the AI search zero-click crisis at its full scope.
Subscriptions have the clearest evidence. The New York Times has 11+ million subscribers and revenue that doesn’t evaporate when search referrals drop. The New Yorker reached record revenue and subscriber numbers in 2025. And 76% of commercial publishers surveyed by Reuters Institute now say subscription and membership is their biggest revenue focus.
Direct audience relationships are the structural equivalent for publishers without paywall leverage. Substack grew 40% year-on-year. Morning Brew‘s newsletter-first model generates 5–7x higher revenue per reader than search-dependent sites. The publishers who are surviving had already stopped treating search referrals as a given before the collapse accelerated.
Brand authority matters too. Publishers cited in AI Overviews as authoritative sources gain a brand signal rather than a click. For subscription businesses, that signal drives direct sign-ups. For CPM advertising, citation without a click produces no impression and no revenue — which is why subscription-based publishers are far better positioned to benefit from AI citation than CPM-dependent ones.
One claim worth treating with scepticism: AIVO (tryaivo.com) reports that visitors arriving via AI search convert at 23x the rate of organic search visitors. AIVO sells AI search analytics — a direct commercial interest in demonstrating AI search visitors are high-value — and the figure hasn’t been independently verified. The direction is plausible, but don’t treat 23x as an established benchmark.
For adaptation strategies that work with AI Overviews rather than against them, GEO and AEO as adaptation strategies for surviving AI search covers the practical implementation.
Yes — and in some verticals, the exposure is actually more acute.
AI Overviews appear in 88% of health-related queries per Semrush. If your business is in HealthTech and you depend on informational Google traffic, you’re more exposed than most publishers. The query types that B2B SaaS, EdTech, and HealthTech companies have built content strategies around — “marketing strategy,” “sales techniques,” “project management tips” — are precisely the informational queries most vulnerable to on-platform AI summarisation.
The structural difference is that SaaS, EdTech, and HealthTech companies measure success in leads and conversions, not ad revenue. That changes things. When an AI Overview cites your company as an authoritative source, organic CTR is 35% higher than for uncited results on the same query. That’s a measurable signal benefit even without the click.
So the practical question for non-publishers isn’t “how do we protect ad revenue” but “how do we ensure our content-led pipeline survives?” The answers — direct community, email lists, brand authority, niche expertise that earns AI citation — translate directly from publisher survival to SaaS content strategy. The situations are more alike than they look.
The publisher crisis is not a media-industry problem you can observe from a distance. For the full scope of the AI-driven traffic collapse, the pillar page covers the complete picture. For the adaptation response, GEO and AEO as adaptation strategies for surviving AI search covers what’s actually working. And for those thinking further ahead, the agentic search layer as an emerging revenue model opportunity examines what happens when the searcher isn’t human at all.
Google’s Q4 2025 search revenue was $63 billion, up 17% year-on-year. Ads now appear alongside AI Overviews rather than alongside publisher content. When AI Overviews answer the query on-platform, publisher ad impressions disappear and Google captures the advertising value that used to flow through the referral chain.
“Going concern” is an accounting term requiring a company to disclose if there’s substantial doubt about its ability to continue operating for the next 12 months. It’s a formal regulatory disclosure with specific legal thresholds — not editorial hyperbole. It signals existential operational threat.
Google-Extended is a control token, not a separate crawler — blocking it doesn’t stop Google from crawling through standard Googlebot. But arXiv 2604.27790v1 shows the opt-out signal reduces AIO appearances even with content fully indexed. You can’t surgically block AI training while remaining visible in AI Overviews.
Cloudflare divided total HTML requests from AI user agent strings by total HTML referrals sent back. OpenAI’s ratio was 1,700:1; Google’s traditional search ratio is approximately 14:1; Anthropic’s was 73,000:1.
For CPM-advertising-dependent content sites: the economics have changed, and not in your favour. For brand-building, lead-generation, and subscription-supported content: the calculus is different. Being cited in AI Overviews provides a measurable 35% CTR boost. The strategic shift is from writing content that ranks to writing content that gets cited — arXiv 2604.27790 confirms that average Jaccard similarity between AI Overview sources and traditional SERP results is below 0.2, meaning traditional ranking doesn’t reliably predict AIO citation.
The April 2026 document is a research briefing — policy analysis, not enacted legislation. The March 2026 European Parliament resolution expresses political will rather than binding law. The December 2025 competition investigation into Google will take several years with appeals extending timelines further. Publishers should plan on a 2–3 year minimum before any enforceable EU remedy.
B2B SaaS companies that built pipeline through informational content ranking in Google face structurally identical dynamics. SaaS companies measure leads and conversions, not ad impressions, so AI citations (even without clicks) may contribute to brand-influenced pipeline in ways CPM-dependent publishers can’t monetise. The practical response is the same: diversify to direct channels, invest in brand authority that earns AI citations, and track AI citation share alongside organic traffic.
Average Jaccard similarity between AI Overview sources and traditional SERP results is below 0.2 across all query subsets, per arXiv 2604.27790. AIO citation and traditional organic ranking are largely decoupled — the foundational justification for treating AEO and GEO as disciplines separate from SEO.
Google AI Search Monetisation — Can Google Fix Zero-Click Without Killing Its BusinessSixty percent of all search queries now end without a click. When AI Overviews appear, that figure climbs to 83 percent. In AI Mode — Google’s full-page conversational interface — it hits 93 percent.
And yet Google Search revenue reached $63 billion in Q4 2025, up 17 percent year-over-year. Alphabet passed Apple in market capitalisation at $3.885 trillion. Its stock rose 65 percent in 2025, despite losing two antitrust cases.
So what’s going on? This is the AI search zero-click crisis — but told from the perspective of the company that engineered it. Google built zero-click search as a deliberate commercial mechanism, backed by sponsored placements inside AI Overviews, a commerce layer called the Shopping Graph, and something called the Universal Commerce Protocol. In this article we’re going to explain that business logic, walk through what Google’s May 2026 AI search updates actually signal, and trace the path toward what Sundar Pichai is calling the “agent manager” future.
The answer is structural: Google moved the advertising inventory inside the zero-click surface.
