AI search statistics (2026)

AI search is widely used but still sends websites little traffic. In February 2026, 49% of U.S. adults said they use AI chatbots (opens in a new tab), and in July 2026 Google's CEO said AI Mode had passed 1 billion monthly active users (opens in a new tab). Yet in a 2025 study of 3,000 sites, 0.17% of the average site's visitors came from AI chatbots (opens in a new tab), and the same prompt rarely returns the same list of brands twice (opens in a new tab).

This page lists the figures most often needed to plan or judge work on AI search, which is also called answer engine optimization (AEO) or generative engine optimization (GEO). The figures are grouped by topic. AI search here means answers that AI systems write after searching the web: Google's AI Overviews and AI Mode, ChatGPT search, Gemini, Claude, Perplexity, and Microsoft Copilot. Each figure links to its original source, which was checked on September 27, 2026. The page is part of AEO HQ's research, and its sources were chosen under the evidence standard in our methodology.

How to read these statistics

Each entry gives four things:

1. The figure, linked to the original source.
2. The sample and date: who or what was measured, how many, and when.
3. The source type, one of the labels below.
4. A caveat, where one changes how the figure should be read.

Source types:

• Peer-reviewed: a journal article or conference paper that passed peer review.
• Preprint: a research paper that has not been peer reviewed. Theses and working papers are included here.
• Vendor, agency, or industry study (or survey): published by a company rather than a research institution. Where the company or an author sells a related product or service, the entry says so. Read these as descriptions, not as proof.
• Research center survey: a survey by a research organization that does not sell a related product, such as Pew Research Center.
• Analyst survey: published by a research and advisory firm such as Gartner or Forrester.
• Journalism research: a study run by a news organization or a journalism school.
• Company statement: a figure a company reports about its own product, usually without a published method.
• Network or panel measurement: traffic data from a hosting network, a browsing panel, or a clickstream panel.

Terms used below:

• Citation: a link or named source attached to an AI answer.
• Referral: a visit that arrives when someone clicks a link in an AI answer.
• Earned media: coverage that a company does not control, such as reviews, comparisons, and editorial articles.
• Jaccard similarity: the share of items that two lists have in common, from 0 (none) to 1 (identical).
• Standard error: the typical size of the error in an estimate.
• API: the programmatic access that developers use, as opposed to the consumer app that people use.

Figures are quoted with the conditions their sources attach to them. Figures that appear only in charts or only in secondary coverage are left out. The evidence window is September 2023 to September 2026.

Use of AI assistants and AI search

• 42% of U.S. adults said they use AI chatbots to search for information, the most common use in the survey (opens in a new tab). Overall, 49% said they ever use AI chatbots such as ChatGPT, Gemini, or Copilot, up from 33% in 2024. Survey of 5,119 U.S. adults on Pew Research Center's American Trends Panel, February 17–23, 2026 (research center survey). Caveat: the 2024 question was worded differently and was asked only of adults who had heard at least a little about chatbots. The 42% is a share of all adults, not of chatbot users.
• 44% of U.S. adults said they use ChatGPT, up from 34% a year earlier (opens in a new tab). Same survey.
• Six in ten U.S. adults said they read the AI summaries at the top of search results (opens in a new tab); three in ten said they do not, and 10% were not sure. Same survey. Caveat: self-reported.
• Google's CEO said AI Mode had passed 1 billion monthly active users, and that the Gemini app had 950 million (opens in a new tab). Alphabet's second-quarter earnings call, July 22, 2026 (company statement). Caveat: Google's own figures, with no published method.
• Only 0.34% of U.S. Google searches went to AI Mode from January to April 2026 (opens in a new tab). Similarweb's U.S. desktop and mobile panel, analyzed by SparkToro and published in June 2026 (panel measurement). Caveat: the author called the share "almost certainly growing fast."
• Average monthly visits to generative AI websites worldwide grew 70% year over year, to 9.5 billion, between June 2025 and May 2026 (opens in a new tab); unique visitors grew 57%, to 655 million. Similarweb's web traffic panel, published July 29, 2026 (vendor study). Caveat: the article does not describe the panel's method.
• ChatGPT's share of visits to generative AI websites fell from about 76% in June 2025 to about 53% in May 2026, while Gemini's rose from under 9% to about 27–28% and Claude's from about 2% to close to 9% (opens in a new tab). Same source. Caveat: these are shares of website visits. Similarweb's app data shows a different pattern.
• ChatGPT enabled web search on 34.5% of queries in February 2026, down from 46% in late 2024 (opens in a new tab), and the monthly share ranged from 15% to 66.3%. U.S. sessions in Semrush's clickstream panel, October 2024 to February 2026 (vendor study). Caveat: a panel of observed sessions, not OpenAI's own data.
• Citations appeared in about 6.8% of U.S. ChatGPT prompts in May 2026, up from about 1.6% in June 2025 (opens in a new tab). The rate was about 23% for travel and hospitality prompts and under 4% for professional services. Similarweb data, published July 29, 2026 (vendor study). Caveat: "citations present" measures something different from "search enabled"; see where the statistics disagree.
• Use of ChatGPT and other generative AI tools for local business recommendations rose from 6% to 45% of U.S. consumers in a year (opens in a new tab), making them the third most used source of such recommendations. Survey of 1,002 U.S. adults (SurveyMonkey panel), published February 11, 2026 (vendor survey; BrightLocal sells local search software). Caveat: consumers choosing local businesses, not B2B buyers.

