How to read the evidence in AI search visibility: sources we trust, sources we don't
A citation share without its count can mislead: listicles fell from 15.47% to 8.19% of retrieved pages on one vendor's transactional prompts while daily retrievals fell only 9.8%. A three-tier source system separates audited evidence from recycled vendor figures and labels single-case studies as provisional.
AI visibility statistics are only as reliable as the method behind them, and a public three-tier system is how this library sorts them. Tier A is citable without qualification, Tier B with attribution and a caveat, and Tier C only as provisional. Accurate numbers can still mislead when a share travels without its count. On one vendor's transactional prompts, listicles fell from 15.47% to 8.19% of retrieved pages while daily retrievals fell only 9.8% Konitzny · Peec AI · 2026. The sections that follow set out the tiers, the figures this library has retired and a four-step check.
Why AI visibility statistics vary so much: circular sourcing and unstable answers
Circular sourcing is the first cause: vendor blogs cite each other until a figure looks like consensus without ever gaining a primary source. A number starts in one case study and travels into agency decks and conference talks. By the time a brand team reads it, the sample size has gone, as happened to the "23×" conversion figure retired below.
AI answers are stochastic, so one run is a sample rather than a constant. In LLM evaluation, run-to-run variance is large relative to the gaps being measured, and the authors recommend 5 to 10 repeated runs Mustahsan · arXiv · 2025. In AI search, reliable citation-share estimates needed roughly 40 to 150 queries per engine, with typical 95% confidence intervals of 3 to 6 percentage points Sielinski · IQRush · 2026. Two studies with different methods can report different numbers on the same outcome and both be accurate.
The three source tiers, and the Tier A sources used most
The tier depends on what a source discloses and how it was produced. Tier A covers peer-reviewed work, enterprise-scale analytics with public methodology, major consultancies, large disclosed first-party studies and government sources. The Princeton GEO paper is Tier A, peer-reviewed at KDD 2024 with its method disclosed Aggarwal et al. · Princeton University / Georgia Tech / Allen Institute for AI / IIT Delhi · 2024.
Tier B is citable with attribution and a one-line caveat. It covers credible trade press, named-company case studies and single-vendor studies with a stated design. The schema study is Tier B: one vendor's matched difference-in-differences test against 4,000 control pages Ahrefs · 2026.
Tier C covers vendor blogs, individual LinkedIn or Substack posts and single-data-point case studies. A Tier C source is used only when a finding is new and nothing stronger exists, and the body labels it provisional. Every source list on this site shows each entry's tier, so a reader can audit an article on the page.
Enterprise-scale telemetry is the largest Tier A group. Pew Research Center observed 68,879 Google searches by 900 US adults rather than asking them Pew Research Center · 2025. Cloudflare compared crawler traffic across a fixed customer cohort to remove growth bias Cloudflare · 2025.
Peer-reviewed and primary sources make up the rest. The GEO-16 analysis audited 1,702 citations with domain-clustered models arXiv · 2025. The Tow Center ran 1,600 queries across eight AI search engines and graded every answer by hand Tow Center for Digital Journalism, Columbia · 2025.
The AI visibility statistics this library no longer cites
The "23× AI conversion rate" is retired. It generalises from a single SaaS case, and the multiple shrinks once both underlying rates are disclosed. A multi-publisher Microsoft Clarity analysis puts AI-referred conversion at roughly 3 times other channels, with its method stated Microsoft Clarity · 2026.
Claims that FAQ schema lifts citations 2.5 to 3.2 times trace to vendor blogs, and so does a "44% lift" with unclear method. Across 1,885 pages, schema produced +2.4% on Google AI Mode and +2.2% on ChatGPT, both indistinguishable from zero Ahrefs · 2026. The matched control design is what makes it the replacement.
Adjacent markup claims fare no better. Across 300,000 domains, the presence of an llms.txt file showed no relationship with how often AI engines cited a domain SE Ranking · 2025. One vendor case still in use reports a 19.72% rise in AI Overview visibility after entity linking van Berkel · Schema App · 2025. It is a single-data-point case study and stays provisional.
How to check an AI visibility statistic in four steps
A reader can run the check without this library, and any one step can fail a statistic. First, find the primary source. If a vendor blog cites another vendor blog, stop, because a claim with no primary source is a rumour with a number attached. Second, check the sample size and method. One customer does not generalise, while the schema study matched 1,885 treated pages to controls Ahrefs · 2026.