When a user searches and gets an AI Overview, they don’t click through to anyone’s site. But ads still appear — within or immediately below those AI-generated responses. Advertisers follow the eyeballs, and the eyeballs stayed on Google. Publishers lose the referral click and the ad revenue that click used to generate. Google keeps both.
Ads now appear in 25.5 percent of AI Overview results, up from 5.17 percent in early 2025 — a 394 percent increase in twelve months. Google VP of Ads Dan Taylor confirmed that AI Overview ads monetise at the same rate as traditional search ads. Paid link CTR in AI Overview environments has fallen from 13 to 6 percent (Seer Interactive). Real declines — but a publisher problem, not a Google revenue problem.
Publisher traffic tells the other side of it: organic visits fell from 2.3 billion in mid-2024 to under 1.7 billion by May 2025. As Jason Aten put it in Inc.: “Search isn’t losing to AI. It’s funding it.”
Two product surfaces get conflated constantly, so let’s be precise about this.
AI Overviews are AI-generated summaries embedded within standard search results pages alongside traditional blue-link results. AI Mode is a separate, full-page conversational interface that replaces traditional results entirely — and that’s where you get the 93 percent zero-click rate.
The monetisation layer runs through two mechanisms. First, standard sponsored placements within or adjacent to AI-generated responses. Second: Direct Offers, a pilot inside AI Mode that lets advertisers present exclusive discounts within AI-generated shopping responses. Petco, e.l.f. Cosmetics, and Samsonite are testing it. The user gets the answer and the offer without navigating anywhere.
Google Pay closes the loop: frictionless checkout within AI Mode for eligible US retailers. Discovery, answer, and transaction — all on Google’s surface, all without a click.
Organic CTR for position-1 pages drops 58 percent when AI Overviews are present (Ahrefs, December 2025). Direct Offers is experimental today, but it’s the template for where this is all heading. This connects directly to the publisher revenue crisis created by Google’s architecture.
Commerce-intent queries — “best noise-cancelling headphones under $300,” “running shoes for wide feet” — are both the most commercially valuable queries Google processes and the most at risk from zero-click erosion. AI Mode is built to answer them directly. So Google needs a structural defence, and it has one.
It’s called the Shopping Graph: over 50 billion product listings refreshing two billion times per hour. Amazon blocks OpenAI’s crawlers entirely — zero Amazon products appear in ChatGPT’s shopping responses. Google has indexed Amazon listings for years. That catalogue advantage isn’t going anywhere quickly.
The Universal Commerce Protocol (UCP) converts that data advantage into something Google has never had before: a transaction layer. Launched at the National Retail Federation conference in January 2026 with Shopify, Etsy, Target, and Wayfair as founding co-developers, UCP is an open standard that lets AI agents complete checkout without the user ever navigating to a retailer’s site.
On April 24, 2026, Amazon, Meta, Microsoft, Salesforce, and Stripe joined the UCP Tech Council. Amazon had declined at launch — UCP threatened to shift the default product search starting point away from its own platform. Three months later, it joined anyway. When the most closed platform in e-commerce starts playing by shared rules, you have your answer about whether agentic commerce infrastructure has arrived.
OpenAI’s competing Agent Commerce Protocol (ACP) launched September 2025. Walmart tested both. ACP-driven purchases converted at one-third the rate of Walmart’s own click-out transactions (per aNavigator, May 2026). Google charges no transaction fees for UCP — every AI agent that implements it queries the Shopping Graph, compounding Google’s data advantage as adoption widens.
On May 7, 2026, Google announced five changes, all oriented toward making links more visible: deep dives (expandable source sections), inline links (source URLs embedded in answer text), subscription content labels (paywalled content flagged with publisher subscription linking), expert advice surfacing (credentialed authors and review sites prioritised), and website previews (pop-up previews when hovering over inline links).
These are strategic signals, not product solutions. No public CTR recovery data exists. The structural logic of AI Overviews does not change because source links are more visually prominent.
What the updates reveal is Google’s trilemma in motion. Three pressures are pulling in different directions: users want direct answers (churning to ChatGPT or Perplexity is Google’s primary competitive threat); advertisers want clicks and conversion tracking; publishers want referral traffic. The May 2026 changes nudge slightly toward advertisers and publishers. The timing is not coincidental — the Ahrefs 58 percent CTR decline data emerged in December 2025, forming part of the zero-click data Google’s May 2026 updates are responding to, right alongside the EU Commission‘s Article 102 TFEU inquiry targeting AI Overviews. As eMarketer put it: “CTRs might not make a comeback, but brand visibility is getting a boost.” These changes are managing pressure, not reversing trajectory.
In an April 2026 interview on the Cheeky Pint podcast, Sundar Pichai said:
“Search would be an agent manager in which you’re doing a lot of things. I use Antigravity today, and you have a bunch of agents doing stuff. I can see search doing versions of those things, and you’re getting a bunch of stuff done.”
Antigravity — formerly called Jet Ski internally — is Google’s internal agent orchestration platform, deployed to the Search team the week of that interview. Not a roadmap item. Infrastructure Google’s own teams are already using.
The architectural shift: Search transitions from a lookup tool to a coordination layer that dispatches AI agents to complete tasks. You can think of it as three commercial eras: the Link Economy (find the page, send the traffic), the Answer Economy (synthesise the response, monetise the surface), and the Action Economy (complete the task, monetise the transaction). UCP, Antigravity, and the Shopping Graph are the plumbing of the Action Economy.
Google’s $175–185 billion 2026 capex — roughly 30 percent higher than Wall Street expected — is the financial signal. That is not experimentation. For what this means for how AI agents discover and interact with your platform, the agentic search infrastructure article covers the full picture.
Three proceedings are running simultaneously. None of them directly threatens Google’s AI search monetisation architecture.
DOJ (Judge Mehta, September 2025): Google found to have maintained an illegal search monopoly. Behavioral remedies only — distribution agreements capped, data-licensing obligations for rivals. No structural divestitures. Google appealed in January 2026.
EU Commission (Article 102 TFEU, December 2025): Targets AI Overviews on two grounds — using publisher content without compensation or opt-out mechanism, and using YouTube content to train AI models while blocking rivals from equivalent access. Enforcement timelines typically run 18–36 months.
UK CMA (October 2025): Designated Google under the Digital Markets and Competition Regime, covering AI Overviews and AI Mode. Currently in consultation phase.