How often Google shows AI Overviews

• 18% of Google searches in March 2025 produced an AI summary (opens in a new tab). The share was 60% for searches that began with a question word, 8% for one- or two-word searches, and 53% for searches of ten words or more. Browsing data from 900 U.S. adults, 68,879 unique Google searches (Pew Research Center; panel measurement). Caveat: the researchers re-ran the searches on April 7–17, 2025 to see which produced a summary, so the summary flag is reconstructed, not observed live.
• AI Overviews were generated for 51.5% of representative, real-user queries (opens in a new tab) in a public benchmark of 11,500 user queries (peer-reviewed; SIGIR 2026). Caveat: the share depends on the query set; it is not a share of all Google searches.
• AI Overviews appeared on 13.7% of 55,393 trending queries, and on 64.7% of queries phrased as questions (opens in a new tab), across 19 topics from March 13 to April 21, 2026 (preprint). Caveat: politically sensitive topics saw markedly lower rates.
• AI Overviews appeared on 36% of informational queries, 8% of commercial queries, and 5% of transactional queries (opens in a new tab) across 53 client accounts and 49,358 tracked queries, January 2025 to February 2026 (agency study). Caveat: the agency notes that its AI Overview label is static rather than tracked in real time.

Clicks when AI answers appear

• Google users clicked a traditional search result on 8% of visits when an AI summary appeared, against 15% when none appeared (opens in a new tab). They clicked a link inside the summary on 1% of visits, and ended their browsing session on 26% of pages with a summary, against 16% of pages without one. Same Pew panel, March 2025. Caveat: observational, and searches with and without summaries differ. A follow-up analysis of the same panel found that the association with fewer clicks held when controlling for the panelist and for query attributes (opens in a new tab) (preprint).
• The presence of an AI Overview correlated with a 58% lower average click-through rate for the top-ranking page (opens in a new tab). 300,000 keywords, December 2025 compared with December 2023, desktop click-through from aggregated Search Console data (vendor study). Caveat: the figure compares actual click-through with a forecast built from keywords without AI Overviews.
• On informational queries in the third quarter of 2025, organic click-through fell 65.2% year over year when the brand was not cited in the AI Overview, 49.4% when it was cited, and 46.2% on queries with no AI Overview (opens in a new tab). 3,119 search terms across 42 client organizations, June 2024 to September 2025 (agency study). Caveat: the agency says it cannot prove that being cited causes higher click-through, because stronger brands may simply be more likely to be cited.
• Organic click-through on queries with an AI Overview fell to 1.31% in December 2025 and rose to 2.36% in February 2026; on queries without one, it was 3.16% and 3.82% (opens in a new tab). 53 brands and 5.47 million queries (agency study). Caveat: a larger and different dataset from the agency's 2025 study above.
• 68.01% of U.S. Google searches ended without a click in January–April 2026 (opens in a new tab), against 60.45% in 2024. Similarweb's U.S. panel (panel measurement). Caveat: the 2024 figure came from a different panel (Datos), so the change mixes panel differences with real change.
• Default availability of AI Overviews reduced search referrals to English Wikipedia by 5.45% and 4.82% (opens in a new tab), compared with German and French Wikipedia (difference-in-differences design; preprint). Caveat: one publisher.