Third, check the date and the run count. AI search moves fast, and single-run measurements are unreliable Mustahsan · arXiv · 2025. Fourth, recast the claim into its parts. A multiple should come with both rates, a share with its count, and any estimate with its confidence interval Sielinski · IQRush · 2026.
Citation share versus citation count: a falling share can hide a stable count
A citation share is a fraction, so it falls whenever the total grows faster than the count. Peec AI's platform data shows the effect after ChatGPT changed its default model in August 2026. Retrieved sources per prompt doubled from about 12 to 24, and listicles fell from 15.47% to 8.19% of retrieved pages Konitzny · Peec AI · 2026. Daily listicle retrievals fell only 9.8%.
A headline saying listicles halved describes the 47% fall in share. The count says they lost 9.8% of their retrievals, about a tenth. Peec discloses no sample size, prompt count or time window, the page-type split covers transactional prompts only, and Peec sells the platform. The magnitudes are one vendor's; the arithmetic holds regardless.
A falling share is sometimes a real fall. Reddit's share of ChatGPT Search citations dropped from 3.83% to 0.52% in mid-August 2026 in Promptwatch's tracking Foppen · Promptwatch B.V. · 2026. Pierview's absolute daily counts fell over the same days, from about 248 to 16 Shaikh · Pierview · 2026. When share and count fall together, the drop is real. Both are vendor panels, and Promptwatch calls its figure provisional.
The unit of count matters as much as the denominator. Across 27,000 ChatGPT responses to 400 prompts over 14 weeks, Resoneo measured a stable 1.26 cited URLs per domain de Segonzac · Resoneo · 2026. A count of cited URLs therefore ran about 26% above the count of distinct sources. Resoneo is an agency with a ChatGPT scraping plugin, does not disclose prompt selection, and covers earlier ChatGPT models only.
What AI visibility research still cannot show
Five questions still lack evidence strong enough for a confident answer. The first is real prompt data: what users type into AI assistants is proprietary to the AI companies, so the category works from public proxies. The second is the long-term effect of model updates on visibility, where no longitudinal Tier A study exists yet.
The third is how agentic browsers reshape attribution. OpenAI documents Atlas as a browser that can complete purchases for the user OpenAI · 2025. Perplexity documents Comet the same way, as a browser that runs agentic tasks across the web Perplexity AI · 2025. Neither launch reports how people actually use the agent.
The fourth is how engines weight first-party against third-party content over time, which current evidence shows only in snapshots. The fifth is regulation. The EU Code of Practice requires signatories to honour machine-readable opt-outs such as robots.txt European Commission / EU AI Office · 2025. The US Copyright Office treats fair use in AI training as fact-specific U.S. Copyright Office · 2025. Neither yet shows an operational effect on retrieval.
FAQFrequently Asked Questions
Sources
Sources are tiered per our methodology & sources page.
Quantifying Uncertainty in AI Visibility: A Statistical Framework for Generative Search Measurement (arXiv 2603.08924)
IQRush · Ronald Sielinski · 2026
Citation visibility behaves as a sample estimate, not a fixed value. Reliable citation-share estimates required roughly 40 to 50 queries on Gemini, about 100 on Perplexity, and 150 or more on SearchGPT. Typical 95% confidence intervals spanned 3 to 6 percentage points, and apparent week-over-week movements often fell inside sampling noise.
Methodology note
Preprint by Ronald Sielinski (IQRush), submitted 2026-06-09. Bootstrap resampling built 95% confidence intervals around citation share and prevalence across Perplexity Search, OpenAI SearchGPT and Google Gemini, using 200 queries per topic across three consumer topics, sampled daily over nine days plus ten-minute-interval runs. Fetched and read directly.
Stochasticity in Agentic Evaluations: Quantifying Inconsistency with Intraclass Correlation
arXiv · Mustahsan · 2025
Quantifies stochasticity in agentic LLM evaluations using intraclass correlation coefficients (ICC). Shows that single-run evaluations of agentic systems are unreliable because run-to-run variance is large relative to the gap between system variants. Recommends a minimum of 5 to 10 repeated runs per evaluation and reports the ICCs for several common agentic benchmarks.
Methodology note
arXiv preprint 2512.06710 (December 2025). Direct fetch on arxiv.org returned the abstract page. The paper applies the intraclass correlation coefficient framework from psychometrics to LLM agent evaluation and reports ICC values across multiple published benchmarks.