Here’s the central point. Behavioral remedies address distribution exclusivity — the pre-AI monopoly mechanism. They do not constrain Antigravity, UCP, or the Shopping Graph, which are products, not contracts. The one material risk: if the EU mandates a functioning publisher opt-out from AI Overviews, Google’s content selection narrows. Think of this as a constraint on Google’s options, not a resolution of the structural problem.
Let’s state this plainly. Google is optimising for its own incentive structure, and the order in which it resolves competing pressures is not arbitrary. Users churning to ChatGPT is the primary competitive threat. Advertiser revenue is the revenue line. Publisher referral traffic is what remains after both are served.
The implication for content investment is structural, not cyclical. Content that trains Gemini, feeds the Shopping Graph, and populates AI Overviews serves Google’s infrastructure — but the traffic that content used to generate is retained by Google’s interface. The gap between those two outcomes is widening, and agentic commerce is accelerating it.
The practical reframe: the question is no longer “how do I rank?” but “how does my content get cited, and what commercial outcome does that citation produce?” The full analysis is in the publisher revenue crisis; agentic discoverability is covered in AI search for agents. For a complete view of the structural transformation of search and content, the series overview covers every dimension of this shift.
Why is Google’s ad revenue still growing if it’s sending less traffic to publishers? Google monetises zero-click surfaces directly through sponsored placements within AI Overviews and Direct Offers within AI Mode. Ads appear in 25.5 percent of AI Overview results at the same rate as traditional search ads. Publisher referral traffic declines; advertiser spend migrates to Google’s AI surfaces.
What is the difference between AI Overviews and AI Mode? AI Overviews are embedded within standard search results alongside traditional blue-link results. AI Mode is a separate, full-page conversational interface replacing traditional results entirely. AI Overviews produce a 58 percent CTR decline for top-ranking pages; AI Mode produces a 93 percent zero-click rate.
What is the Universal Commerce Protocol and why does Amazon joining matter? UCP is an open standard enabling AI agents to discover products and complete checkout within Google’s AI surfaces. Amazon joined the UCP Tech Council on April 24, 2026 — after initially declining. Amazon’s reversal signals UCP has moved from pilot to infrastructure.
What did Sundar Pichai mean by “agent manager”? In the April 2026 Cheeky Pint podcast, Pichai described Search’s future as an orchestration layer dispatching AI agents to complete tasks rather than returning links. Antigravity is the internal platform implementing this, deployed to the Search team the week of the interview.
What are Google’s May 2026 AI search updates and did they fix the zero-click problem? Five changes on May 7, 2026: deep dives, inline links, subscription content labels, expert advice surfacing, and website previews. No public CTR recovery data exists. These are responses to regulatory and advertiser pressure, not confirmed solutions to publisher referral decline.
What did the DOJ antitrust ruling change for Google’s AI search strategy? Judge Mehta’s September 2025 ruling imposed behavioural remedies — distribution exclusivity restrictions, data-licensing obligations — but no structural divestitures. Google’s AI search architecture was not directly affected.
What is Antigravity and how does it relate to Google Search? Antigravity (formerly Jet Ski) is Google’s internal agent orchestration platform. Pichai confirmed its deployment to the Search team in April 2026. It is the internal implementation of the “agent manager” model.
What is the Shopping Graph and why does it matter? The Shopping Graph is Google’s product database — over 50 billion listings refreshing two billion times per hour. It powers AI Mode commerce responses and UCP-enabled agentic checkout, giving Google a structural competitive advantage AI-native competitors cannot quickly replicate.
AEO and GEO — Two New Disciplines, One New Optimisation TargetSixty percent of searches now end without a click. When users do engage, it’s with AI-generated summaries — not individual pages. Organic CTR is no longer the right thing to be measuring at the top of your funnel.
Three terms have emerged to describe what comes next: SEO, GEO, and AEO. They’re not synonyms. They describe distinct practices with very different operational implications.
Conductor’s analysis of 17 million AI citations found only 17% overlap between AI-cited sources and traditional top-10 organic rankings. Five out of every six AI citations come from content that isn’t on page one. Ranking well doesn’t reliably predict whether you get cited. That’s a significant problem if you’re running the old playbook. This guide is part of our comprehensive look at the AI search zero-click crisis, where we examine what the data means and how to respond.
In this article we’ll explain how all three disciplines fit together, why entity building is the thing you need to get right first, and how to decide which platforms to prioritise. It’s written for people who need to brief a team or allocate budget — not implement schema markup themselves. To understand the zero-click data that makes these disciplines necessary, the statistics article in this series has the full breakdown. If you want the full background on the structural shift driving this new optimisation landscape, the pillar article has everything you need.
SEO is still the prerequisite. Pages need to be technically sound, indexable, and authoritative before AI systems can evaluate them at all. GEO and AEO don’t change that.
What’s changed is the relationship between ranking and visibility. For queries where an AI Overview appears, organic CTR has dropped 61%. There are businesses that have held their rankings and still lost 40–70% of their traffic in a single year.
Historically, 76% of AI Overview citations came from top-10 organic results. That number is now 38% and falling. For ChatGPT, 80–90% of cited pages rank outside the top 10. When 51% of B2B software buyers now start research with an AI chatbot more often than Google, SEO alone isn’t enough anymore.
Conductor analysed 17 million AI responses and 100 million citations. Only 17% of AI-cited content overlaps with the sites that traditionally appear on page one. BrightEdge tracked this independently across 16 months and found 16.7% — flat the whole time. The divergence isn’t a blip. It’s structural.
It gets more striking when you dig into ChatGPT specifically. Only 6.82% of ChatGPT results overlap with Google’s top-10 organic results. One Ahrefs study found 28.3% of ChatGPT’s most-cited pages have zero organic visibility. A page that ranks nowhere can still be ChatGPT’s primary source on a topic.
And if you’re in B2B, consider this: G2’s April 2026 research found 85% of B2B buyers think more highly of a vendor when AI includes them in an answer. If half your buyers start research in ChatGPT or Perplexity, a strategy focused exclusively on ranking is a strategy for a shrinking share of the journey.
One honest caveat: the Conductor report page is JavaScript-rendered and wasn’t directly accessible during research. The 17% figure is consistent with BrightEdge’s independently published 16.7% and corroborated across multiple sources. Treat it as directionally solid — not as a number to quote in a board deck without a footnote.