AI referral traffic to websites

How to count AI referral traffic in analytics is covered in a separate guide.

• 0.17% of the average website's visitors, and 0.12% of its page views, came from AI chatbots (opens in a new tab). 3,000 sites, published February 6, 2025 (vendor study). Caveat: Ahrefs notes that AI visits can be counted as direct traffic, so the true share is likely higher.
• ChatGPT referrals were less than 0.2% of all traffic, about 200 times smaller than Google's organic search (opens in a new tab). 973 e-commerce websites with more than 50,000 ChatGPT-referred transactions, August 2024 to July 2025 (peer-reviewed; Marketing Science). Caveat: the first author is affiliated with the company that supplied the data (Grips Intelligence).
• In a mostly B2B brand panel, ChatGPT's share of AI referrals fell from 89.1% (May–August 2025) to 62.6% (March–April 2026), Claude's rose from 1.4% to 18.5%, and Perplexity's from 3.1% to 7.3%; Gemini's reached 10.6% (opens in a new tab). GA4 referrer data from 41 brand sites in 2025 and a new brand panel in 2026, published May 21, 2026 (vendor study; Goodie sells AEO software). Caveat: the size of the 2026 panel is not given, and the 2026 figures average brands equally. Google's AI Overviews and AI Mode cannot be seen separately in GA4.
• After ChatGPT began showing brand names as links in May 2026, homepage visits rose from about 26–29% of ChatGPT's referral traffic to about 62–63% (opens in a new tab). Similarweb weekly referral data through late May 2026 (vendor study).
• First-touch attribution credited AI with only 28 of the 189 leads (15%) who named an AI tool when asked how they had heard of the company (opens in a new tab). One organic growth agency's own records, published August 28, 2026 (agency data). Caveat: a single company.

Conversion figures conflict:

• ChatGPT referrals converted above paid social but below every other traditional channel, and organic search had a 13% higher conversion likelihood (opens in a new tab). Same 973-site study (peer-reviewed).
• AI-referred visitors to U.S. retail sites converted 42% more than non-AI traffic in March 2026, after converting at nearly half the rate a year earlier (opens in a new tab). Adobe Digital Insights, based on visits to U.S. retail sites (vendor study). Caveat: a later period and a different sample from the peer-reviewed study.
• AI-referred visits to technology and software sites had a 30% higher engagement rate and a 40% lower bounce rate than other visits, with 40% more time and 23% more pages per visit (opens in a new tab). Same report, March 2026. Caveat: engagement, not purchases; the report states its sample only for retail sites.
• AI search brought 0.5% of Ahrefs' traffic and 12.1% of its signups over 30 days, a conversion rate 23 times that of traditional organic search (opens in a new tab). Published June 16, 2025. Caveat: one software company measuring its own site.

