Malte Landwehr

521 AI Search Statistics for 2026

How many people use AI search, what it does to clicks, which sources answer engines cite, and what actually moves visibility. Every figure names its source and the month it was measured.

Last updated .

Key takeaways

Why these numbers matter

AI search stopped being an experiment some time in 2025. Close to 4 billion people now use a language model every month, monthly sessions across AI tools are 56% the size of search worldwide, and in the six largest European markets visits to the leading LLM platforms went from near zero at the end of 2023 to 1.4 billion in May 2026 while search volume stayed flat. For a marketing team the question is no longer whether answer engines matter. It is how visibility inside them works, and which part of it you can actually measure.

This page collects 521 statistics on AI search in 33 sections. They come from primary sources: Pew Research Center, Adobe, Ahrefs, Similarweb, Semrush, Graphite, BCG, Cloudflare, Google, OpenAI, Peec AI, academic papers, court filings and company earnings calls. Every figure links to its source and states the month it was measured, so you can check it rather than take my word for it. Numbers without a date are worthless in this field: most of what follows has a shelf life measured in months.

Read the numbers with the usual caution. The studies below use different samples, different engines, different markets and different definitions of a search, so two figures on the same topic can disagree by a wide margin and both still be right about what they measured. Where that happens I have kept both rather than picking a favourite.

If a figure here is out of date or wrong, tell me. I would rather fix it than leave it on the page.

How big AI search has become

  1. About half of U.S. adults (49%) say they ever use AI chatbots in 2026, up from 33% in 2024. Pew Research Center,
  2. Almost 4 billion people use LLMs globally every month. Auto Trader,
  3. Monthly sessions of AI tools are now 56% the size of search worldwide, and 34% the size of search in the US. Graphite,
  4. Search-related usage of AI is now 28% the size of search worldwide and 17% in the US. Graphite,
  5. From December 2024 to December 2025, AI usage in the US increased by 300%. Graphite,
  6. 83% of AI usage worldwide happens in mobile apps. In the US it is 75%. Graphite,
  7. ChatGPT accounts for 20% of search-related traffic worldwide and 12% in the US. Graphite,
  8. 10% to 15% of all search is already happening in AI search. seoClarity,
  9. Monthly visits to the leading LLM platforms across six European markets (UK, France, Germany, Spain, Italy, Portugal) grew from near zero at the end of 2023 to 1.4 billion in May 2026. SimilarWeb via BCG,
  10. Monthly search engine visits across those same six European markets stayed broadly flat while LLM usage scaled, at 13.0 billion at the end of 2023 and 12.5 billion in May 2026. SimilarWeb via BCG,
  11. ChatGPT’s share of LLM visits across six European markets fell from about 90% in late 2024 to 57% in May 2026. SimilarWeb via BCG,
  12. Gemini’s share of LLM visits across six European markets rose from 6% in September 2025 to 27% in May 2026. SimilarWeb via BCG,
  13. Claude’s share of LLM visits across six European markets rose from 3% to 10% between March 2026 and May 2026. SimilarWeb via BCG,
  14. DeepSeek peaked at a 5% share of LLM visits across six European markets in March 2025. SimilarWeb via BCG,
  15. In May 2026 the share of LLM visits across six European markets was ChatGPT 57%, Gemini 27%, Claude 10%, DeepSeek 2%, and 1% each for Perplexity, Copilot and Grok. SimilarWeb via BCG,
  16. Both Grok and Character AI are now bigger than Perplexity by usage. Third-party usage data via Malte Landwehr,
  17. Claude traffic grew 3x from January to March 2026 and overtook Perplexity as number three in AI search. Malte Landwehr, MonthlyLandwehr podcast,
  18. B2B websites already receive similar click volumes from Claude and Gemini, despite Gemini being 4x bigger in overall web usage. Malte Landwehr, MonthlyLandwehr podcast,
  19. Google AI Overviews reaches 2.5 billion monthly users. Google I/O 2026,
  20. Google Gemini has 0.9 billion monthly users. Google,
  21. Google AI Mode has surpassed 1 billion monthly active users globally. Google, AI Mode U.S. Insights,
  22. ChatGPT reached 700 million weekly active users in July 2025. The Information,
  23. ChatGPT had more than 500 million weekly active users in July 2025 and 400 million in February 2025. Axios and CNBC,
  24. ChatGPT handles 2.5 billion prompts per day globally. Axios,
  25. OpenAI had 3 million paying business users in June 2025 and 5 million in August 2025. Brad Lightcap, OpenAI COO,
  26. OpenAI reported 2 million paying enterprise customers in February 2025. CNBC,
  27. From June 2024 to June 2025, usage of AI chat and search tools grew 121% in the US and 158% in the EU and UK, while traditional search engine usage in the US declined 8%. Datos and SparkToro, State of Search Q2
  28. From June 2024 to June 2025, the number of monthly Google searches by US desktop users declined by 5%. Datos and SparkToro, State of Search Q2
  29. From June 2024 to June 2025, the share of UK desktop users using ChatGPT increased from 20.5% to 30.5%. Datos and SparkToro, State of Search Q2
  30. In June 2025, 5.6% of U.S. search traffic on desktop browsers went to an AI-powered large language model such as ChatGPT or Perplexity. WSJ and Datos,
  31. Between January and September 2025 in the UK, Google had 77.3 million users versus ChatGPT’s 16.9 million, and 99% of ChatGPT users also used Google. SimilarWeb, BrightonSEO
  32. 25% of Google users have tried AI Mode, both in the US and the UK. SimilarWeb, BrightonSEO
  33. Google’s AI Mode is the fastest chatbot ever to reach 100 million users. SimilarWeb, BrightonSEO
  34. On desktop in August 2025, 46% of US and 38% of UK AI Mode users used it on more than one day. SimilarWeb, BrightonSEO
  35. When a person starts using ChatGPT, their Google usage measured by page views decreases 2% after 3 months and 8% after 23 months. Coatue,
  36. 66% of students globally now use ChatGPT. Digital Education Council Global AI Student Survey
  37. 72% of people use AI tools for search, and 14% use them for search daily. HigherVisibility,
  38. 44% of people prefer LLM-based search and answer engines. Only 31% prefer classic search engines. McKinsey,
  39. 62% of brands are invisible in AI answers. Tactixx 2026,