OpenAI launched ChatGPT Atlas, an AI-native web browser that embeds ChatGPT directly into browsing, summarises pages, answers questions in the sidebar, and can carry out multi-step tasks on the user's behalf such as filling forms, comparing products, and completing purchases. (agent inferred)
Methodology note
First-party product launch announcement from OpenAI. The post introduces Atlas as a Chromium-based browser with ChatGPT integrated as the default interface, agentic capabilities for browsing and transacting, and memory of past sessions. Initial availability is on macOS with other platforms to follow.
AI Answer Engine Citation Behavior: An Empirical Analysis of the GEO-16 Framework
arXiv · 2025
Three on-page properties showed the strongest association with whether a page got cited by AI answer engines: metadata and freshness, semantic HTML markup, and structured data. Pages that scored at least 0.70 on the GEO-16 quality score and met at least 12 of 16 quality pillars were cited at substantially higher rates than pages that did not.
Methodology note
70 product-intent prompts were run across Brave Summary, Google AI Overviews, and Perplexity, producing 1,702 citations across 1,100 unique URLs. The researchers audited each cited page against a 16-pillar framework and used logistic models with domain-clustered standard errors. The study focuses on English-language B2B SaaS pages. Published September 2025.
Google Users Are Less Likely to Click on Links When an AI Summary Appears in Search Results
Pew Research Center · 2025
When a Google search result page includes an AI summary, users click on a traditional link in roughly 8% of visits. On result pages without an AI summary, they click in roughly 15% of visits. Users rarely click on the citations inside the AI summary itself, doing so on about 1% of visits.
Methodology note
Pew Research panel study covering 900 US adults and 68,879 Google searches conducted between March and May 2025. Sessions were tracked through opt-in browser participation; click behaviour was observed directly rather than self-reported. Published July 2025.
Code of Practice for General-Purpose AI Models (Copyright Chapter)
European Commission / EU AI Office · 2025
The EU General-Purpose AI Code of Practice provides a voluntary route for AI model providers to demonstrate compliance with the AI Act's obligations on copyright, transparency, and safety. The Copyright Chapter requires signatories to honour machine-readable opt-out signals such as robots.txt and TDM reservations, to publish a summary of training data, and to put a complaint mechanism in place for rightsholders.
Methodology note
Official European Commission policy page hosting the Code of Practice for General-Purpose AI Models, developed by independent experts under the EU AI Act process and published in 2025. The Code covers Safety and Security, Transparency, and Copyright chapters, and is signed by major AI providers as a way to show compliance with the AI Act.
Introducing Comet: Browse at the Speed of Thought
Perplexity AI · 2025
Perplexity launched Comet, an AI-native web browser built around Perplexity's answer engine. Comet replaces the search bar with a conversational assistant, summarises pages, answers questions about open tabs, and can execute agentic tasks across the web on behalf of the user. (agent inferred)
Methodology note
First-party product launch announcement from Perplexity. The post positions Comet as a Chromium-based browser with Perplexity's assistant available across every tab, supporting research workflows, product comparisons, and multi-step actions. Initial access was offered to Perplexity Max subscribers.
From Googlebot to GPTBot: Who's Crawling Your Site in 2025
Cloudflare · 2025
Across Cloudflare's network, search and AI crawler traffic rose 18% from May 2024 to May 2025. Googlebot grew 96% in raw requests and now accounts for 50% of crawler traffic. GPTBot rose 305% in requests, with its share climbing from 2.2% to 7.7%. ChatGPT-User requests jumped 2,825%, and PerplexityBot grew 157,490% off a tiny base. About 14% of top domains now use robots.txt rules targeting AI bots specifically.
Methodology note
Cloudflare Radar analysis published July 2025, comparing crawler activity in May 2024 against May 2025 across a fixed cohort of customers to remove growth bias. The team matches user-agent tokens against an open-source list of AI crawlers and analyses robots.txt files on 3,816 of the top 10,000 domains. Methodology and limits are documented in the post.
Copyright and AI Part 3: Generative AI Training (pre-publication report)
U.S. Copyright Office · 2025
The US Copyright Office concludes that generative AI training raises copyright questions at several points: data collection, model training, retrieval-augmented generation, and outputs. Fair use is fact-specific and depends on transformativeness, commerciality, the amount used, and effects on the market for the original work, including market dilution and lost licensing. The Office recommends voluntary licensing markets rather than compulsory licensing schemes.
Methodology note
Pre-publication version of Part 3 of the Copyright Office's Report on Copyright and Artificial Intelligence, released May 2025 by the Register of Copyrights. The report draws on more than 10,000 comments submitted in response to a 2023 Notice of Inquiry, plus existing case law and international approaches. Sections cover technical background, prima facie infringement, fair use, and licensing options.