Where AI answers get their sources

• Google's organic results, AI Overviews, and Gemini retrieved substantially different sources, with an average Jaccard similarity below 0.2 (opens in a new tab). Benchmark of 11,500 user queries (peer-reviewed).
• On average, 53% of the domains that AI Overviews consulted were not in Google's organic top 10, and 27% were not in the top 100 (opens in a new tab). 4,706 queries issued in English in September 2025 (preprint).
• Nearly 30% of the domains cited in AI Overviews did not appear among the first-page results shown with them (opens in a new tab). 55,393 trending queries, March–April 2026 (preprint).
• In July 2025, 76.10% of pages cited in AI Overviews ranked in Google's top 10 (opens in a new tab) (1.9 million citations from 1 million AI Overviews, counting the top three citations in each). In March 2026, 37.9% of cited URLs appeared within the first 10 blocks of Google's results, or 37.1% counting organic rankings only (opens in a new tab) (863,000 keyword results pages and 4 million AI Overview URLs). Both vendor studies. Caveat: Ahrefs improved its citation parsing between the two studies, so they are not directly comparable.
• 28.6% of the URLs Perplexity cited ranked in Google's top 10 for the same prompt, against about 8% for the other AI assistants tested (opens in a new tab). 15,000 long-tail prompts, data collected in early July 2025 (vendor study). Caveat: the article's headline "12%" averages all five assistants, Perplexity included.
• 87% of ChatGPT search citations matched Bing's top organic results for the same question, against 56% for Google's results, at a median Google rank of 17 (opens in a new tab). About 100 queries and more than 500 citations, published February 6, 2025 (agency study). Caveat: the agency calls its sample limited and its findings directional.
• Of all the domains returned by LLM-based and traditional search engines, 37% appeared only in the LLM-based engines' results, and only 38% appeared in both (opens in a new tab). 55,936 queries taken from daily Google and X trends, July 16 to August 10, 2025, run on six LLM-based search engines, Google, and Bing (preprint). In the same study, 38% of Gemini's responses and 82% of Grok's contained no cited website (opens in a new tab).
• In API tests with ranking-style prompts, earned media made up 93.5% to 95.1% of the domains ChatGPT cited, 86.3% to 87.3% for Claude, 67.4% to 73.4% for Perplexity, and 63.4% to 66.4% for Gemini (opens in a new tab). Social platforms made up 17.5% to 23.8% of Perplexity's cited domains. Published September 2025 (preprint). Caveat: API versions, not consumer apps; each domain's type was labeled by an OpenAI model; the ranges span well-known and niche brands.
• Only 2.9% of 149,912 citations pointed to the tracked brand's own domain; ranked "best-of" listicles made up about 21%, Wikipedia 2.6%, and software review sites such as G2 and Capterra 1.1% (opens in a new tab). 102 brands on five AI engines, queried through their APIs, March–May 2026 (preprint). Caveat: the author co-founded and holds equity in the tracking platform studied.
• "Best X" blog lists made up 43.8% of the page types that ChatGPT cited (opens in a new tab) in answers to 750 prompts about software, products, and agencies. 26,283 source URLs, published December 4, 2025 (vendor study).
• Reddit appeared in about 62% of AI answers to decision-stage buying prompts, YouTube in about 25%, and G2 in about 5% (opens in a new tab). About 1,000 prompts and 57,095 citations from ChatGPT, Perplexity, Gemini, and Google's AI Overviews and AI Mode, collected January 29 to February 4, 2026 (agency study using a tracking vendor's data). Caveat: one week of data.
• Wikipedia, YouTube, and Reddit together made up 15% of the sources listed in Google's AI summaries, against 17% in standard results, and 88% of the summaries cited three or more sources (opens in a new tab). Pew panel, March 2025.
• URLs cited by ChatGPT averaged 958 days since publication, the newest of the platforms measured, while the top three AI Overview citations averaged 1,432 days, the oldest (opens in a new tab). 16.975 million cited URLs, published July 28, 2025 (vendor study). Caveat: the article's headline figure, that AI assistants cite pages 25.7% "fresher", uses 1,432 days for organic results, but the article's own table gives 1,416 days for organic results and 1,432 for AI Overviews.
• For the same query, the ChatGPT and Gemini interfaces shared only 5.4% of cited domains on average, with no domain in common in 76.7% of comparisons (opens in a new tab). 1,536 responses to product queries, published September 2026 (preprint).

How much AI answers vary between runs

• There is a less than 1 in 100 chance that ChatGPT or Google's AI, asked the same question 100 times, gives the same list of brands in any two responses; Claude is slightly more likely to (opens in a new tab). Seeing two lists in the same order took about 1 in 1,000 runs. 600 volunteers ran 12 prompts through ChatGPT, Claude, and Google's AI Overviews or AI Mode 2,961 times, November–December 2025 (industry study). Caveat: a co-investigator works for an AI visibility tracking vendor.
• About 65% of cited sources changed from one day to the next, and cited-source sets overlapped by a Jaccard similarity of only 0.32 to 0.43 between runs (opens in a new tab). Daily panel of German-language prompts on ChatGPT, Gemini, Google AI Mode, and Perplexity, January 24 to March 20, 2026 (preprint). Caveat: the lead author is affiliated with a tracking vendor.
• The standard error of a brand's detection rate was 0.370 with one run per prompt, 0.081 with seven runs, and 0.062 with eight (opens in a new tab). Same panel. 57.8% of ChatGPT's runs returned no citations (opens in a new tab).
• When the same queries were run two months apart, only 18% of the web pages in AI Overviews were the same in both runs, against 45% for organic results (opens in a new tab). Runs in July–August and September 2025 (preprint).
• 77.5% of brand, prompt, and engine combinations were either always or never mentioned across repeated runs, and whether a brand was framed positively or negatively flipped about 6.7 times more often than whether it was mentioned (opens in a new tab). 102,025 API responses, March–May 2026 (preprint; same conflict of interest as above).
• The API and the consumer interface of the same assistant shared only 12.0% (ChatGPT) and 14.8% (Gemini) of cited domains (opens in a new tab). Published September 2026 (preprint). Caveat: API tests are rough proxies for what buyers see.

