Generative engine optimization statistics at a glance
| Category | Statistic | Source |
|---|---|---|
| Clicks | Users click a traditional result on 8% of visits to result pages with an AI summary, against 15% without | Pew Research Center, July 2025 |
| Clicks | AI Overview presence correlates with a 58% lower CTR for the top-ranking page, up from 34.5% a year earlier | Ahrefs, February 2026 |
| Citations | YouTube takes 22.9% of AI Overview mentions, Reddit 18.5%, Facebook 10.1% | Ahrefs, September 2026 |
| Citations | Reddit was cited in about 60% of ChatGPT answers in early August 2025 and about 10% six weeks later | Semrush, November 2025 |
| Overlap | Only 12% of URLs cited by AI assistants rank in Google's top 10; about 80% rank nowhere in the top 100 | Ahrefs, August 2025 |
| Overlap | AI Overview citations pulled from the top 10 fell from 76.1% to 37.9% in eight months | Ahrefs, July 2025 and March 2026 |
| Signals | Branded web mentions correlate 0.664 with ChatGPT visibility; Domain Rating manages 0.326 and links less | Ahrefs, December 2025 |
| Traffic | AI assistants sent 770.7 million referral visits a month worldwide, up 117.4% year over year | Similarweb, September 2026 |
| Usage | Gemini has 950 million monthly users; AI Mode has passed 1 billion | Alphabet, July 2026 |
| Research | The paper that named GEO tested 9 methods on 10,000 queries and reported up to 40% more visibility | Aggarwal et al., arXiv, 2023 |
How often do AI answers appear, and what do they cost in clicks?
AI Overviews appeared on 18% of US Google searches in March 2025, and every independent measurement since agrees they suppress clicks. The size of the loss ranges from 34.5% to 61% depending on method.
| Statistic | Source |
|---|---|
| AI Overviews appeared on 18% of US Google searches in March 2025. On searches starting with who, what, when or why, 60% produced one. | Pew Research Center, July 2025 |
| Users clicked a traditional result on 8% of visits to pages with an AI summary, against 15% without, and clicked a source inside the summary on 1% of visits. Sample: 900 US adults, 68,879 searches. | Pew Research Center, 2025 |
| The top-ranking page had a 34.5% lower CTR when an AI Overview was present, across 300,000 keywords. 99.2% of AI Overview keywords were informational. | Ahrefs, April 2025 |
| Ahrefs reran the same 300,000-keyword study (December 2023 against December 2025) and found the gap had widened to 58%. | Ahrefs, February 2026 |
| Organic CTR on AI Overview queries fell 61% (1.76% to 0.61%) and paid CTR 68% (19.7% to 6.34%), June 2024 to September 2025. Sample: 3,119 queries at 42 organisations. | Seer Interactive via Search Engine Land, 2025 |
| Brands cited inside the AI Overview earn 120% more organic clicks per impression than brands that are not. Sample: 53 brands, 5.47 million queries, 2.43 billion impressions. | Seer Interactive, April 2026 |
| 68.01% of US Google searches ended without a click in January to April 2026, up from 60.45% in 2024. Only 0.34% of searches reached Google's AI Mode. | SparkToro / Similarweb, June 2026 |
| Google says AI Overviews drive over a 10% increase in search usage for the queries showing them in the US and India. No sample or method is published. | Google, May 2025 |
| Google states that total organic click volume from Search to websites is "relatively stable" year over year, and has published no figure, percentage or method to support it. | Google, August 2025 |
The click loss stays a range on purpose. Three teams measured three different things and landed in the same direction, which is the finding. The most-quoted claim in this category, Gartner's predicted 25% fall in search volume by 2026, is missing because I could not find a public Gartner document stating it.
Which sources do LLMs cite?