ChatGPT’s real search volume versus Google

  1. ChatGPT handles roughly 600 million searches per day. Peec AI,
  2. ChatGPT likely processes between 500 million and 1.7 billion searches per day. Peec AI,
  3. ChatGPT’s search volume is between 4% and 12% of Google’s search volume. Peec AI,
  4. Google is 8x to 22x bigger than ChatGPT in search volume. Peec AI,
  5. Most search-style queries on ChatGPT need fewer than 2 messages. Peec AI,
  6. Google processes approximately 14 billion searches per day. Google, via Peec AI,
  7. 24% of ChatGPT prompts can be classified as information-seeking search queries. OpenAI and Harvard, via Peec AI,
  8. 65% of ChatGPT usage qualifies as search. Ahrefs,
  9. ChatGPT has 12% of Google’s search volume. Ahrefs,
  10. ChatGPT has 1.2 to 1.4 billion search-like tasks per day, roughly 9% of Google’s daily searches. seoClarity,
  11. ChatGPT has 4% to 9% of Google’s search volume. Sistrix,
  12. ChatGPT represents only 0.6% of Google’s search volume when measured by outgoing clicks to websites. Ahrefs,
  13. Google sends 190x more traffic to websites than ChatGPT. Ahrefs,
  14. Traffic share in July 2025 was 41.91% for Google and 0.19% for ChatGPT. ChatGPT-vs-Google (Ahrefs),
  15. In Germany, 40% of ChatGPT prompts are searches or comparable to searches, 40% are to create or edit text, and 20% are emotional, roleplay or creative. Hanns Kronenberg / GPT-Insights,
  16. 38% of ChatGPT queries are generative (“write or make something”) and therefore not search. Profound,
  17. 12% of ChatGPT queries are no-intent (“thank you”, “please”) and therefore not search. Profound,
  18. 30% of prompts have search intent. SparkToro,
  19. 31% of ChatGPT prompts trigger a web search. For commerce it is 41%, for travel 22%, under 20% for credit cards and almost 60% for local. Nectiv Digital,
  20. ChatGPT conducted a web search in 100% of more than 8,000 chats about trending news topics. Peec AI (Malte Landwehr),