Which brands AI answers name

• On their first visibility run, global household names appeared in 73% of relevant AI answers, established mid-market and regional brands in 44%, and niche and small brands in 11% (opens in a new tab). 100,000+ prompt responses across 100+ brands, March–May 2026 (preprint; same conflict of interest as above).
• YouTube mentions correlated at about 0.737 with a brand's visibility across ChatGPT, AI Mode, and AI Overviews, and branded web mentions at 0.709 (AI Mode), 0.664 (ChatGPT), and 0.656 (AI Overviews), while link metrics showed "very weak correlations" (opens in a new tab). 75,000 brands, published December 12, 2025 (vendor study). Caveat: correlation, not cause.
• Asked about startups by name, ChatGPT (gpt-4o-mini) and Perplexity recognized them 99.4% and 94.3% of the time, but in discovery-style questions they named them only 3.32% and 8.29% of the time (opens in a new tab). Visibility on Perplexity correlated with referring domains (r = +0.319) and with community presence on Reddit (r = +0.395), and websites' GEO scores showed no correlation with discovery (opens in a new tab). 112 startups from the 2025 Product Hunt leaderboard, 2,240 queries (preprint based on a master's thesis). Caveat: correlations, not causes.
• Well-known brands were recommended 100% of the time when all products had the same specifications, but this lead disappeared when a competitor had a rating advantage of less than 0.1 stars (opens in a new tab). Lab experiments with GPT-4o-mini, Claude Sonnet, and Gemini 3 Flash on skincare products (preprint).
• A top guest rating raised the chance that an AI assistant recommended a hotel by 31.6 percentage points, a high price lowered it by 30.0 points, and list position alone was worth about $12 per night (opens in a new tab). Randomized experiment across twelve models (preprint). Caveat: hotels, not B2B services.
• ChatGPT stated a first-person product preference in 79% of the responses that recommended products, against 7% for Gemini and 2% for AI Overviews (opens in a new tab). Same product-advice audit (preprint).

Accuracy of AI answers and citations

• On average, 51.5% of generated sentences were fully supported by their citations, and 74.5% of citations supported the sentence they were attached to (opens in a new tab). Human evaluation of Bing Chat, NeevaAI, perplexity.ai, and YouChat (peer-reviewed; Findings of EMNLP 2023). Caveat: the engines tested have since changed.
• 11.0% of 98,020 atomic claims in Google AI Overviews were not supported by the pages they cited, with omission the most common failure (opens in a new tab). March–April 2026 (preprint).
• Almost half of AI answers about the news had at least one significant issue, a third showed serious sourcing problems, and a fifth contained major accuracy issues (opens in a new tab). More than 3,000 responses from ChatGPT, Copilot, Gemini, and Perplexity, evaluated by journalists at 22 public service media organizations in 18 countries and 14 languages; published October 21, 2025 (journalism research). Caveat: news questions only.
• Eight AI search tools answered more than 60% of 1,600 queries incorrectly when asked to identify the source of a news excerpt; Perplexity answered 37% incorrectly and Grok 3 94% (opens in a new tab). Tests in February 2025, published March 6, 2025 (journalism research).
• Google's AI Overview and featured snippet on the same results page were inconsistent with each other in 33% of cases (opens in a new tab). 1,508 baby care and pregnancy queries (peer-reviewed; ICWSM 2026). Caveat: one health topic.
• 64% of B2B software buyers said they encounter inaccurate AI chatbot recommendations often or very often (opens in a new tab). Survey of 1,076 B2B software buyers and decision-makers, March 2026 (vendor survey; G2 runs a software review site).
• 27% of U.S. B2B professionals who use AI at work said AI vendor recommendations don't reflect real pricing or contract structures (opens in a new tab). 519 respondents, March–April 2026 (vendor survey).

