TL;DR
- AI Overviews appear on 49.43% of non-personalized US search results and 65.07% of personalized ones. Clickstream data measuring what users actually experience puts it at 43%.
- Click-through on AI Overview queries stopped falling in early 2026. It bottomed at 1.3% in December 2025 and recovered 85% to 2.4% by February 2026.
- Only 38% of AI Overview citations come from Google's top 10, against 76% in mid-2025.
- 68.01% of US Google searches ended without a click in early 2026, up from 60.45% in 2024.
- ChatGPT citation volumes fell 86–94% across five markets between February and April 2026, then recovered by June. Citation visibility is far less stable than rankings ever were.
- AI traffic to US retailers converted 42% better than non-AI traffic in March 2026, having converted 38% worse a year earlier.
How often AI Overviews appear in US results
AI Overviews appear on 49.43% of non-personalized US search results, and 65.07% of personalized ones
Xponent21 analysed Advanced Web Ranking's dataset of 8,000 US keywords as updated on 2 March 2026. The non-personalized figure is what a brand-new user with no history encounters; the personalized figure is what Google serves once it factors in history, location and behaviour.
The gap between those two numbers is itself a finding. The more Google knows about a user, the more likely it is to answer rather than link.
Clickstream data puts the figure at 43% of searches actually performed
Similarweb's US panel, covering June 2025 to May 2026, measures a different thing: the share of real search sessions that included an AI Overview, reported by TechCrunch as up from roughly 15% at the start of that window. Similarweb's own writeup publishes the 43% but doesn't detail panel methodology, so treat it as directional.
Why prevalence estimates range from 43% to 65%
They measure three different populations, and the difference is not noise.
| Figure | What it measures |
|---|---|
| 43% | Share of real US search sessions that included an AI Overview (clickstream) |
| 49.43% | Share of tracked keywords triggering one for a user with no search history |
| 65.07% | Share of tracked keywords triggering one once personalization is applied |
Keyword-set measurements depend entirely on which keywords are in the set. Session measurements depend on who's in the panel. Any article presenting one of these as "the" AI Overview prevalence number is presenting a sample as a fact.
Commercial-intent AI Overviews grew 71% in six months
The reassurance that AI Overviews only affect informational content has expired. Semrush tracked 600,000+ US desktop keywords across 10 industries from November 2025 to April 2026 and found commercial-intent AI Overviews up 71% — led by Finance (+231%), Computers & Electronics (+108%) and Games (+77%). Transactional-intent AI Overviews fell 5% over the same period.
Keywords with an AI Overview also carry higher commercial value than those without: in Jobs & Education, $5.02 average CPC against $1.51.
Comparison queries trigger an AI Overview 95.4% of the time
Seer Interactive's analysis of 5.47 million queries across 53 brands, running to February 2026, puts comparison-format queries at 95.4%, question-format at 85.9% and "near me" queries at 76.9%. Informational queries overall sit at 36%, commercial at 8% and transactional at 5%.
If your content programme leans on comparison pages and buying guides — as most B2B programmes do — that's close to universal exposure. Before that changes any planning, it's worth knowing which prompts in your category actually carry volume, since trigger rates only matter on questions people ask.
What AI Overviews do to clicks
Click-through on AI Overview queries recovered 85% between December 2025 and February 2026
Seer Interactive's 2026 update is the largest first-party dataset in this field: 53 brands, 5.47 million queries, 2.43 billion organic impressions, drawn from consenting clients' Search Console data and modelled with per-segment regression.
Organic CTR on AI-Overview queries bottomed at 1.3% in December 2025 and recovered to 2.4% by February 2026. Queries without an AI Overview sat at roughly 3.3%.
This is the finding most AI SEO content hasn't caught up with. The decline was real. It also stopped.
Being cited roughly doubles your clicks on an AI Overview query
The same dataset separates three outcomes on informational queries:
| Outcome | Organic CTR | Clicks per million impressions |
|---|---|---|
| No AI Overview | ~3.3% | ~33,500 |
| AI Overview present, your brand cited | ~2.1% | ~20,743 |
| AI Overview present, your brand not cited | ~0.9% | ~9,445 |
Being cited doesn't recover what the AI Overview took. It's the difference between losing a third of your clicks and losing three-quarters of them — which makes citation, not position, the variable worth optimising once an AI Overview is present.
One 2026 study says position-one CTR was cut by 59%
SISTRIX analysed 100 million German keywords and found position-one CTR falling from 27% without an AI Overview to 11% with one. Other positions ran consistently 25–30% below their AI-Overview-free average. Scaled up, they estimate 265 million organic clicks lost per month in Germany, or 6.6% of all organic clicks.