Zero clicks, CTR and the end of the referral

  1. ChatGPT’s zero-click rate is roughly 99%. Google AI Mode’s is about 95%. Peec AI (Malte Landwehr),
  2. 35% to 40% of Google searches lead to a click. In ChatGPT and Google AI Mode only 1% to 5% of search-like prompts result in a click. Peec AI (Malte Landwehr),
  3. The zero-click rate in search has risen from 60% to 65% to 70%. Peec AI (Malte Landwehr),
  4. Web analytics misses more than 95% of the impact of AI search. Peec AI (Malte Landwehr),
  5. Google’s CTR is 29.2%. ChatGPT’s CTR is 0.84%, which is 96% lower. Ahrefs,
  6. ChatGPT’s CTR is 3.5% for pure-search prompts, 1.6% for asking-intent prompts and 1.3% for broad-search prompts. Ahrefs,
  7. CTR in ChatGPT is below 1%. Leaked publisher report,
  8. An OpenAI software engineer wrote about ChatGPT’s source links: “No matter how prominently we show the links, users won’t click.” Publishers’ unredacted brief, NYT et al. v. OpenAI and Microsoft,
  9. Microsoft’s own data shows click-through rates from Bing Chat and Copilot are 87% to 93% lower than from Bing search for New York Times properties, 83% to 91% lower for Daily News domains and 51% to 94% lower for Ziff Davis properties. Microsoft internal data in NYT et al. v. OpenAI and Microsoft,
  10. Google users click a link inside an AI Overview in only 1% of visits where an AI Overview appears. Pew Research Center,
  11. On 20 September 2026, 26.2% of Google AI Overviews carried an external link inside the answer text. Before 11 September 2026 it was close to 0%. Peec AI (Malte Landwehr),
  12. The median CTR per task in Google’s AI Mode is 0%, and 78% of AI Mode sessions have a CTR of 0%. Growth Memo,
  13. Google AI Mode has a click-through rate of 3.8% to 5.4%. iPullRank,
  14. Google AI Mode has a click-through rate of 7% per session. Semrush,
  15. 24% of Google Search sessions result in a click, versus 5% of Google AI Mode sessions. iPullRank and Similarweb,
  16. Google’s AI Mode is 92% to 94% zero-click. Semrush,
  17. Google’s AI Mode has 90% fewer clicks than organic search. Traditional search on Google has a 20% CTR, AI Mode has 3%. SimilarWeb, BrightonSEO
  18. In June 2025, 57% of Google searches on desktop in the US did not result in a click to an external website. In the EU and UK it was 53%. Datos and SparkToro, State of Search Q2
  19. Zero-click searches rose from 56% to nearly 69% between May 2024 and May 2025, after the launch of AI Overviews. SimilarWeb,
  20. Roughly 40% of Google searches result in a click to an external website. SparkToro,
  21. The measured CTR impact of Google AI Overviews across more than 15 studies ranges from +19% to -85%. Malte Landwehr meta-analysis,
  22. The average CTR impact of Google AI Overviews is -40%. The median is -37%. Malte Landwehr meta-analysis,
  23. Google AI Overviews reduce clicks to number-one-ranked content by 58%. Position 10 loses 19.4% of clicks. Ahrefs AI Search Benchmark Report Q1
  24. Organic CTR on Google dropped 24% from 2024 to 2025, and on some positions by up to 62%. seoClarity,
  25. Measured through logfile hits, CTR in AI search runs below 0.1% for top-of-funnel pages, 1% to 1.5% for purchase-intent pages, 9% for downloads and below 0.5% for the homepage. PromptWatch, BrightonSEO
  26. From June 2024 to June 2025, the share of US desktop Google searches resulting in an organic click declined 6%, while the share resulting in a paid click increased 10%. Datos and SparkToro, State of Search Q2
  27. From June 2024 to June 2025, the share of EU and UK desktop Google searches resulting in an organic click declined 4%, while paid clicks increased 50%. Datos and SparkToro, State of Search Q2