B2B buyers and AI

• 51% of B2B software buyers now start their research with an AI chatbot more often than with Google, up from 29% in April 2025 (opens in a new tab). In the same survey, 71% said they rely on AI chatbots for software research, against 60% seven months earlier; 69% chose a different vendor than planned based on AI chatbot guidance; 33% bought from a vendor they were not familiar with; and 45% said citations from software review sites are the most confidence-inspiring signal in an AI answer. 1,076 B2B software buyers and decision-makers in North America, EMEA, and APAC, March 2026 (vendor survey). Caveat: G2 says generative AI models helped design the survey and analyze the results.
• 82% of B2B software buyers had sourced software recommendations from an AI chatbot in the previous 24 months, and review sites (38%) overtook AI chatbots (37%) as the top source influencing which vendors make the shortlist (opens in a new tab). Only 9% were comfortable letting AI agents make purchases within approved guardrails, and 2% without pre-approval (opens in a new tab). Survey of 1,038 B2B software decision-makers, June 2026 (vendor survey).
• 63% of technology buyers used AI to research their software purchase, and 94% of those who used AI said they fact-check its responses at least some of the time (opens in a new tab); 83% shortlisted three or fewer products. 1,862 technology buyers, January 2026 (vendor survey; TrustRadius runs a review platform).
• After an AI mentions a vendor, 71% of U.S. B2B professionals who use AI at work say they visit the vendor's website, 63% search for the company on Google, and 38% check reviews on G2 or similar sites (opens in a new tab). In the same survey, 66% regularly use AI to research products, vendors, or solutions, 92% say AI has shaped their vendor shortlist, and agencies and service providers were the category most often researched with AI (51%). For product research, 71% use ChatGPT, 61% Gemini, 45% Microsoft Copilot, 18% Perplexity, and 14% Claude. 519 respondents, March–April 2026 (vendor survey).
• 67% of B2B buyers prefer a rep-free buying experience, and 45% used AI during a recent purchase (opens in a new tab). Survey of 646 B2B buyers, August–September 2025 (analyst survey). A later release on a survey of 645 buyers from the same months reported that 69% prefer to validate AI-generated insights with sales reps (opens in a new tab).
• 94% of business buyers report using AI during their buying process (opens in a new tab). Forrester's Buyers' Journey Survey of nearly 18,000 global business buyers, 2025 (analyst survey). Caveat: the published summary does not define "using AI".
• 94% of B2B buying groups ranked their preferred vendors before first contact, and bought from that preliminary favorite 77% of the time (opens in a new tab). More than 4,000 buyers in North America, EMEA, and APAC, published November 12, 2025 (vendor survey).

AI crawlers

A crawler is a program that fetches web pages. Each AI company runs its own, such as OpenAI's GPTBot and Anthropic's ClaudeBot.

• OpenAI's GPTBot made 569 million requests and Anthropic's Claude crawler 370 million on Vercel's network in one month, together about 20% of Googlebot's 4.5 billion (opens in a new tab). None of the major AI crawlers rendered JavaScript; ChatGPT's and Claude's crawlers fetched JavaScript files in 11.50% and 23.84% of their requests but did not execute them (opens in a new tab). Published December 17, 2024 (network measurement by a hosting platform). Caveat: one network, late 2024.
• In July 2025, Anthropic's crawlers made about 38,065 page requests for every referral, down from 286,930 in January; OpenAI's ratio was 1,091 (from 1,217), and Perplexity's rose to 195 (from 54) (opens in a new tab). Cloudflare network data, published August 29, 2025 (network measurement). Caveat: traffic referred by Claude's native app carries no Referer header, so the ratios may be overstated, by an unclear amount (opens in a new tab).
• GPTBot's share of AI and search crawler traffic grew from 4.7% in July 2024 to 11.7% in July 2025, and ClaudeBot's from 6% to nearly 10%, while Googlebot accounted for 39% (opens in a new tab). Same source.
• About 38,000 of 137,210 domains had published an llms.txt file, and 97% of those files received no requests in May 2026 (opens in a new tab). SEO audit tools made 21.7% of the requests for these files, more than any single AI bot (opens in a new tab). Server logs from Ahrefs Web Analytics, published June 15, 2026 (vendor study). See llms.txt: what it is and whether it matters.
• 12% of all new URLs clicked in Bing's search results were first discovered through IndexNow (opens in a new tab). IndexNow is a protocol that tells search engines when a page changes. September 2023 (company statement).