YouTube, Reddit and Wikipedia dominate every published citation set, but the concentration is less settled than it looks: the top three sources hold 5.6% to 51.5% of citations depending on whether the study counts every citation or normalises against the top 50 domains.
| Statistic | Source |
|---|---|
| In Google AI Overviews, YouTube takes 22.9% of mentions, Reddit 18.5%, Facebook 10.1%, Google 8.8%. The top 3 domains hold 51.5%. Sample: over 3 million US queries, September 2026. Mention share is measured against the combined citations of the top 50 domains. | Ahrefs, September 2026 |
| In June 2025 the leading domain was Wikipedia on ChatGPT (16.3%), YouTube on Perplexity (16.1%) and YouTube in AI Overviews (9.5%). Sample: 76.7 million AI Overviews, 957,000 ChatGPT prompts, 953,500 Perplexity prompts. | Ahrefs, June 2025 |
| Measured against all citations rather than a top-50 subset, the leading domain holds 7.8% on ChatGPT (Wikipedia), 6.6% on Perplexity (Reddit) and 2.2% in AI Overviews (Reddit). Sample: 680 million citations, August 2024 to June 2025. | Profound, June 2025 |
| Reddit was cited in roughly 60% of ChatGPT answers in early August 2025 and roughly 10% by mid-September; Wikipedia fell from about 55% to under 20% in the same window. Sample: 230,000 prompts and over 100 million citations across 13 weeks, July to October 2025. | Semrush, November 2025 |
| ChatGPT cites 88.46% of the URLs it retrieves from search, 12.01% of news URLs, 1.93% of Reddit URLs, 0.51% of YouTube URLs and 0.40% of academic URLs. Sample: 1.4 million prompts. | Ahrefs, April 2026 |
The Reddit collapse is the most useful fact here. In six weeks ChatGPT went from citing Reddit in three answers out of five to one in ten, with no announcement. Anyone who built a GEO programme on "get mentioned on Reddit" in August 2025 had a dead strategy by October.
Do AI citations follow Google rankings?
Partly, and it depends on which surface you measure. Across the six studies with a stated sample, an average of 51.7% of AI-cited URLs also rank in Google's top 10, but the range runs from 12% to 82%.
| Statistic | Source |
|---|---|
| Only 12% of URLs cited by AI assistants also rank in Google's top 10 for the same query, and about 80% rank nowhere in the top 100. Sample: 15,000 long-tail queries across four assistants, July 2025. | Ahrefs, August 2025 |
| By assistant, top-10 overlap was 28.6% for Perplexity, 8.6% for Copilot, 8.2% for ChatGPT in-text citations and 8.0% for Gemini. | Ahrefs, 2025 |
| 76.1% of AI Overview citations came from pages ranking in Google's top 10; 14.4% came from pages that ranked nowhere. Sample: 1.9 million citations from 1 million AI Overviews. | Ahrefs, July 2025 |
| Eight months later the same measurement gave 37.9% from the top 10 and 31.0% from beyond the top 100. Sample: 863,000 keyword SERPs, 4 million AI Overview URLs. | Ahrefs, March 2026 |
| Google AI Mode citations overlapped Google's top 10 on 35% of URLs and 54% of domains. Sample: 5,000 keywords and over 150,000 citations. | Semrush, July 2025 |
| In the same study Perplexity showed 82% URL and 91% domain overlap, AI Overviews 67% and 86%. ChatGPT had the lowest overlap of the four surfaces. | Semrush, July 2025 |
| The share of AI Overview citations that rank organically anywhere rose from 32.3% in May 2024 to 54.5% in September 2025 across nine industries, but only 16.7% came from the organic top 10. Most of the growth came from positions 21 to 100. No sample size is published. | BrightEdge, September 2025 |
| Industry convergence ranges from 75.3% in healthcare to 19.2% in restaurants. No sample size is published. | BrightEdge, 2025 |
Two things are true at once and most GEO advice picks one. Ranking is a strong shortcut into Google's own AI surfaces, where over half of citations come from pages that rank somewhere. It is a weak shortcut into ChatGPT, where four in five cited URLs miss the top 100 entirely.
What predicts whether an LLM cites you?
