Yahoo's CEO Jim Lanzone called AI Mode the biggest threat to web traffic, on the logic that answer engines depend on content while sending fewer clicks back. Google's VP of Search, Liz Reid, published a lengthy rebuttal arguing that third-party measurements "inaccurately suggest dramatic declines in aggregate traffic." Both statements were made in good faith. Both are, in a narrow sense, defensible.
That contradiction is the actual story here — and most coverage of AI Mode resolves it by simply picking a side. This piece does not. What follows is what has structurally changed, what the traffic data does and does not support, where the evidence genuinely conflicts, and what any of it means for how you work.
The short version
AI Mode breaks one query into many parallel sub-queries, which decouples ranking from visibility — you can rank first and still be skipped. Publisher traffic has fallen sharply, though the headline figures vary wildly by content type and methodology, and at least one credible dataset suggests the decline is levelling off. The durable shift is not "less traffic," it is that visibility, measurement, and the value of a click have all been redefined at once.
MechanicsWhat AI Mode Actually Does Differently
Strip away the framing and one architectural change drives most of the downstream disruption: query fan-out.
When someone submits a query in AI Mode, Google does not run one search. It decomposes the question into multiple related sub-queries — reporting suggests up to sixteen running in parallel — retrieves and ranks sources independently for each, then synthesizes everything into a single answer with inline citations. The user can then ask a follow-up and the system retains context.
The consequence is easy to state and awkward to absorb: ranking first for the original query is no longer sufficient. Your page can hold position one and still be absent from the answer, because a page ranking sixth happened to contain a cleaner passage answering one of the sub-queries. Ahrefs found in an analysis of 173,020 URLs that pages were around 161% more likely to be cited when they also ranked for the fan-out queries, not just the head term.
Ranking is now an eligibility condition, not an outcome. Fan-out decides who actually appears.
Two further changes matter for context. The model powering all of this has moved fast — Gemini 3 became the default for AI Overviews globally on January 27, 2026, replacing a lighter model that only routed hard queries to heavier reasoning. AI Mode moved to Gemini 3.5 Flash as its global default at I/O on May 19, 2026. And at that same event, Google announced AI Overviews and AI Mode are merging into what Liz Reid described as one seamless AI Search experience.
Scale is no longer speculative either. AI Overviews reaches roughly two billion monthly users; AI Mode passed one billion. This is not an emerging surface. It is the default experience for a substantial share of the web's search demand.
The DamageWhat the Traffic Data Actually Shows
Here is where most articles either catastrophise or dismiss. The honest answer is that the numbers are large, real, and much more variable than any single headline suggests.
| Source | Finding | Scope |
|---|---|---|
| Reuters Institute / Chartbeat (Jan 2026) | −33% global publisher search traffic in year to Nov 2025; U.S. publishers −38% | Broad publisher panel |
| Peer-reviewed research | ~−15% direct AI Overview effect on publisher page traffic | Controlled academic study |
| Digital Content Next | Majority of members losing between 1% and 25%; ~10% aggregate decline | ~40 major publishers |
| HubSpot (disclosed) | −70–80% organic traffic | Single company |
| Business Insider (disclosed) | −55% Apr 2022 to Apr 2025; 21% staff cut | Single company |
| Chegg (disclosed) | Revenue −24% YoY; sued Google in Feb 2025 citing AI Overviews | Single company |
| Reuters Institute forecast | News publishers expect −43% search traffic by 2029 (median); ~20% expect worse than −75% | Industry survey |
Notice the spread: from 10% to 80% depending on who you ask. That is not measurement error. It reflects a genuine, uneven distribution of harm.
The pattern in the research is consistent: informational and definitional content took the heaviest losses, because a synthesized paragraph substitutes for the click almost perfectly. Health explainers, how-to tutorials, reference material, recipes, and travel guides sit at the painful end. STEM and technical content saw smaller declines — users still need the full source. Transactional and commercial queries have been comparatively insulated so far.
If your content is mostly "what is X" and "how do I Y," you are in the exposed category. If it is specification-heavy, transactional, or requires the actual artifact, you are not — yet.
There is also a quality argument, and it is not purely Google spin. Peer-reviewed work found AI-referred visitors converted roughly 42% better, spent about 48% longer on site, and generated around 37% higher revenue per visit. Seer Interactive, tracking 3,119 terms across 42 organizations, found brands cited inside an AI Overview earned 35% more organic clicks and 91% more paid clicks than uncited brands on the same queries.