Two caveats matter. It's German data, and AI Overviews appear on roughly 20% of German keywords against roughly half of US ones — a fundamentally different exposure rate. And SISTRIX doesn't publish a collection window on the page, though it appears in their February–March 2026 reporting.
Why the two 2026 CTR studies disagree
Seer says click-through is recovering. SISTRIX says it's been halved. Both are 2026 data, and both are right about what they measured.
| Seer Interactive | SISTRIX | |
|---|---|---|
| Finding | CTR recovering, 1.3% → 2.4% | Position-one CTR cut 27% → 11% |
| Data source | First-party Search Console clicks | Third-party rank panel, modelled CTR |
| Unit | Aggregate CTR across all positions held | Position one specifically |
| Design | 14-month time series | Single-point comparison |
| Market | US | Germany |
| Sample | 53 brands, 5.47M queries | 100M keywords |
Three of those rows explain most of the gap. Seer's 2.4% and SISTRIX's 11% aren't the same quantity — one is an average across every position a client ranks in, the other is position one alone. And a single-point comparison structurally cannot detect a recovery, because it has no time axis. SISTRIX isn't disputing Seer's rebound; it isn't measuring for one.
There's also a selection effect worth naming. Seer's 53 brands are enterprise SEO clients, more likely than average to be cited inside AI Overviews — and by Seer's own numbers, being cited more than doubles CTR. Some of the "recovery" may be more client pages getting cited rather than the underlying penalty easing.
A third 2026 dataset leans Seer's way: Advanced Web Ranking's Q1 2026 CTR analysis found desktop click-through for the top five positions rose a combined 10.54 percentage points against Q4 2025. It carries no AI Overview segmentation, so it can't settle the question — but it's moving in the same direction.
What no 2026 study currently shows
There is no published study with a 2026 collection window that gives US organic CTR broken down by position and split by AI Overview presence. Seer has the AI Overview split without positions. Advanced Web Ranking has the positions without the split. SISTRIX has both, in Germany.
The widely-circulated per-position table — 58% loss at position one, 50.8% at two, and so on — is December 2025 data. It's the best position-level evidence that exists. It just isn't 2026, and anyone presenting it as current is describing the market as it was fourteen months ago.
If your rankings have held while traffic hasn't, that gap between position and visibility is the thing to investigate, because position-level reporting can no longer explain it.
Zero-click search in the US
68.01% of US Google searches ended without a click in early 2026
SparkToro and Similarweb's US clickstream analysis covering January to April 2026 puts the zero-click rate at 68.01%, against a 60.45% baseline in 2024.
The composition matters as much as the total. Clicks to the open web fell 9.51 percentage points in two years — a 22.9% relative decline — while clicks to Google-owned properties rose 7.2 points. The study excludes the Google mobile app, so the true figure is likely higher.
83% of AI Overview searches end without a click, and AI Mode refers traffic at a tenth of Google's rate
Similarweb's June 2026 analysis puts the no-click rate on AI Overview searches at 83%. AI Mode, still only 0.34% of Google searches, refers traffic onward at 1.6–2.5% against 17–19% for traditional results.
Publishers expect a 43% decline in search traffic within three years
Reuters Institute's 2026 trends and predictions survey found publishers forecasting a 43% fall in search referrals over three years, with a fifth expecting losses above 75% (coverage).
Why zero-click figures range from 68% to 83%
Total searches and AI Overview searches are different populations. 68.01% describes every US Google search; 83% describes only those with an AI Overview attached. Blending them describes something that doesn't exist — and the older 92–94% figure for AI Mode that circulates widely is May–July 2025 data, from a study whose original URL is now dead.
Do rankings still drive AI citations?
Only 38% of AI Overview citations come from Google's top 10
Ahrefs analysed 863,000 keyword SERPs and 4 million AI Overview URLs in a study published in March 2026. 37.9% of citations came from top-10 results, 31.2% from positions 11–100, and 31.0% from pages ranking nowhere in the top 100. On an organic-only view, more citations came from outside the top 100 (36.7%) than from the top 10 (37.1%).
For comparison, Ahrefs' July 2025 run of the same measurement put top-10 pages at 76.1%. That baseline is 2025 data and is included only to size the change.
| Ranking band | Jul 2025 baseline | Mar 2026 |
|---|---|---|
| Top 10 | 76.1% | 37.9% |
| 11–100 | 9.5% | 31.2% |
| Outside top 100 | 14.4% | 31.0% |
Ahrefs doesn't publish a collection window for the 2026 study, so it's dated by publication. Two different samples also make the exact size of the drop less reliable than its direction — and the direction is unambiguous. Ranking in the top 10 has gone from a near-prerequisite for citation to roughly a coin flip.