Brand mentions across the open web, not backlinks. Ahrefs' study of 75,000 brands found branded web mentions correlate 0.664 with ChatGPT visibility, while Domain Rating manages 0.326 at best and backlinks are weaker still.
| Statistic | Source |
|---|---|
| Spearman correlation between branded web mentions and brand visibility: 0.709 in Google AI Mode, 0.664 in ChatGPT, 0.656 in AI Overviews. Sample: 75,000 brands with Domain Rating above 40. | Ahrefs, December 2025 |
| Classic SEO metrics correlate far more weakly. Domain Rating reaches 0.326 in AI Overviews, 0.285 in AI Mode and 0.266 in ChatGPT. Ahrefs calls the backlink and URL rating correlations very weak and publishes no coefficient for either. Content volume sits at about 0.194. | Ahrefs, December 2025 |
| URLs with natural-language slugs were cited by ChatGPT 89.78% of the time against 81.11% for URLs without. The median cited page was about 500 days old. Sample: 1.4 million prompts. | Ahrefs, April 2026 |
A 0.664 correlation is not a mechanism. What makes it worth acting on is the gap: mentions beat the best link metric roughly two to one, on the same dataset, on all three surfaces. That points at digital PR and review sites rather than another link-building retainer.
How much traffic does AI search send?
Not much yet: AI assistants send roughly 6 referral visits for every 1,000 clicks Google's organic results still send to the open web.
| Statistic | Source |
|---|---|
| AI platforms sent an average of 770.7 million referral visits a month worldwide between June 2025 and May 2026, up 117.4% year over year. Every tracked industry doubled. | Similarweb, September 2026 |
| ChatGPT's share of worldwide generative AI web traffic fell from 76% to 53% over that year while Gemini rose from under 9% to 27 to 28%. | Similarweb, July 2026 |
| The share of US ChatGPT answers containing web citations rose from 1.6% to 6.8% over the same year, hitting about 23% in travel and under 4% in professional services. | Similarweb, July 2026 |
| LLM referrals were 0.08% of total traffic for publishers and 1.71% for SMB sites in May 2026. Sample: 166 GA4 properties and 6.77 million LLM sessions. | Previsible, July 2026 |
| AI search accounts for under 1% of referral traffic across the sites BrightEdge tracks, Fortune 100 brands included. No sample size is published. | BrightEdge, March 2026 |
| AI search visitors converted 4.4 times better than traditional organic visitors across 500-plus digital marketing topics. | Semrush, June 2025 |
| Bain attributes a 15% to 25% reduction in organic web traffic to AI search behaviour, with about 60% of searches ending without a click. No sample size is published. | Bain & Company, February 2025 |
| Gemini has 950 million monthly active users and AI Mode has passed 1 billion, per Alphabet's Q2 2026 earnings remarks. Gemini daily users tripled year over year. | Alphabet, July 2026 |
| Anthropic's crawlers made 70,900 HTML requests for every referral they sent back to a site in late June 2025. | Cloudflare, July 2025 |
The gap between attention and volume is the whole story. A billion people use AI Mode, and AI referrals are still under 1% of traffic almost everywhere a real panel has measured it. Fund GEO from the brand budget, not from the channel paying your bills today.
What did the original GEO paper measure?
The 2023 paper that named generative engine optimization tested 9 methods on a 10,000-query benchmark and reported visibility gains of up to 40%, with the largest lifts from adding quotations, statistics and cited sources.
| Statistic | Source |
|---|---|
| GEO-bench contains 10,000 queries and the paper tested 9 optimization methods. | Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, arXiv 2311.09735 |
| Headline result: GEO methods boosted visibility by up to 40% on the position-adjusted word count metric. | Aggarwal et al., 2023 |
| Quotation addition, statistics addition and cite sources each gained 30% to 40% on the position-adjusted word count metric, the paper's own band. Cited sources lifted 5th-ranked websites by 115.1% and cut 1st-ranked visibility by 30.3%. | Aggarwal et al., 2023 |
| Submitted November 2023, last revised June 2024. The measurements were taken on 2023-era generative engines. | arXiv 2311.09735 |
This is the one deliberately old set of numbers here, included because most GEO articles quote the 40% without saying when it was measured or on what. It predates AI Mode, ChatGPT Search and the citation infrastructure the rest of this page describes.
The paper is good work. The marketing built on top of it is not. A 40% lift measured on 2023 engines, on a synthetic benchmark, gets recycled in 2026 as though someone tested it on ChatGPT last week, and nobody has published a replication. Stop quoting it as a current benchmark.
How we calculated the original numbers
Four figures here appear nowhere else. Here is the arithmetic, so you can check it or change the inputs.