GEO — Generative Engine Optimisation — is the practice of structuring content, entity relationships, and authority signals so that AI models cite your brand when synthesising responses.
Traditional SEO targets ranking algorithms that score pages holistically. GEO targets RAG pipelines — AI systems that retrieve individual content chunks and synthesise a response from them. Because retrieval happens at the chunk level, page-level optimisation is no longer the whole game. That’s a meaningful difference in how you actually approach the work.
Lumar’s framework organises GEO into three layers. Entity GEO: who you are, resolved in the Knowledge Graph. Brand Authority GEO: whether you can be trusted — backlinks, media mentions, E-E-A-T signals. Content-Level GEO: passage structuring, schema, FAQ format.
The terminology is contested, for what it’s worth. Graphite calls it “AEO”; AthenaHQ calls it “GEO”; SurferSEO uses “AI SEO.” That last label has 250,000 monthly searches versus 1,300 for GEO — but it conflates two operationally distinct tracks. In this article we use GEO for AI-summary citation and AEO for direct-answer citation, because the distinction changes what you actually do day-to-day.
AEO — Answer Engine Optimisation — is the practice of optimising content to be selected when AI platforms generate direct, single-intent answers. Voice-style queries, factual questions, comparisons where the user wants one authoritative answer rather than a synthesised summary.
It grew out of Position Zero and Featured Snippet strategies that were prominent around 2016–2019. First Page Sage became the first agency to formally offer AEO as a named service in 2023 — their own description. The term spread as Perplexity branded itself an “answer engine” and practitioners needed a term to distinguish direct-answer optimisation from the broader GEO discipline.
In short: GEO handles complex multi-source summaries. AEO handles direct-answer retrieval. AEO requires shorter, more declarative passages optimised for voice extraction. In practice, most content teams end up optimising for both at the same time — question-format headings, declarative openings, and FAQ sections serve both tracks without needing to pick one.
Before a generative AI system can cite your content, it needs to answer one question: who are you?
This is Step 0. If your brand isn’t a resolved entity in Google’s Knowledge Graph, you don’t enter the candidate pool for retrieval. Without entity resolution, content-level optimisation produces inconsistent results no matter how well-structured your content is.
Lumar’s framework identifies five mechanisms to get this right: schema markup for Organisation, Product, and Person; consistent brand descriptions across your website, LinkedIn, Google Business Profile, and industry directories; Wikipedia inclusion (the most-cited domain across AI Overviews, AI Mode, and ChatGPT responses); B2B review platform listings on G2 and Trustpilot; and topical authority content clusters.
One measurement nuance worth knowing before you start tracking results: ChatGPT cites sources 87% of the time but mentions brand names in only 20.7% of answers. These ghost citations — source link, no brand mention — mean raw citation rate can significantly overstate actual brand visibility. Entity GEO reduces that risk.
On schema specifically: position.digital’s finding that “schema markup has no major impact on AI visibility” applies to schema as a general ranking booster. Schema for entity signals — Organisation, Product, Person — operates at the entity resolution layer. Implement those for entity resolution. Don’t rely on schema alone as a citation driver.
For the full evidence on the empirical research on what actually predicts AI citation, the research article in this cluster goes deeper.
There is no single platform-agnostic strategy that works across all three. They have different citation patterns and a different relationship to traditional SEO.
Google AI Overviews have the strongest correlation with traditional SEO of any AI platform — though it’s falling fast, from 76% to 38%. Organic CTR is 35% higher when your brand is cited in an AI Overview, versus 61% lower when you’re absent. If your buyers come via Google search, your existing SEO investment transfers most directly here.
Google AI Mode is a separate interface with a 93% zero-click rate. AI Mode and AI Overviews share only 13.7% of cited domains — despite both being Google products. Treating “Google AI” as a single optimisation target is a good way to misallocate your resources.
ChatGPT drives 87.4% of all AI referral traffic. Its most-cited sources are Wikipedia (47.9%), Reddit (11.3%), and Forbes (6.8%). Entity signals and third-party mentions are stronger predictors here than ranking.
Perplexity weights recency heavily — 50% of its citations are content published in 2025 alone. For research-intensive B2B queries, structured declarative content performs better here than on Google.
Microsoft Copilot is worth adding to the list if you’re a B2B team working in Microsoft 365 environments. Claude has expanded web search to all users as of May 2025 — check that your robots.txt doesn’t block ClaudeBot or Claude-SearchBot.
For measurement, tools like Profound, Peec AI, Ahrefs Brand Radar, HubSpot AI Search Grader, and Otterly track AI citation share across platforms. (What changes when the searcher is an AI agent rather than a human is explored in the agentic search article in this cluster; the prioritisation question shifts again there.)
AI engines don’t evaluate pages holistically. They retrieve individual content chunks via RAG pipelines.
44.2% of all LLM citations come from the first 30% of a page’s text. A declarative opening sentence is citation infrastructure, not a stylistic preference. In practice that means: declarative opening sentences per section, self-contained sections of 120–180 words, question-format headings, inline statistics, and FAQ sections formatted for AI extraction.
Content freshness matters too. Content updated in the past three months averages 6 citations versus 3.6 for outdated pages. Refreshing priority pages every 60–90 days is a practitioner recommendation, not an experimentally validated interval — but the directional signal is consistent across sources.
The third-party finding is the one that surprises most content teams: brands are 6.5 times more likely to be cited through third-party sources than their own domains. Building presence on Reddit, Wikipedia, G2, and industry publications is a more direct path to AI citation than publishing more owned content. That’s a genuine shift in where your effort should go.
Here’s an honest summary of where the evidence stands: entity building is well-validated, passage-level structuring is well-corroborated, and third-party citation building has multiple independent studies behind it. Specific refresh intervals, specific GEO content length targets, and update frequency signals borrowed from SEO without transfer validation are still hypothesis. GEO and AEO are young disciplines. The foundation is solid. The fine-tuning is still being worked out.
GEO targets AI models synthesising multi-source summaries in response to complex queries. AEO targets direct-answer retrieval — factual or single-intent questions where users expect one authoritative answer. Most structural practices serve both tracks at once. AEO additionally requires shorter, more extractable passages for voice output.