18% of citations from non-ranking pages are YouTube URLs
From the same study: among AI Overview citations pointing at pages that don't rank in Google's top 100, 18% are YouTube. A meaningful share of AI visibility is being earned on a platform most SEO programmes neither own nor measure.
Google now cites itself in 17.42% of AI Mode citations
SE Ranking analysed 1,321,398 citations across 68,313 keywords collected on 12 February 2026. Google.com accounts for 17.42% of all AI Mode citations, up from 5.7% in June 2025 — roughly tripled in eight months. Of those self-citations, 59% point to organic search results and 36.1% to Google Business Profiles.
ChatGPT cites around 15 sources per response; Gemini cites about 3
Semrush's 2026 AI Visibility Index analysed 126 million US AI search prompts between January and April 2026 across ChatGPT, Gemini, AI Mode and AI Overviews.
Two findings stand out beyond the citation counts. On Gemini, the overlap between brands mentioned and brands cited runs as low as 30% — your brand can be named in an answer without being linked, which is invisible to anything measured through referral traffic. And only 36 brands globally held top-100 visibility across all four platforms in every month of the study.
That gap is why tracking the same prompt across every model gives a different picture than analytics does. It's also worth understanding how ChatGPT decides what to retrieve and cite, since its selection behaviour is the least Google-like of the major surfaces.
There is no 2026 data on cross-platform citation overlap
Every published comparison of how much ChatGPT, Gemini, Perplexity and AI Mode citations overlap with Google's top 10 uses data collected in mid-2025. Given that the AI Overviews figure alone halved in the eight months after that, those overlap numbers should be treated as historical rather than current.
Citation visibility is far less stable than rankings
This is the finding with no equivalent in traditional SEO, and it deserves more attention than it gets.
ChatGPT citation volumes fell 86–94% across five markets between February and April 2026
seoClarity tracked citations across the US, UK, Canada, Germany and Italy from 8 February to 27 April 2026, with monitoring continuing into May. Citation volumes fell 86–94% across all markets over that period, with a single drop of more than 80% around the 19 April update.
In the US, the zero-citation rate — answers containing no citations at all — doubled in March, from 28% to 48%. In Germany it reached 85% by April.
Citations then rebounded in May and had recovered toward pre-March levels by June.
What that means for measurement
Nothing about those sites changed. No content was removed, no rankings moved. A model update altered how often ChatGPT cited anything at all, and every brand's visibility moved with it.
Two practical consequences. A month-over-month drop in AI citations may be a platform event rather than a content problem, and reading it as the latter leads to work that fixes nothing. And any AI visibility measurement based on a single snapshot is close to meaningless — trend over weeks is the only reliable read.
What correlates with getting cited
Traditional SEO metrics predict AI topic ownership barely better than chance
Semrush's ChatGPT topic authority study tracked 50,000+ brands across 1,094 US categories, 220,000+ domains and 600,000+ citations, monthly from January to June 2026.
| Signal | How often the topic owner ranked higher on it |
|---|---|
| Branded search volume | 55.7% of pairs |
| Authority Score | 52.5% of pairs |
| Organic traffic | 48.4% of pairs |
A coin flip is 50%. Organic traffic performed slightly worse than one. Semrush's own conclusion is that domain-level metrics "correlate with ownership only about half the time," and that deep topical coverage plus relevant mentions were the stronger determinants.
This is worth reading carefully rather than as "SEO is dead." It says something narrower and more useful: the metrics that summarise a domain don't predict which brand owns a topic in AI answers. Topic-level coverage does.
Only 15.2% of categories have a clear AI topic owner
From the same study: 15.2% of the 1,094 categories analysed had a clear owner, 31.2% had an emerging leader, and 53.7% were unsettled. Among high-volume topics, only 11.3% had a clear owner.
But where ownership exists, it's sticky — owners held position in 90.4% of month-over-month comparisons. Most categories are still open, and the ones that close appear to stay closed.
75% of pages cited by LLMs were updated within the last year
Seer analysed 7,683 dated pages and 47,097 citations across ChatGPT, Gemini and Perplexity between March and June 2026. Three-quarters of cited pages had been updated within a year; 88% within two.
The more actionable finding is which date matters. Judged by last update, 72% of cited pages looked fresh. Judged by original publish date, only 42% did. Refreshing works — but only if the updated date is actually exposed on the page.
Tolerance varies by engine: Gemini is strictest at 78% of citations updated within a year, ChatGPT 73%, Perplexity most forgiving at 65%.