Google AI Overviews: coverage and behaviour

  1. AI Overviews appeared in 56.9% of Google searches in April 2025 and in 86.7% in April 2026. Peec AI, 500,000 prompts,
  2. AI Overviews appear for 88.5% of decision-stage prompts. Peec AI,
  3. AI Overviews appear in 93% to 97% of queries with a clear intent pattern such as short fact, comparison or definition. Peec AI,
  4. AI Overviews appear in 67% of queries with ambiguous intent. Peec AI,
  5. AI Overviews appear in 76% of Google searches in the EU and 90.3% outside the EU. Peec AI,
  6. AI Overviews appear in 91.4% of Google searches in the US, 91.7% in the UK, 86.5% in Australia, 82.9% in Poland and 75.6% in Germany. Peec AI,
  7. AI Overviews appear in 0% of Google searches in France, because neither AI Overviews nor AI Mode is served there. Peec AI,
  8. One-word queries trigger an AI Overview almost 80% of the time, two-word queries 64.6%, five-word queries 78.5%, 6 to 7 word queries 84.2%, 8 to 10 word queries 87.7%, 11 to 15 word queries 89.1%, and queries above 15 words between 87% and 91%. Peec AI,
  9. Across all query word counts, AI Overviews appear in 87% of Google searches. Peec AI,
  10. 29.1% of 321 tracked news searches in the US triggered a Google AI Overview. Peec AI (Malte Landwehr),
  11. Google’s AI Overviews appear for 21% of all keywords. Ahrefs AI Search Benchmark Report Q1
  12. Google’s AI Overviews appear on 9.5% of single-word queries and 46.4% of queries with seven or more words. Ahrefs AI Search Benchmark Report Q1
  13. 99.9% of Google AI Overviews appear on informational queries. Ahrefs AI Search Benchmark Report Q1
  14. Shopping queries trigger a Google AI Overview only 3.2% of the time. Ahrefs AI Search Benchmark Report Q1
  15. Indonesia leads the world in Google AI Overview coverage at 37.2% of queries. The US rate is 20.5%. Ahrefs AI Search Benchmark Report Q1
  16. AI Overview coverage per country: Indonesia 37%, Philippines 29%, Mexico 29%, India 27%, Nigeria 25%, Brazil 25%, Japan 25%, USA 21%, Canada 19%, UK 19%, Germany 19%. Aleyda Solis,
  17. 18% of searches trigger an AI Overview. Conductor,
  18. 20% of keywords in the US and UK show AI Overviews. seoClarity,
  19. In Germany, across 1.2 million analysed keywords, 11% of keywords have an AI Overview. Metrics Tools,
  20. In Germany, AI Overviews have a median of 9 sources, mostly between 6 and 11, and rarely more than 30. Metrics Tools,
  21. 37.9% of URLs cited in Google AI Overviews rank in the SERP top 10, down from about 76% in July 2025. Ahrefs AI Search Benchmark Report Q1
  22. 60% of Google AI Overview citations are not in Google’s top 20. Semrush,
  23. 59% of Google AI Overviews contain no brands or entities. For AI Mode it is 34.66%. Ahrefs AI Search Benchmark Report Q1
  24. Google AI Overviews have a 70% chance of changing between observations, yet stay semantically consistent at 0.95 cosine similarity. Ahrefs AI Search Benchmark Report Q1
  25. The average length of content cited in Google AI Overviews is 1,282 words, with near-zero correlation between word count and being cited. Ahrefs AI Search Benchmark Report Q1
  26. 53.4% of pages cited in Google AI Overviews are under 1,000 words. Only 16% are over 2,000 words. Ahrefs AI Search Benchmark Report Q1
  27. The share of Google’s “People Also Asked” answers that are AI-generated rose from 18% in July 2025 to 27% in August and 38% in September 2025. AlsoAsked,
  28. 80% of queries on Google in the US show the “AI Mode” tab. Brodie Clark and Nozzle,
  29. One site with AI Overviews gained 46% more Search Console impressions and 80% more visibility in SEO tools, while its clicks dropped 33%. Campixx 2026,