No. BrightEdge’s tracking found only 16.7% overlap between AI-cited sources and traditional top-10 organic rankings, flat across 16 months. For ChatGPT, 80–90% of cited pages rank outside the top 10. SEO is necessary but doesn’t reliably predict AI citation. A separate GEO/AEO effort is required.
Entity resolution means an AI system can unambiguously identify and describe your brand — a resolved node in the Knowledge Graph. Without it, AI engines may misattribute your content, cite a similarly-named competitor, or omit you entirely. Entity GEO is the practice of achieving entity resolution before you attempt content-level optimisation.
A ghost citation occurs when an AI response links to your domain but doesn’t name your brand in the answer text. ChatGPT cites sources 87% of the time but mentions brand names in only 20.7% of answers. AI citation rate is not the same as AI brand mentions — teams tracking citation share may be significantly overcounting their actual visibility.
For entity signals — Organisation, Product, Person schemas — yes. position.digital’s finding that “schema markup has no major impact on AI visibility” applies to schema as a broad ranking booster, not as an entity-signal mechanism. Implement Organisation and Product schema for entity resolution. Don’t expect schema alone to drive citation frequency.
Start with Google AI Mode if your current traffic comes from Google search — entity signals and passage structure transfer to AI Overviews as well. Prioritise ChatGPT entity signals — Wikipedia inclusion, G2 listings, Reddit mentions — if your buyers research in AI assistants. ChatGPT cites lower-ranked pages 80–90% of the time, so SEO rank is a weaker predictor there. Either way, entity GEO is the prerequisite.
“AI SEO” positions the discipline as a direct evolution of traditional SEO. GEO and AEO treat AI citation optimisation as materially different from ranking optimisation, because the evidence shows ranking and citation are largely decoupled. “AI SEO” has the highest search volume (250K/month) but the conflation creates real operational confusion for teams trying to allocate effort between the two tracks.
Track AI citation share: the percentage of relevant AI-generated responses that cite your brand or domain. Tools like Profound, Peec AI, Ahrefs Brand Radar, HubSpot AI Search Grader, and Otterly track this across platforms. Run prompt harnesses simulating buyer queries across Google AI Mode, ChatGPT, and Perplexity. And remember: a source link is not the same as a brand mention.
G2’s April 2026 research found 51% of US B2B software buyers now start research with an AI chatbot more often than Google. For B2B SaaS, the AI citation moment arrives earlier in the buyer journey than most teams expect. Microsoft Copilot is disproportionately relevant for B2B teams in Microsoft 365 environments — a platform most B2C practitioners don’t need to think about.
GEO and AEO are new disciplines but they rest on a clear sequence: entity resolution first, passage-level structuring second, third-party citation building third. The Conductor 17% finding tells you why the sequence matters — the old ranking-to-visibility relationship is broken, and a separate optimisation track is the only way to close the gap.
For a complete overview of the structural shift driving this new optimisation landscape — including the full data picture, the mechanism behind the click collapse, and how the publisher revenue crisis connects — see our AI search zero-click crisis guide.
AI Overviews Kill 61% of Clicks — The Mechanism Behind the DropYour Google Search Console shows stable rankings. Your Google Analytics shows a traffic cliff. The answer is architectural — and it’s not going to fix itself.
Google’s AI Overviews (AIOs), powered by Gemini, now appear for a growing share of queries and answer the user’s question before they have any reason to click through. Position-1 click-through rate has collapsed from 7.3% to 1.6% when an AI Overview is present — a 78% drop in two years (Ahrefs, December 2025). Across 300,000+ keywords, organic CTR falls 61% on average whenever an AIO appears.
This article explains the architectural mechanism behind those numbers — written for a technical audience that wants the why, not just the what. We’ll cover query fan-out, Gemini’s citation selection logic, the Jaccard similarity finding, and why the featured snippet playbook no longer applies. All of this connects to the broader AI search zero-click crisis.
When you submit a search, Google’s AI system doesn’t return a single ranked list. It decomposes your query.
The process — documented in three Google patents (US20240289407A1, US12158907B1, US11663201B2) — works like this. A single user query is broken into 8–12 parallel sub-queries. These fire simultaneously across multiple retrieval surfaces: the Web index, Knowledge Graph, Shopping Graph, Maps, and YouTube. Each sub-query generates its own results. Passages are then extracted via semantic chunking — documents broken into candidate answer fragments, each evaluated against the relevant sub-query. Gemini synthesises the retrieved passages into a single AI Overview, selecting which passages to cite based on relevance scoring, not traditional ranking order.
💡 Retrieval-Augmented Generation (RAG) is an architecture where a language model retrieves relevant content from external sources before generating its response — the model reads before it writes, which is why passage structure in your content directly affects whether it gets cited.
Two details are worth paying attention to here. First, the fan-out sub-queries are not stable — only 27% are consistent across repeated searches of the same head query (Surfer SEO). Second, this isn’t unique to AI Overviews: AI Mode uses the same mechanism with far more extensive fan-out.
Think of it as a research team sent out to answer a question. They split up, each searches a different related angle, and they bring back the most useful passages. The final answer draws from all of those angles — not from whoever ranked first for the question as originally typed. That decoupling is the mechanism that breaks the old ranking-to-traffic chain.
For the zero-click statistics that quantify this impact at scale, see Article 1.
When an AI Overview is present, position-1 CTR falls from 7.3% to 1.6% — a 78% collapse. And the drop isn’t confined to position 1. Ahrefs’ data shows position 2 down 50.8%, position 5 down 32.6%, position 10 down 19.4%. The effect cascades across the whole SERP.
The mechanism is straightforward. The AIO satisfies query intent on-page. The user gets a synthesised answer with cited sources visible inline. There’s no reason to click. Your page keeps its ranking impression in Search Console — but the click never happens.
This is structurally different from the featured snippet era. Featured snippets gave users one extractable answer and they often clicked through for more context. AIOs deliver complete synthesised answers. Your organic result shifts from primary destination to fallback content for users who want to go substantially deeper.
AI Overviews now appear for approximately 13–25% of all queries (Q1 2026). For pure informational queries, saturation is near-total. And the effect is spreading: Semrush shows transactional AI Overview presence grew from 1.98% to 13.94% between 2024 and 2025 — funnel-stage traffic is now exposed.