Community platforms take 52.5% of AI citations, against 47.5% for brand-owned domains
Otterly.ai analysed over one million citations across ChatGPT, Perplexity and Google AI Overviews in January and February 2026. Community platforms — chiefly Reddit and Quora — took the majority. News and media sites accounted for 20.3%.
Brand-domain preference varies sharply by platform: Google AI Overviews 59.8%, ChatGPT 44.7%, Perplexity 28.9%. Optimising your own site is a materially better strategy on AI Overviews than on Perplexity, where more than seven citations in ten point somewhere you don't control.
The same study found reference-grade, chunked content earning 3–5× more citations, and 73% of sites carrying technical barriers that block AI crawler access.
Wikipedia and Reddit together supply over a quarter of US ChatGPT citations
5W Public Relations' Q1 2026 Citation Source Audit, covering data through April 2026, puts Wikipedia at 13.15% of US ChatGPT citations and Reddit at 11.97%. LinkedIn accounts for 14.3% of ChatGPT Search responses.
The Wall Street Journal, New York Times, Bloomberg and Financial Times don't appear in the top 20 at all. Whatever ChatGPT treats as authoritative, it isn't legacy publishing brands — which makes where community citations actually come from a different question from which publications carry prestige.
If citation rate is the metric you're working on directly, there's a fuller treatment of growing your citation rate.
What doesn't work
Two tactics have now been tested at scale with 2026 data. Both come back empty, and both are still standard advice.
97% of llms.txt files received zero traffic
Ahrefs checked 137,210 domains in May 2026. 28% published an llms.txt file; 97% of those received no requests at all that month.
Of the requests that did arrive, 96% came from bots — mostly SEO audit tools (21.7%), unknown crawlers (14.9%) and tech-profiling services (11.6%). AI retrieval bots accounted for 1.1%.
The detail that settles it: zero requests arrived from AI bots for llms.txt files that don't exist. The bots aren't checking. We reached the same conclusion on llms.txt before data at this scale existed, and everything since has pointed the same way.
Adding schema markup produced no citation lift, and a small negative effect in AI Overviews
Ahrefs ran the only controlled study on this: 1,885 pages that added JSON-LD between August 2025 and March 2026, matched against 4,000 control pages drawn from an initial pool of 6 million URLs, using difference-in-differences across 30-day pre and post windows.
| Platform | Effect of adding schema |
|---|---|
| Google AI Overviews | −4.6% |
| Google AI Mode | +2.4% |
| ChatGPT | +2.2% |
Independent corroboration: searchVIU tested five AI systems and found none consume JSON-LD during direct retrieval — they extract visible HTML.
The authors' own caveat belongs with it. The tested pages were already heavily cited; schema may still help pages not yet in the consideration set, and that case is untested. Schema also retains its value for traditional rich results. It just isn't the AI citation lever it was widely sold as.
AI referral traffic and what it's worth
AI referral traffic grew 9.9× in 19 months, and 92.4% of it is ChatGPT
Previsible tracked 166 GA4 properties and 6.77 million LLM sessions through May 2026, across SaaS, ecommerce, finance, legal, health, insurance, education, publishing and ticketing. Monthly LLM referral sessions went from 65,249 to 644,478.
The platform split has consolidated hard. ChatGPT grew 12.8× and now supplies 92.4% of trackable LLM referral traffic. Claude grew 64× and overtook Perplexity in March 2026. Perplexity is down 61% from its peak; Copilot down 96%.
AI traffic to US retailers converted 42% better than non-AI traffic in March 2026
Adobe Analytics measured over one trillion visits to US retail sites, plus a survey of 5,000+ US consumers. AI referral traffic to US retailers rose 393% year over year in Q1 2026 (TechCrunch coverage).
The quality flip is the more important half. Engagement rate up 12%, time on site up 48%, pages per visit up 13%, revenue per visit up 37%. In March 2025 the same traffic converted 38% worse than non-AI traffic. The reversal took twelve months.
Only 66% of US retail product pages are machine-readable
Same study, scoring AI content visibility: homepages 75%, category pages 74%, product pages 66%. Best retailer 82.5%, worst 54.2%.
The pages closest to revenue are the least readable by the systems now sending the best-converting traffic. That's fixable, and it starts with seeing what the crawl-to-citation path looks like on your own site rather than assuming pages that render for humans render for agents.
49% of US adults use AI chatbots, and 42% of them use one to search for information
Pew Research surveyed 5,119 US adults between 17 and 23 February 2026. Chatbot use rose from a 33% baseline in 2024. ChatGPT leads at 44%, then Gemini (24%), Copilot (17%) and Meta AI (14%). 24% use them daily.