AI Search Has a Citation Problem (Tow Center Report)
Tow Center for Digital Journalism, Columbia · 2025
Across eight AI search engines tested, more than 60% of news-attribution queries received incorrect answers. Perplexity got 37% wrong; Grok 3 got 94% wrong. Premium paid models were no more accurate than free ones, and often produced confidently incorrect answers without flagging uncertainty. Several engines retrieved content from publishers that had explicitly blocked their crawlers.
Methodology note
1,600 queries were run across ChatGPT Search, Perplexity, Perplexity Pro, DeepSeek Search, Microsoft Copilot, Grok-2, Grok-3, and Google Gemini. The researchers selected 10 articles from each of 20 publishers, used direct excerpts as queries, and asked each chatbot to identify the headline, publisher, publication date, and URL. Responses were manually graded against six categories.
GEO: Generative Engine Optimization
Princeton University / Georgia Tech / Allen Institute for AI / IIT Delhi · Pranjal Aggarwal et al. · 2024
Adding citations, quotations, and statistics to content can increase its visibility in AI-generated answers by up to 41% on average. Pages ranked outside the top of traditional search saw the largest gains. The effect varies by content domain and by AI engine, but the lift from evidence-style content elements is consistent across the conditions tested.
Methodology note
10,000 questions were run through generative search engines. The researchers compared answers before and after applying nine content optimisation strategies, including citations, quotations, statistics, and authoritative language. They measured visibility as the share of the AI answer attributable to the optimised page, using both word position and word count metrics. Peer-reviewed at KDD 2024.
Reddit Citations Fell Off a Cliff in ChatGPT, While Perplexity Doubled Down (Pierview)
Pierview · Naqui Shaikh · 2026
Across 44,488 tracked Reddit citations from 1 July to 19 August 2026, ChatGPT's daily Reddit citations fell from about 248 in early August to 16 over 14-19 August, a 94% collapse in absolute counts, not share. In the same window Perplexity's rose 93% (512 to 988 per day) and supplied 71% of all tracked Reddit citations, while Gemini stayed roughly flat.
Methodology note
Vendor study by Pierview, an AI-visibility platform, authored by Naqui Shaikh and last updated 24 August 2026, fetched directly on 28 August 2026. Continuous real-browser session tracking of 50,000-plus prompts across six AI surfaces; Reddit citations counted daily per engine. Prompt-set composition, category mix and confidence intervals are not disclosed.
Why Did ChatGPT Stop Citing Reddit? Inside the August 2026 Citation Collapse (Promptwatch)
Promptwatch B.V. · Klaas Foppen · 2026
Reddit's share of ChatGPT Search citations averaged 3.83% from 18 July to 7 August 2026, then fell to 0.52% across 14-17 August, an 86.4% relative drop. Over the same window Google AI Overviews fell only 11.3% (2.37% to 2.10%) and AI Mode 30.5% (2.22% to 1.54%), both gradually. Promptwatch calls the drop provisional and cannot rule out its own collection error.
Methodology note
First-party vendor telemetry from Promptwatch, an AI-visibility tracker, authored by co-founder Klaas Foppen and published 20 August 2026, fetched directly on 28 August 2026. Reddit's share is computed over ChatGPT Search responses containing at least one citation. No prompt-panel size, panel composition or confidence intervals are disclosed.
ChatGPT's New Default Model GPT-5.6: More Retrieval, More Fan-Outs, More First-Party Content
Peec AI · David Konitzny · 2026
After GPT-5.6 became ChatGPT's default, single-fan-out prompts fell from 94.0% to 43.5% and average retrieved sources per prompt doubled from about 12 to 24. On transactional prompts, listicles fell from 15.47% to 8.19% of retrieved pages while daily retrievals fell only 9.8%; product pages became the top type at 16.39%. The site: operator rose from 0.3% to 23% of fan-outs.
Methodology note
Platform analysis by David Konitzny, GEO Researcher at Peec AI, published 10 August 2026 on LinkedIn Pulse and read from PDF on 2026-08-24. Aggregate telemetry from Peec's tracked prompts with no sample sizes, prompt counts, confidence intervals or time window disclosed. Page types use the author's five-category scheme on transactional prompts only, and Peec sells the platform the data comes from.
We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved
Ahrefs · 2026
Across 1,885 pages that added JSON-LD between August 2025 and March 2026, schema produced no meaningful uplift in AI citations. Matched difference-in-differences tests against 4,000 control pages showed +2.4% on Google AI Mode and +2.2% on ChatGPT (both statistically indistinguishable from zero) and a small 4.6% decline on Google AI Overviews. 53% of AI-cited pages already carry schema, but this reflects overall site quality.