Google AI Mode

  1. Google AI Mode queries have more than doubled every quarter since the U.S. launch in May 2025. Google, AI Mode U.S. Insights,
  2. The average Google AI Mode search query is triple the length of a traditional search query. Google, AI Mode U.S. Insights,
  3. Follow-up queries in Google AI Mode have increased by more than 40% on average per month in the U.S. Google, AI Mode U.S. Insights,
  4. More than 1 in 6 Google AI Mode searches in the U.S. are multimodal, meaning non-text. Google, AI Mode U.S. Insights,
  5. Google AI Mode searches with image input are growing by more than 40% month over month since launch. Google, AI Mode U.S. Insights,
  6. Brainstorming-related queries in Google AI Mode have grown 30% faster than AI Mode queries overall since launch. Google, AI Mode U.S. Insights,
  7. Queries starting with “which” in Google AI Mode have grown 40% faster than AI Mode queries overall over the past six months. Google, AI Mode U.S. Insights,
  8. Planning-related queries in Google AI Mode have grown 80% faster than AI Mode queries overall over the past six months. Google, AI Mode U.S. Insights,
  9. Image creation queries on Google AI Mode have more than tripled since the start of 2026. Google, AI Mode U.S. Insights,
  10. Google’s AI Mode cites 12.6 sources on average, versus roughly 3 for AI Overviews. Growth Memo,
  11. There is only a 10.7% URL overlap between Google’s AI Mode and AI Overviews. Growth Memo,
  12. Google’s AI Mode and AI Overviews reach the same conclusions 86% of the time but have only 13.7% URL overlap in their citations. Ahrefs AI Search Benchmark Report Q1
  13. A page cited in a Google AI Overview has a 61% chance of also appearing in AI Mode. Ahrefs AI Search Benchmark Report Q1
  14. Only 15.6% of Google AI Mode citations overlap with the Google Top 10, down from 40% a year earlier. Internetwarriors GEO Study
  15. Domain authority still matters at domain level, not URL level: 51% of Google AI Mode citations come from Top 10 domains, just from different pages. Internetwarriors GEO Study
  16. 68% of queries have at least one Google AI Mode citation from the top 20 organic results, but only 20% of total AI Mode citations come from the top 20. seoClarity,
  17. Share of organic rank positions appearing in Google AI Mode: rank 1 27%, rank 2 21%, rank 3 17%, rank 4 11%, rank 5 14%, rank 6 14%, rank 7 10%, rank 8 12%, rank 9 9%, rank 10 9%. seoClarity,
  18. The overlap of organic ranking in Google and citation in Google AI Mode is 42%. For mentions it is only 9%. Semrush,
  19. Google’s AI Mode is more likely to recommend smaller or challenger brands than traditional Google search. SimilarWeb, BrightonSEO
  20. 66% of “knowledgeable” respondents find Google’s AI Mode results more helpful than Google Search. Oppenheimer,
  21. Time on site after a click: Google 8.5, AI Mode 7.2, ChatGPT 15.3. Page views after a click: Google 9.7, AI Mode 7.5, ChatGPT 13.1. SimilarWeb, BrightonSEO
  22. Pictures in your Google account influence Google AI Mode results. LLM Mastery featuring Charles Floate,