- 23.5% median top-3 citation concentration. I took every measurement of top-three citation share from the three studies that publish their method and sample. Ahrefs September 2026, AI Overviews: 51.5%. Ahrefs June 2025: 25.3% (AI Overviews), 23.5% (ChatGPT), 31.3% (Perplexity). Profound, 680 million citations: 10.7% (ChatGPT), 9.6% (Perplexity), 5.6% (AI Overviews). Sorted, those seven values are 5.6, 9.6, 10.7, 23.5, 25.3, 31.3 and 51.5, so the median is 23.5%, and it is a median across mixed denominators rather than a measurement of the web: Ahrefs normalises against the top 50 domains, Profound counts every citation. Third-place shares in the June 2025 set are read off chart labels, so treat the median as good to about half a point.
- 51.7% mean top-10 overlap for AI-cited URLs. Six measurements have a stated sample: Ahrefs assistants 12%, Ahrefs AI Overviews July 2025 76.1%, Ahrefs AI Overviews March 2026 37.9%, Semrush AI Mode 35%, Semrush AI Overviews 67%, Semrush Perplexity 82%. (12 + 76.1 + 37.9 + 35 + 67 + 82) / 6 = 51.7%. Split by surface, Google's own AI products average 54.0% and standalone assistants 13.4%. BrightEdge's 16.7% sits near the bottom but is excluded because no sample size is published. One caution: the 12% averages four assistants while the other five are single surfaces, so the mean under-weights standalone assistants.
- 6.4 AI referral visits per 1,000 Google organic open-web clicks. Numerator: Similarweb's 770.7 million average monthly AI referral visits worldwide (June 2025 to May 2026). Denominator: Google states more than 5 trillion searches a year, which is 416.7 billion a month, and SparkToro measured 374 open-web clicks per 1,000 US searches in 2024 and a 22.9% fall in click rate by early 2026, so 374 x 0.771 = 288 per 1,000. 416.7 billion x 0.288 = 120.0 billion organic clicks a month. 770.7 million / 120.0 billion = 0.642%, or 6.4 per 1,000. Three mismatches to watch. Google's search count is global, SparkToro's click rate is US. The inputs cover different windows: May 2025 for Google, January to April 2026 for SparkToro, June 2025 to May 2026 for Similarweb. And SparkToro publishes the 22.9% fall on its any-click metric, not on open-web clicks alone, so 288 is an estimate.
- 6.5 years for AI referrals to match Google's organic clicks. AI referrals would need to grow 156 times (120.0 billion / 770.7 million). At Similarweb's measured 117.4% annual growth, that is a factor of 2.174 a year, so ln(156) / ln(2.174) = 6.50 years from the end of Similarweb's window in May 2026, or late 2032. That assumes growth never slows, which it already has for ChatGPT.
What the 2026 numbers actually say
GEO is two jobs wearing one name. Getting into Google's AI Overviews is mostly downstream of ranking: 37.9% to 76.1% of those citations come from the top 10. Getting into ChatGPT is not: 8.2% of its in-text citations rank in the top 10. Anyone selling one GEO service for both surfaces is selling you one of them.
The citation graph is unstable and nobody is warning you. ChatGPT cited Reddit in three answers out of five, then one in ten six weeks later. AI Overview citations from the top 10 halved in eight months. Optimising for the current shape of a citation set is optimising for a photograph. The signals that held steady are boring: be mentioned widely, answer directly, cite your sources.
The traffic case is weak and the intent case is strong. Under 1% of referral traffic, converting several times better. Both halves are load-bearing. Fund GEO by promising traffic and you will be embarrassed in two quarters. Fund it as brand distribution into the systems 950 million people now ask first, and the numbers point the same way.
FAQ
What is generative engine optimization?
Generative engine optimization is the practice of getting content cited inside AI-generated answers. The term comes from a 2023 paper by Aggarwal and colleagues (arXiv 2311.09735), which tested 9 methods on a 10,000-query benchmark and reported visibility gains of up to 40%, with quotations, statistics and cited sources all landing in the 30% to 40% band.
Do AI search engines cite the pages that rank on Google?
Sometimes. Averaged across the six studies with a stated sample, 51.7% of AI-cited URLs also rank in Google's top 10, and the range runs from 12% to 82%. Ahrefs found 12% overlap across 15,000 queries on standalone assistants (2025); Semrush found 82% for Perplexity (2025).