The catch — and it is a serious one — is scale. Chatbot referrals to publishers grew over 200%, but still account for less than 1% of all referrals. Better-converting traffic at a fraction of the volume does not make a publisher whole. For a commercial site with a narrow funnel, the maths can work. For an ad-funded content business, it usually cannot.
RestructuringFive Things That Have Genuinely Changed
Visibility and ranking have separated into different problems
For twenty years, rank was a reliable proxy for visibility. That relationship has broken. Analysis published in mid-2026 indicates only around 38% of AI Overview citations now come from Google's organic top 10 — meaning the majority of citations go to pages that were not the traditional winners.
Practically, this means an SEO report showing stable rankings can coexist with collapsing visibility, and nobody notices until revenue moves. Rank tracking alone has become an incomplete instrument.
Off-site brand signals now outweigh links
Ahrefs analysed 75,000 brands and found branded web mentions correlated with AI Overview brand visibility at roughly 0.664, and YouTube mentions at about 0.737 — against backlinks at roughly 0.218. Off-site brand language beat the traditional link signal by two to three times.
Separately, University of Toronto research found roughly 91% of AI-generated answers cite third-party content rather than brand-owned sites. Both findings push the same direction: the centre of gravity has moved off your own domain. We covered the implications in detail in brand visibility in Gemini AI Overviews.
This has a slightly uncomfortable implication for the industry: digital PR, which spent a decade being treated as link-building's less measurable cousin, turns out to be the higher-leverage activity.
The content that wins has inverted
Thin, keyword-targeted pages built to capture a single query are the most exposed asset class in the ecosystem — they are exactly what a synthesized answer replaces. Meanwhile the Princeton and IIT Delhi GEO research found keyword stuffing performed below the do-nothing baseline for AI visibility, while citing sources and adding attributed statistics delivered 30 to 40% relative improvements.
What holds value now is content an AI cannot synthesize from elsewhere: original data, first-party research, genuine operational experience, opinionated comparison built on real criteria, and tools that do something rather than describe something. If a competent model could write your page from general knowledge, that page is a depreciating asset.
Measurement has quietly become the hardest part of the job
Sessions are no longer a clean success metric, because the same brand impact now produces fewer of them. But most reporting infrastructure — and most client expectations, and most internal targets — are still built on session counts.
Teams that adapt are tracking citation rate, brand mention frequency, branded search volume, and direct traffic alongside conversions, treating traffic as one input rather than the outcome. Teams that do not are having a difficult conversation every quarter about numbers that look like failure but may reflect a channel behaving differently rather than performing worse.
This is genuinely hard, and worth saying plainly: nobody has fully solved AI visibility measurement yet. The tools are young and most of the published benchmarks come from vendors selling those tools.
Blocking the crawler is not a defensive option
The intuitive response to "AI is taking my traffic" is to block it. The research suggests this backfires. Studies have found AI Overviews retrieve fewer sources from sites blocking Google's AI crawler — but the answer still gets generated, from someone else's content. You remove yourself from the citation without recovering the click.
Blocking is a coherent choice only as part of a licensing or legal strategy, which is roughly what large publishers negotiating directly with AI companies are doing. For everyone else it accelerates the loss.
DisagreementWhy Smart People Are Reaching Opposite Conclusions
It is worth understanding why the industry has not converged on an answer, because it tells you how much confidence to place in anyone's advice — including ours.
The measurement methodologies genuinely differ. Keyword-level analyses, clickstream panels, publisher-reported analytics, and Google's own aggregate data all measure different things. SparkToro's keyword-level zero-click figure and Datos' clickstream figure are not the same quantity, so pointing out that they disagree is not the gotcha it appears to be.
The impact really is unevenly distributed. A reference publisher and a B2B SaaS company are having genuinely different experiences. Both are describing reality accurately. Neither generalises.
Everyone has an interest. Google has an interest in the impact looking modest. Publishers negotiating licensing deals have an interest in it looking severe. Vendors selling AI-visibility software have an interest in urgency. This does not make anyone dishonest, but it does mean you should weight primary data over framing.
Notably, several publishing executives told Digiday they are deliberately not overhauling their approach in response to GEO hype — the view being that strong journalism and sound SEO fundamentals remain the correct strategy regardless. That is a defensible position, and the fact that experienced operators hold it should temper anyone's certainty in the opposite direction.
PracticallyWhat This Means for How You Work
Setting aside the ecosystem argument, here is what we would actually change:
- Audit your content by substitutability. Sort pages by the question: could a competent AI produce this from general knowledge? Everything that could is at risk. Prioritise defending or replacing it.