Methodology note
Ahrefs identified 1,885 URLs that transitioned from no JSON-LD to having JSON-LD between August 2025 and March 2026, using its crawler database. Each treated page was matched to three control pages from different domains with similar pre-period citation levels. Citation changes were measured 30 days before and after the schema-add date across AI Overviews, AI Mode and ChatGPT using four statistical tests including matched difference-in-differences.
The Bigfoot Effect: How ChatGPT's Search Visibility Collapsed Overnight (Resoneo)
Resoneo · Olivier de Segonzac · 2026
Across 27,000 responses to 400 prompts over 14 weeks, ChatGPT's switch to GPT-5.3 Instant on 4 March 2026 cut cited URLs per response by about 20%, and unique domains per response fell from about 19.6 to 15.5. The URL-to-domain ratio held at 1.26, so results are effectively deduplicated by domain. The free tier, used by over 90% of weekly users, searches and cites less than paid tiers.
Methodology note
Longitudinal tracking study by Resoneo, published March 2026 and read from PDF on 2026-08-24: 27,000 responses across 400 prompts over 14 weeks spanning the 4 March default-model switch. Prompt selection, vertical mix and confidence intervals are undisclosed. Resoneo is an AI-search agency that distributes a ChatGPT scraping plugin, and the scope is GPT-5.3 Instant and earlier.
AI Traffic Converts at 3× the Rate of Other Channels (Study)
Microsoft Clarity · 2026
Visitors arriving from AI assistants convert at roughly 3 times the rate of visitors from other channels, and at up to 11 times the rate in certain publisher segments. AI traffic still represents a small share of total visits, but its per-visitor commercial value is materially higher than traditional search or social.
Methodology note
Analysis of Microsoft Clarity user-session data across a multi-publisher dataset. The study compared conversion rates of sessions originating from AI assistants against sessions from other referral channels. Published January 2026. Single-vendor study with disclosed methodology, downgraded to Tier B in v1.1.
LLMs.txt Shows No Clear Effect on AI Citations (300K domains)
SE Ranking · 2025
Across 300,000 domains, only 10.13% had an llms.txt file. Adoption is roughly flat across traffic tiers, with high-traffic sites slightly less likely (8.27%) to use it than mid-tier ones (10.54%). Statistical tests and an XGBoost model found no relationship between the presence of llms.txt and how often a domain is cited by AI engines. Removing the variable from the model actually improved its accuracy.
Methodology note
SE Ranking study of nearly 300,000 domains, published November 2025. The team checked each domain for an llms.txt file, segmented adoption by monthly traffic, and modelled citation frequency using Spearman correlation, XGBoost regression and SHAP analysis. The conclusion is based on whether llms.txt presence improved or degraded model predictions of LLM citations.
What 2025 Revealed About AI Search and the Future of Schema Markup
Schema App · Martha van Berkel · 2025
In 2025, Google and Microsoft publicly confirmed they use Schema markup for generative AI features, and ChatGPT confirmed it uses structured data to decide which products appear in results. Schema App reported a 19.72% rise in AI Overview visibility on its own site after deploying Entity Linking, and customer InSinkErator a 69% rise in clicks on non-branded queries.
Methodology note
First-party essay by Schema App's CEO. The piece argues structured data should be treated as a knowledge graph rather than a rich-result trick, and uses examples from Schema App's own site and named customers (InSinkErator, Wells Fargo) plus public statements from Google, Microsoft, and ChatGPT to support the case.
About the author Max Ackermann
Max Ackermann is founder and Managing Director of info.link, the product data platform that makes brands visible in AI search and connects every physical product to the web through GS1 Digital Link. He writes about AI search and generative engine optimization (GEO), AI-powered commerce, and how brands can structure product data for ChatGPT, Gemini, Perplexity, and retailer AI assistants like Amazon Rufus. For the past two years he has built the pipelines that put structured product data into AI answers, and run the experiments that test what actually moves AI citations.
Max has 20+ years of experience building digital products and businesses. He previously led McKinsey's Corporate Venture and Design teams across Europe, and as Managing Director of a leading US digital agency he built platforms with Nike, Google, Meta, and Airbnb. He founded the UX Design program at Central Saint Martins College, University of the Arts London, and is a Fellow of the UK's Higher Education Academy. Based in Hamburg, he works closely with GS1 on Digital Link adoption; info.link is headquartered in Hamburg and Berlin and counts GS1 Germany among its investors.
Follow Max on LinkedIn.