Citation selection is not a ranking operation.
Gemini doesn’t look at the top organic result and cite it. It evaluates passages retrieved from fan-out sub-query SERPs — which may not include the head-query SERP at all — and selects based on passage-level relevance scoring. A URL that ranks number one for the head query may not appear in any of the 8–12 sub-query result sets. It’s simply not in the retrieval pool.
Ahrefs’ February 2026 analysis of 863,000 keyword SERPs confirms this: only 37.9% of URLs cited in AI Overviews appear in the top-10 organic results for the same query. That was 76% in July 2025. The overlap has nearly halved in seven months.
The Knowledge Graph functions as a prerequisite filter. It’s Google’s structured entity database — recording and resolving canonical identities for people, organisations, places, and concepts. Pages referencing entities that are verified and resolved in the Knowledge Graph are more likely to pass the initial retrieval threshold. Pages referencing unresolved entities get filtered out earlier.
Two citation patterns confirm that the logic is decoupled from traditional ranking. YouTube accounts for 18.2% of AI Overview citations among pages not in the top 100 organic results — because video is retrieved via a separate fan-out surface. Reddit accounts for approximately 21% of citations overall, despite lacking traditional domain authority. Both point to a system drawing its citation pool from a far wider retrieval surface than the original query’s top-10.
For what the citation data actually shows about surviving AI Overviews, see Article 7.
Jaccard similarity measures overlap between two sets: intersection divided by union, from 0 (no shared elements) to 1 (identical sets).
Research published in arxiv 2604.27790v1 measured the Jaccard similarity between traditional SERP top-10 rankings and the URLs cited in AI Overviews for the same queries. The average score is below 0.2. In plain terms: if your page ranks in the top 10, there’s an 80%+ probability it won’t appear as a citation in the AI Overview for that same query.
Multiple independent datasets back this up. Ahrefs (863K SERPs) shows 37.9% of AIO citations coming from top-10 results — down from 76% in July 2025. Surfer SEO (173,902 URLs) found that 68% of AI-cited pages were not in the top-10 organic results for the same query.
This isn’t a calibration issue — it’s an accelerating divergence. AI Overviews are relying less on direct search results and more on sources turning up in fan-out query SERPs. The direct implication: optimising for traditional SERP ranking is necessary but no longer sufficient for AI search visibility. The two optimisation targets are structurally diverging, and the gap is widening fast.
The fan-out architecture retrieves at the passage level, not the page level.
Semantic chunking breaks a document into candidate answer fragments. Each is evaluated independently against a specific sub-query. The system isn’t asking “is this a good page?” — it’s asking “does this passage directly answer this specific sub-question?”
A page at position 1 may contain no passage that addresses any of the 8–12 fan-out sub-queries. A page at position 40 may contain several. Wellows research (December 2025) provides the operational data: the optimal passage length for AI Overview extraction is 134–167 words. Cosine similarity above 0.88 between a passage and its target sub-query produces 7.3x higher citation rates.
💡 Cosine similarity measures how closely two pieces of text are aligned in meaning when converted into mathematical vector representations — a score of 0.88 is very high, meaning the passage and sub-query are semantically very close.
Surfer SEO’s analysis of 173,902 URLs found a Spearman correlation of 0.77 between topical coverage — how many fan-out sub-query variants a page addresses — and AIO citation probability. Fan-out coverage, not head-query ranking authority, is the structural predictor.
This rewrites the old content model entirely. Old model: one page, one target query, position zero. New model: one page, multiple structured passages, multiple fan-out sub-queries. Each H2 section should function as a self-contained answer — a declarative opening sentence, a specific factual claim, a data reference — completable in 134–167 words without requiring context from surrounding paragraphs.
The optimisation disciplines that operationalise this are the subject of GEO and AEO — the optimisation disciplines this mechanism created.
AI Overviews and AI Mode are not two different products. They’re two deployments of the same architecture at different intensity levels. Both use query fan-out, RAG passage extraction, and Gemini synthesis. AI Overviews run 8–12 sub-queries and produce a zero-click rate of approximately 83%. AI Mode issues potentially hundreds of sub-queries and produces a 93% zero-click rate (Semrush).
The distinction that matters most for content strategy: AI Mode and AI Overviews cite the same URL only 14% of the time for similar answers (Digital Applied). Despite sharing the same underlying mechanism, they draw from largely separate retrieval pools. A content strategy targeting one cannot assume it’s targeting the other.
AI Mode is currently primarily a US feature; AI Overviews are globally deployed. If you’re outside the US, the optimisation work required for AI Overviews — passage-level structure, entity resolution, topical coverage — is exactly the preparation you need for AI Mode when it arrives.
The traditional search value exchange was predictable: produce quality content, earn rankings, receive traffic. That chain is architecturally broken — not cyclically depressed. The ranking system still functions. But ranking no longer guarantees the click.
The new citation logic operates on three layers.
Entity Verification: Is the brand or source a resolved entity in the Knowledge Graph? Pages referencing verified, canonical entities pass the initial retrieval threshold more reliably.
Passage-Level Relevance: Does the content contain passages that directly address fan-out sub-queries at the right semantic distance? Cosine similarity above 0.88, passage length 134–167 words, declarative structure.
Topical Coverage: Does the content address enough sub-query variants to appear across the fan-out retrieval pool? The Spearman correlation of 0.77 between topical coverage and citation probability makes this the strongest structural predictor.
These three layers replace “rank well → get traffic” as the working mental model. AIO citations are probabilistic, not deterministic — only 27% of fan-out sub-queries are stable across repeated searches. So AI citation share — how often a brand appears across many query executions — is the correct measurement metric. Tools worth knowing about: Ahrefs Brand Radar, Semrush AI Visibility Toolkit, Profound, Peec AI, and Otterly.
Traditional SEO remains a baseline requirement. For the 75–87% of queries that don’t generate an AI Overview, organic ranking remains the primary traffic driver. The practical approach: maintain SEO as a floor and layer passage-level structuring and entity building — GEO and AEO — as the differentiated layer that determines AI citation outcomes.
This article has explained why the old model is broken. GEO and AEO — the optimisation disciplines this mechanism created covers the optimisation disciplines that replace it. Article 7 examines what the citation data actually shows about surviving AI Overviews. For the full scope of the zero-click shift, the pillar article covers the complete landscape.