- Add citation and mention tracking to reporting now, even crudely — a fixed set of twenty prompts checked monthly beats waiting for perfect tooling. This also protects you when rankings look fine and revenue does not.
- Rebalance budget toward off-site brand work. If the correlation data holds, digital PR and earned coverage deserve a larger share than most current SEO budgets allocate.
- Write for topic coverage, not query coverage. Fan-out rewards depth across a subject; it punishes pages that answer exactly one question and stop. Practical detail in keyword research for AI search.
- Reset expectations with whoever you report to, before the numbers force the conversation. A quarter of declining sessions alongside stable conversions is a different story than it looks like on a dashboard — but only if you have framed it in advance.
- Do not block the crawlers unless you have a licensing strategy. Check robots.txt for
Google-Extendedand confirm it is deliberate.
If you are diagnosing a specific drop rather than planning strategy, why your website isn't showing up in AI and why your competitor is showing up instead of you are the more operational companions to this piece. The systematic version of the work sits in generative engine optimization and AI Overviews optimization.
HonestlyWhat Nobody Knows Yet
Three open questions worth holding in mind rather than pretending are settled.
Whether the decline plateaus or compounds. Datos' data suggests stabilisation; the Reuters Institute survey suggests publishers expect it to get considerably worse through 2029. Both could be right about different time horizons. We do not know.
Whether the economics find a new equilibrium. Google needs a healthy web to summarise. Publishers need revenue to produce what gets summarised. That tension is unresolved, and the resolution — licensing, regulation, some new referral mechanism, or simply a smaller open web — will shape the next decade of this discipline more than any optimization tactic.
Whether current tactics survive the next model. Gemini 3 changed citation behaviour in January. Gemini 3.5 Flash changed it again in May. Any specific claim about how sources get selected is a snapshot of a system Google is actively rebuilding. The structural principles — depth beats thinness, brand recognition beats links, being in the answer beats ranking below it — have held across several model generations. The tactical details have not.
The most useful posture is probably neither panic nor dismissal, but treating search as a channel whose rules are being rewritten while you operate in it — building the things that survive rule changes, and re-checking the rest quarterly.
Find out where you actually stand in AI search
A free audit covering citation rate, brand mention rate, and the specific queries where competitors appear and you don't — across AI Mode, AI Overviews, ChatGPT, and Perplexity.
Get Your Free AI Visibility AuditQuestions first? Talk to us. Or see the industries we work in.
Frequently asked questions
Is SEO dead because of Google AI Mode? +
What is query fan-out in Google AI Mode? +
How much traffic have publishers actually lost to AI search? +
Are AI Overviews and AI Mode merging? +
Does blocking Google's AI crawler protect my traffic? +
Is AI-referred traffic actually worth anything? +
What should SEO teams actually change in response to AI Mode? +
Sources & references
- Reuters Institute / Chartbeat — 2026 report (January 13, 2026): global publisher search traffic −33% in year to November 2025; U.S. −38%; 2029 forecasts
- Datos — Q1 2026 State of Search (April 27, 2026): U.S. zero-click rate 24.5% → 22.4%, clickstream methodology
- Digital Content Next — member publisher survey on AI Overview traffic impact, as reported by Digiday
- Peer-reviewed research on AI Overview traffic effects and AI-referral conversion quality, summarised by Authority Tech (~15% traffic reduction; +42% conversion, +37% revenue per visit)
- Ahrefs — query fan-out analysis (173,020 URLs; 161% higher citation likelihood when ranking for fan-out queries)
- Ahrefs — brand visibility correlation study, 75,000 brands (branded mentions ~0.664, YouTube ~0.737, backlinks ~0.218)
- Seer Interactive — AIO Impact on Google CTR (3,119 terms, 42 organizations)
- Aggarwal et al., Princeton / Georgia Tech / IIT Delhi / Allen Institute — "GEO: Generative Engine Optimization" (arXiv:2311.09735, KDD 2024)
- Google / Liz Reid — I/O announcement (May 19, 2026) on merging AI Overviews and AI Mode; public response to third-party traffic measurement
- Publisher disclosures: HubSpot, Business Insider, CNN, Chegg — as compiled by Search Engine Journal
- Google Search Central — AI features and your website
Sourcing note: the figures in this article come from methodologically incompatible sources — keyword-level analyses, clickstream panels, publisher-reported analytics, academic studies, and vendor research. We have kept them attributed and separate rather than averaging them into a single number, because doing so would imply a precision the underlying data does not support. Where a source has a commercial interest in the finding, we have said so.