The other half of that finding stays relevant: 51% of US adults don't use AI chatbots at all.
How the impact varies by market and vertical
| Measure | Range across categories | Source |
|---|---|---|
| Commercial-intent AI Overview growth, Nov 2025–Apr 2026 | Finance +231% to transactional overall −5% | Semrush |
| ChatGPT citation rate by vertical, May 2026 | Travel & Hospitality ~23%, Automotive ~20%, Professional Services <4% | Similarweb |
| Top-3 brand share of AI visibility, Jan–Apr 2026 | News & Media 82.9%, Consumer Electronics 76.9%, Industrial 42.2%, Finance 41.4% | Semrush |
| ChatGPT zero-citation rate, April 2026 | US 48%, Germany 85% | seoClarity |
Two reads. Finance and industrial are where an unknown brand can still break in; media and consumer electronics are effectively closed. And the market differences in citation behaviour are large enough that US benchmarks shouldn't be applied to European sites, or the reverse.
Methodology
The rule: every statistic on this page comes from research whose data collection window ends in calendar 2026. Publication date in 2026 is not sufficient — several widely-cited "2026" studies run on 2025 data, and they're excluded here.
Baselines: where a pre-2026 figure appears, it's labelled as a baseline for measuring a 2026 change — the 2025 top-10 citation share, the 2024 zero-click rate, and similar — and never presented as a current figure.
Conflicts: where studies disagree, both figures appear with the methodological reason. Nothing is averaged.
Flagged inclusions: two studies here don't publish a collection window — Ahrefs' March 2026 citation analysis and SISTRIX's German CTR study. Both are dated by publication and labelled as such. Otterly's report carries a publication date that predates the end of its stated study window; it's cited as early 2026.
Exclusions: the widely-quoted 58% position-one CTR loss (data ends December 2025), the 75,000-brand correlation study (December 2025), all cross-platform citation overlap figures (mid-2025), the 8%/15% Pew click-rate comparison (March 2025), and the 92–94% AI Mode zero-click figure (July 2025).
Review cadence: quarterly. Given that ChatGPT citation volumes moved 90% and back inside four months, anything older than a quarter on this page should be assumed to have moved.
Frequently asked questions
Do rankings still matter for AI search?
Less than they did. In Ahrefs' March 2026 analysis of 863,000 SERPs, 38% of AI Overview citations came from top-10 results, against 76% in mid-2025 — and roughly as many citations came from pages ranking nowhere in the top 100. Rankings still help. They're no longer the mechanism.
How much traffic do AI Overviews actually cost?
On US data through February 2026, click-through on AI-Overview queries ran at about 2.4% against 3.3% for queries without one — and being cited inside the answer roughly doubles your share of what's left. Larger figures circulate, but the most-quoted ones use 2025 data or non-US markets.
What percentage of US searches end without a click?
68.01% in early 2026, against a 60.45% baseline in 2024, per SparkToro and Similarweb clickstream data. On AI Overview searches specifically, Similarweb puts it at 83%.
Is AI search traffic worth less than organic traffic?
Not per visit, and the evidence flipped recently. AI traffic to US retailers converted 42% better than non-AI traffic in March 2026, having converted 38% worse a year earlier. Volume is the constraint, not quality.
Does llms.txt help with AI search visibility?
There's no evidence it does. Across 137,210 domains, 97% of llms.txt files received zero traffic in May 2026, and AI retrieval bots accounted for 1.1% of the requests that did arrive.
Does schema markup increase AI citations?
Not according to the only controlled study on it. Ahrefs found a statistically significant −4.6% effect in AI Overviews and no measurable effect in AI Mode or ChatGPT. Schema retains its value for traditional rich results.
Why do my AI citations fluctuate so much?
Because the platforms change underneath you. Between February and April 2026, ChatGPT citation volumes fell 86–94% across five markets and the US zero-citation rate doubled from 28% to 48% in a single month — then recovered by June. Nothing about the affected sites changed. Track the trend over weeks; a single snapshot tells you almost nothing.
How do I track whether my brand is cited in AI answers?
You need prompt-level tracking across models, because the same question returns different sources on different platforms and changes between runs. Analytics alone won't show it: on Gemini, brand mentions and citations overlap by as little as 30%, so much of your visibility never produces a referral to measure. There's a comparison of the tools that do this if you're evaluating options.
How old is too old for an AI search statistic?
About two quarters. Three of the largest figures on this page reversed direction inside six months: AI Overview click-through, top-10 citation overlap, and ChatGPT citation volumes. Anything from 2024 describes a different product.