Which websites do LLMs cite most?
YouTube, Reddit, Wikipedia and Facebook dominate. In Google AI Overviews, YouTube takes 22.9% of mentions and Reddit 18.5% (Ahrefs, September 2026). On ChatGPT, Wikipedia led with 16.3% (Ahrefs, June 2025).
How much traffic does AI search send to websites?
AI platforms sent 770.7 million referral visits a month worldwide in the year to May 2026, up 117.4% (Similarweb). That is roughly 6 visits per 1,000 organic clicks Google still sends to the open web, or 0.08% to 1.71% of a site's traffic (Previsible, 2026).
Do AI Overviews reduce organic clicks?
Yes, by 34.5% to 61% depending on method. Ahrefs measured a 58% lower CTR for the top-ranking page in February 2026, up from 34.5% in April 2025. Pew found clicks fell from 15% to 8% of visits (2025), and Seer measured a 61% fall in organic CTR (2025).
What signals make an LLM cite a brand?
Brand mentions across the web, more than links. Branded web mentions correlate 0.664 with ChatGPT visibility and 0.709 with AI Mode, against 0.326 for Domain Rating and less for backlinks, across 75,000 brands (Ahrefs, December 2025).
Sources
- Pew Research Center, Google users click fewer links when an AI summary appears (July 2025)
- Ahrefs, AI Overviews Reduce Clicks by 34.5% (April 2025)
- Ahrefs, AI Overviews Reduce Clicks: 2026 Update (February 2026)
- Seer Interactive via Search Engine Land, AI Overviews and CTR (2025)
- Seer Interactive, AIO Impact on Google CTR: 2026 Update (April 2026)
- SparkToro / Similarweb, Less than a third of Google searches send a click (June 2026)
- Google, More opportunities on Google Search (May 2025)
- Google, AI in Search is driving more queries (August 2025)
- Ahrefs, The 50 Most-Cited Websites in Google AI Overviews (September 2026)
- Ahrefs, The 10 most mentioned domains across assistants (June 2025)
- Profound, AI Platform Citation Patterns (June 2025)
- Semrush, The Most-Cited Domains in AI: A 3-Month Study (November 2025)
- Ahrefs, Why ChatGPT Cites One Page Over Another (April 2026)
- Ahrefs, Only 12% of AI Cited URLs Rank in Google's Top 10 (August 2025)
- Ahrefs, 76% of AI Overview Citations Pull From the Top 10 (July 2025)
- Ahrefs, Update: 38% of AI Overview Citations Pull From The Top 10 (March 2026)
- Semrush, How Google's AI Mode compares to search (July 2025)
- BrightEdge, AI Overview citations and organic rankings (September 2025)
- Ahrefs, Top Brand Visibility Factors in ChatGPT, AI Mode and AI Overviews (December 2025)
- Aggarwal et al., GEO: Generative Engine Optimization, arXiv 2311.09735 (November 2023, revised June 2024)
- Similarweb, AI Referral Traffic by Industry (September 2026)
- Similarweb, AI Search Stats 2026 (July 2026)
- Previsible, AI Traffic Report (July 2026)
- BrightEdge, AI Search Visits Surging in 2025 (March 2026)
- Semrush, AI Search SEO Traffic Study (June 2025)
- Bain & Company, Consumer reliance on AI search results (February 2025)
- Alphabet, Q2 2026 earnings, CEO's remarks (July 2026)
- Cloudflare, The crawl before the fall of referrals (July 2025)
Methodology. Numbers were collected in September 2026 and checked against the original publisher's page, not against other statistics roundups. Vendor claims with no stated sample or method were excluded, and where a source publishes a claim with no supporting figure (Google on organic click volume, BrightEdge and Bain on their samples) the row says so. The widely quoted Gartner prediction of a 25% fall in search volume by 2026 is not here because no citable primary document could be located. No ChatGPT user count is quoted, because OpenAI's own pages refused automated requests. Company figures appear only where the company published them in a fetchable primary, as Alphabet did for Gemini. Where two credible studies disagree, both appear with their method and sample. Last updated 8 September 2026.