When an AI Overview is present, position-1 CTR falls from 7.3% to 1.6% without any ranking change. The AIO satisfies user intent on-page — the user reads the synthesised answer and doesn’t click through. Your page still registers an impression in Search Console; the click simply doesn’t happen. Rankings and traffic are now decoupled.
Instead of returning one ranked list, Google’s AI decomposes your query into 8–12 related sub-questions, searches each in parallel across the web index, Knowledge Graph, YouTube, and other surfaces, then synthesises the most relevant passages into one AI Overview answer. The specific wording of your original query matters less — the AI rewrites it before retrieval begins.
No. Only 37.9% of AI Overview citations come from top-10 organic results for the same query (Ahrefs, February 2026) — down from 76% in July 2025. Citations are drawn from fan-out sub-query SERPs, not the head-query SERP. The two lists are largely non-overlapping, and increasingly so.
AI search systems break your page into passage-sized chunks and evaluate each against a specific sub-query. Each H2 should open with a direct declarative sentence, followed by specific data, completable in under 170 words without requiring surrounding context. Cosine similarity above 0.88 between a passage and its target sub-query produces 7.3x higher citation rates (Wellows, December 2025).
Yes — it’s a baseline requirement, not a sufficient strategy on its own. For the 75–87% of queries that don’t generate an AI Overview, organic ranking remains the primary traffic driver, and organic signals (E-E-A-T, topical authority, site health) remain relevant inputs to citation selection. The practical approach: maintain SEO as a floor and layer passage-level content structuring and entity building (GEO/AEO) as the differentiated layer that determines AI citation outcomes.
60% Zero-Click — The Numbers That Rewrite Every Traffic AssumptionYou have seen the headlines. 60% of Google searches end without a click. Another study says 83%. A third says 93%. All three numbers appear in serious publications, all three come from credible researchers — and all three are correct.
That is not a contradiction. It is a measurement problem. The 60% figure covers all query types across all sessions. The 83% applies only to queries that triggered an AI Overview. The 93% measures what happens inside Google’s AI Mode interface. In this article we are going to reconcile all three so they make sense together.
The sources include SparkToro, Datos, Semrush, Ahrefs, Pew Research Center, BrightEdge, and Chartbeat. Together they paint a consistent picture: zero-click is accelerating, and the trajectory through 2027 points one direction. This article is part of our comprehensive guide to the AI search zero-click crisis — the structural shift that is rewriting how content-led businesses think about search.
Zero-click statistics diverge because researchers are measuring different things. They are not competing claims — they are parallel measurements of different slices of the same ecosystem.
Clickstream panel data (SparkToro and Semrush, both using Datos) tracks real user behaviour via opt-in browser extensions — what happened after the search, at session scale.
Keyword-level CTR analysis (Ahrefs) measures click-through rate on specific keywords using aggregated Google Search Console data across 300,000 keywords. It answers a different question: what fraction of searchers on a given term clicked an organic result?
Consumer panel observation (Pew Research Center) directly watches 900 US adults across 68,879 real sessions — no browser extensions, no inferred data.
There is also a definitional wrinkle. SparkToro counts clicks to YouTube, Google Maps, and Shopping as zero-click — the user never left Google’s ecosystem. That choice pushes their headline rate higher than alternatives that count all destination clicks.
Here is where the main studies land. SparkToro / Datos (all queries, open-web clicks only) — 64.82% zero-click (US, 2024). Semrush / Datos (AIO-triggered queries only) — approximately 83%. Ahrefs (per-keyword CTR, 300,000 keywords) — 58% CTR reduction at position 1. Pew Research Center (68,879 sessions) — 8% click rate with AIO versus 15% without. Searchlab (cross-study synthesis) — 93% zero-click in AI Mode.
SparkToro measures all Google search sessions — navigational, branded, informational, local, transactional. Their 2024 study found 64.82% of US sessions ended without a click to the open web. For every 1,000 queries, only 374 visits reach the open web.
This is not an AI Overview shock. It is an acceleration of a multi-year trend: 43.9% zero-click in 2016, 50.3% in 2019, 58.5% in 2022–2023, 64.8% in 2024–2026. The zero-click rate has grown three times faster than total search volume since 2018 (SimilarWeb). AI Overviews are accelerating a structural shift already driven by featured snippets, knowledge panels, People Also Ask, and local packs — not creating it from scratch.
One important caveat: SparkToro’s sample includes navigational searches. When someone types a brand name and goes straight to that site, that counts as zero-click — but those users were already converted. Use 64.82% when describing aggregate reality across all Google searches. Do not use it as your personal exposure estimate.
Semrush studied 10 million+ keywords and 200,000+ queries using Datos clickstream data. Result: approximately 83% of searches that trigger an AI Overview end without an organic click. Important caveat: AI Overviews trigger on roughly 15–25% of all Google searches. The compound maths matters — if 20% of your queries trigger an AIO and 83% of those end without a click, you are losing clicks on roughly one in six target queries, not every search.
AI Mode is structurally different. AI Overviews appear above a traditional SERP that still shows organic links below. AI Mode — launched March 2025 and broadly available in the US from May 2025 — replaces the SERP entirely with a Gemini-powered conversational interface. Organic URLs are not surfaced the same way. Result: 93% zero-click (Searchlab, compiled from Semrush and position.digital). The gap between 83% and 93% is simply the difference between a SERP that shows links and one that does not.
Your exposure to the 83% rate depends on how often your target queries trigger AIOs today. The 93% is the forward-looking risk if AI Mode becomes the default interface.
Here is where the abstraction becomes concrete. Ahrefs’ February 2026 update across 300,000 keywords found that when an AI Overview is present, position-1 CTR for informational queries drops from approximately 7.6% to 1.6% — a 58% reduction for the same ranking position. Your page has not moved. The SERP around it has changed.
Pew Research Center’s independent panel confirms this at the session level: across 68,879 sessions, users clicked any result in only 8% of visits to AIO pages versus 15% of visits to pages without one. Two entirely different methodologies, the same conclusion — AIO presence roughly halves the probability of any organic click occurring.
So a top-3 ranking in an AIO-saturated query space is worth significantly less than it was in 2023. Not worthless — but you need to recalibrate what that ranking actually delivers.