Research on optimization tactics

• The 2024 paper that introduced generative engine optimization (GEO) reported that GEO "can boost visibility by up to 40% in generative engine responses" (opens in a new tab) (peer-reviewed; KDD 2024). A 2026 review of 45 studies found those gains "valid within its experimental setting but conditional on a source already being present in a fixed context" (opens in a new tab) (preprint).
• Out of 54 cases, only three showed statistically significant ranking improvements from conversational SEO methods, and adding statistics lowered rankings in 19 of 24 settings (opens in a new tab). Tested with gpt-4o-mini, Claude 3.5 Haiku, o3, and o4-mini (peer-reviewed; NeurIPS 2025 Datasets and Benchmarks).
• Across 252,000 controlled trials in six models, topical relevance and list position were the biggest drivers of being cited first; explicit prices and recent timestamps also helped consistently, while formatting-only edits had little impact (opens in a new tab) (peer-reviewed; SIGIR 2026). Caveat: a laboratory test in which the authors work for a software company, Sprinklr.
• After 1,885 pages added JSON-LD schema markup, their AI Overview citations fell 4.6% relative to about 4,000 matched control pages, while changes in AI Mode (+2.4%) and ChatGPT (+2.2%) were statistically indistinguishable from zero (opens in a new tab). August 2025 to March 2026, published May 11, 2026 (vendor study). Caveat: every page already had 100 or more AI Overview citations before the change. See whether schema markup helps AEO.
• On one website, total ChatGPT referrals grew 5.7 times while untreated pages on the same site grew 3.5 times; the answer-first changes were associated with a 1.82-fold rise (95% confidence interval 1.31 to 2.54), but a placebo test gave p = 0.16, so the effect is "suggestive, not conclusive" (opens in a new tab) (preprint). Caveat: both authors work for the company whose site was studied.
• An estimated 8.90% of 10,095 pages in Google Search and Gemini results for 1,000 real-user queries were optimized for generative engines, rising to 16.36% of pages modified in 2026 (opens in a new tab) (preprint). Caveat: an estimate from an automated detector.
• A critical survey of 45 studies found no reviewed technique with a stable, longitudinal, cross-platform causal effect on organic discoverability (opens in a new tab). Studies published November 2023 to July 2026 (preprint).

Which of these findings hold up across studies is covered in what replicates in AEO and GEO research.

Where the statistics disagree

Several questions have more than one published answer. The table shows the main disagreements and why they arise.

What these statistics do not show

• How long a new company takes to appear in AI answers. No study we reviewed measures it, and no reviewed technique shows a stable, longitudinal, cross-platform causal effect (opens in a new tab).
• Cause. The figures on brand mentions and referring domains are correlations.
• What buyers see. Many studies use APIs, which share few cited domains with the consumer apps (opens in a new tab).
• Stable values. AI answers and products change often. A figure from 2025 may not describe 2026.
• All markets. Most figures describe U.S. users or English-language queries.
• Independence. Most figures come from vendors, agencies, or preprints. Few are peer-reviewed, and several authors sell tools or services in this field, as noted in each entry.

These statistics describe averages across many sites and queries. They do not predict the result for any one company. To measure a single company's AI visibility, AEO HQ recommends repeating a fixed set of buyer questions on each engine and reporting rates with intervals, as described in our methodology.

Change log

• September 28, 2026: First published.

Next steps

To see how AI assistants answer your buyers' questions, AEO HQ's Instant AEO Audit ($499) is an automated check of your site and up to three competitors. Its AI results come from one model answering five questions once each, so they are a first look rather than a measurement; the methodology lists its limits.