BrightEdge’s 12-month longitudinal analysis found healthcare queries trigger an AI Overview 88% of the time, education 83%, B2B tech 82%. The cross-industry average is 48%.
If you are running a SaaS, FinTech, HealthTech, or EdTech company, run the compound calculation: 82% AIO coverage rate applied to your informational query volume, then 83% conditional zero-click rate applied to the AIO-triggered portion. Roughly four in five informational queries in B2B tech now compete with an AI-generated on-SERP answer — and most of those sessions end without a click.
Query intent matters as much as vertical. Semrush data: informational queries hit 74% zero-click, commercial investigation 46%, transactional 31%. Transactional and branded queries are structurally protected — Google’s commercial model depends on routing buying intent to advertisers. Zero-click hits hardest at the top of the funnel. The bottom is largely intact.
Chartbeat tracked Google Search referral traffic to 2,500+ news and media publisher sites between November 2024 and November 2025 and found a 33% year-over-year global decline — 38% in the US. Reported by Press Gazette and the Reuters Institute.
Not a model. A direct measurement. And it is not evenly distributed: small publishers lost 60% of search referral traffic over two years, medium publishers 42%, large publishers 22%.
Reuters Institute notes the combined effect — AI Overviews plus the broader shift away from search toward social, newsletters, and AI chatbots — makes clean attribution difficult. For a B2B tech company, Chartbeat is the upper bound: what happens when zero-click exposure is near-total. See how publishers are losing revenue to this shift.
The direction is unambiguous. 43.9% zero-click in 2016, 50.3% in 2019, 58.5% in 2022–2023, 64.8% in 2024–2026, approximately 83% for AIO-triggered queries today, 93% in AI Mode. These are not a single linear series, but they trace a consistent structural direction — each new Google SERP feature that answers queries on-page adds another slice of searches ending without an open-web click.
AI Mode is the forward signal: 75 million daily active users, available in 53+ languages, with a 93% zero-click rate that represents where the standard SERP is heading.
There is one structural brake worth noting: Google’s ad revenue. Ad integration in AI Overviews began in Q1 2026, and Google needs to route transactional intent to advertisers — so the trend’s endpoint is not 100%. The 2026–2027 outlook is that informational content’s organic CTR will keep declining. The question is rate of decline, not direction.
Here is the reconciled picture. The 64-65% SparkToro figure is the baseline across all Google searches. The 83% is your conditional exposure when informational content competes with AI Overviews. The 93% is the trajectory if AI Mode becomes the dominant interface. Which number applies to you depends on what proportion of your target queries are informational and what proportion of those trigger an AIO.
There are two things to hold alongside the traffic numbers.
Zero-click brand impressions have value. Brands cited in AI Overviews earn 35% higher organic CTR and 91% higher paid CTR on those queries compared to uncited brands (Seer Interactive / ALM Corp). The gap is in attribution — existing analytics stacks do not capture AI Overview impressions. Appearing in AI Overview responses matters; you just need different tools to measure it.
The metrics that matter are changing. SERP Share of Voice — how often your brand appears in AI-generated answers — is replacing click volume as the KPI that actually predicts market share. It correlates with market share at 0.87, compared to 0.71 for click volume alone (Semrush / Nielsen).
The inbound funnel implication is not a traffic collapse — it is a channel-mix shift. Informational content’s role in search discovery is diminishing. Direct branded search, AI citation presence, community, and email carry more of the load. For the strategic response, start with how Google’s AI Overviews actually suppress clicks and what the data shows about ranking in AI responses. Understanding the zero-click crisis reshaping content strategy starts with the numbers — and now you have them.
Approximately 64-65% of all Google searches end without a click to the open web (SparkToro/Datos 2024). For queries triggering an AI Overview, that rises to approximately 83% (Semrush); in AI Mode it reaches 93% (Searchlab 2026).
A zero-click search is a Google session that ends without the user clicking through to any website outside Google’s own properties — answered on the SERP via an AI Overview, featured snippet, knowledge panel, or similar feature.
Yes. Searchlab data shows 77% zero-click on mobile versus approximately 56% on desktop, because mobile SERPs more frequently display features that fully answer queries without requiring a click.
AI Overviews are AI-generated summary boxes sitting above organic results — links are still visible below. AI Mode (launched March 2025, broadly available in the US from May 2025) replaces the SERP entirely with a Gemini-powered conversational interface. AI Overviews produce approximately 83% zero-click; AI Mode produces 93%.
SparkToro partnered with Datos to analyse opt-in browser session data from US and EU users, tracking whether users clicked any website outside Google’s own properties. The 2024 study covers all query types — navigational, branded, informational, transactional — which is why their headline rate is higher than studies that isolate specific categories.
Pew Research published its findings in July 2025, based on a March 2025 panel of 900 US adults across 68,879 search sessions — 8% click rate on AIO pages versus 15% without. Available at the Pew Research Center website.
Because the SERP environment has changed around your stable ranking. When an AI Overview appears above your result, position-1 CTR for informational queries drops from approximately 7.6% to 1.6% (Ahrefs, February 2026, 300,000 keywords) — AI Overview presence intercepts most clicks before users reach your result.
BrightEdge’s 12-month analysis found B2B tech queries trigger an AI Overview approximately 82% of the time — comparable to healthcare (88%) and education (83%), and well above the cross-industry average of 48%.
SERP Share of Voice measures how frequently a brand appears in AI-generated search answers. It correlates with market share at 0.87, compared to 0.71 for click volume alone (Semrush / Nielsen), making it a more reliable forward-looking KPI as click volume declines.
Chartbeat tracked 2,500+ publisher sites and found a 33% year-over-year global decline in Google Search referrals between November 2024 and November 2025 (38% in the US), reported by Press Gazette and the Reuters Institute.
Transactional queries average only 31% zero-click (Semrush) and branded navigational queries carry a 57% position-1 CTR, because Google routes buying intent to advertisers. Zero-click is primarily a threat to informational and educational content.
Audit your keyword portfolio by intent type using Semrush or Ahrefs. Work out what proportion are informational versus transactional or branded. Then check what percentage of those informational queries trigger an AI Overview and apply the approximately 83% conditional zero-click rate. That is your exposure.