Which domains AI engines actually cite

  1. Reddit.com is the most-cited domain in AI search across ChatGPT, Google AI Mode, Gemini, Perplexity and AI Overviews. Peec AI, 30 million sources,
  2. The ten most-cited domains in AI search are Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium and TechRadar in that order. Peec AI,
  3. Reddit is the number 1 or number 2 most-cited source in every AI engine tested. Peec AI,
  4. YouTube is the most-cited domain in both Google AI Mode and Google AI Overviews. Peec AI,
  5. Reddit is the most-cited domain in Gemini and in Perplexity. Peec AI,
  6. Wikipedia is the most-cited domain in ChatGPT. Peec AI,
  7. G2, TechRadar and PC Mag are the sources most cited by LLMs in B2B software. Peec AI (Malte Landwehr),
  8. Reuters and AP News both place in the top 5 most-cited domains in every news category in ChatGPT. Peec AI (Malte Landwehr),
  9. Reuters holds 38% of ChatGPT citations for international politics news prompts and 32% for business and economy prompts. Peec AI (Malte Landwehr),
  10. AP News is the most-cited domain for crime and safety news prompts in ChatGPT with 27% of citations. Peec AI (Malte Landwehr),
  11. In the health news category, CDC, FDA, NIH and WHO have unusually high citation rates in ChatGPT. Peec AI (Malte Landwehr),
  12. YouTube appears more often than any other domain in ChatGPT’s sources sidebar for news prompts, but never receives an in-line citation across thousands of chats. Peec AI (Malte Landwehr),
  13. LinkedIn is a top 20 LLM source and more influential on LLM answers than Slashdot, Medium and SourceForge. Peec AI (Malte Landwehr), 1.2 million citations,
  14. LinkedIn Pulse is responsible for 75% of LinkedIn citations in LLMs for software purchase prompts. Peec AI (Malte Landwehr), 1.2 million citations,
  15. LinkedIn content with just 3 likes can perform well in LLMs. Peec AI (Malte Landwehr),
  16. Grok’s top cited sources are 29% X, 2% Reddit, 0.7% Forbes, 0.6% Quora and 0.5% CNET. Profound,
  17. Reddit is the number 1 cited source across the big answer engines. YouTube is number 2. Profound,
  18. 6% of all ChatGPT citations are Reddit. Profound, pre
  19. ChatGPT cites Reddit for 25% of questions, Perplexity for 22% and Gemini for 0.2%. AirOps, pre
  20. 17.4% of citations on ChatGPT are user-generated content. Profound,
  21. More ChatGPT citations go to domains with a domain authority of 20 to 40 than to domains with a domain authority of 80 to 100. AirOps,
  22. 74% of ChatGPT citations go to sites with a domain authority under 80. AirOps,
  23. SEO link metrics such as backlinks and authority score account for only about 4% of citations. Profound,
  24. 28.3% of ChatGPT’s 1,000 most-cited pages have zero organic keywords in Google. Ahrefs AI Search Benchmark Report Q1
  25. Only 32.3% of ChatGPT’s most-cited pages are content types brands can influence, such as blogs, reviews and news. Ahrefs AI Search Benchmark Report Q1
  26. Across 75,000 brands, YouTube mentions correlate more strongly with AI brand visibility (about 0.737) than any other factor. Ahrefs AI Search Benchmark Report Q1
  27. 99% of Gemini and ChatGPT prompts cite earned media. Bottled Imagination, BrightonSEO
  28. 60% of Gemini prompts cite industry-related media and 42% cite “best of” lists. Bottled Imagination, BrightonSEO
  29. 30% of Gemini prompts cite national media, versus only 6% of ChatGPT prompts. Bottled Imagination, BrightonSEO
  30. 90% of ChatGPT citations are not in Google’s top 20. For Perplexity it is 50%, for Google AI Overviews 60% and for Gemini 70%. Semrush,
  31. There is only a 10% overlap between URLs ranking in Google and URLs cited in ChatGPT. Profound,
  32. On domain level, there is only an 11% overlap of citations between ChatGPT and Perplexity. Profound, pre
  33. Pages ranking number 1 in Google are cited 3.5x more by ChatGPT than pages outside the top 20. AirOps,
  34. The overlap of organic ranking in Google and citation is 38% for Perplexity and 34% for ChatGPT. For mentions it is 9% and 15%. Semrush,
  35. For B2B brands, comparing the top 100 mentions and the top 100 sources, only 8% of brands appear in both on Google AI Mode, 7% on Perplexity and 13% on ChatGPT. Semrush,
  36. For B2C brands, comparing the top 100 mentions and the top 100 sources, only 8% of brands appear in both on Google AI Mode, 8% on Perplexity and 9% on ChatGPT. Semrush,

How fresh cited content has to be

  1. 89.7% of ChatGPT’s most-cited pages were last updated in 2025. Ahrefs AI Search Benchmark Report Q1
  2. 95% of ChatGPT citations come from content updated in the last 10 months. Semrush,
  3. 50% of content cited by LLMs is less than 13 weeks old, and 75% is less than 16 weeks old. Profound,
  4. 50% of media sources used in Gemini were from 2024 onwards. For ChatGPT it was 40%. Bottled Imagination, BrightonSEO
  5. 50% of Perplexity citations are from 2025. Seer Interactive,
  6. 88% of Google AI Overview citations for YouTube videos are recent. Profound,
  7. Refreshed content is 3x more likely to be cited by LLMs. AirOps via Brainlabs AI Report
  8. ChatGPT cites Wikipedia articles and PDF files up to 6 months after they have been deleted. Peec AI (Malte Landwehr),
  9. ChatGPT kept citing a deleted Wikipedia article for 4 months after deletion with no drop in citation frequency. Peec AI (Malte Landwehr),
  1. Reddit’s share of all ChatGPT citations jumped from 2.6% to more than 11% in the week of 8 May 2026. Peec AI (Malte Landwehr),
  2. Reddit grew 4x to 5x as a retrieved AI Overview source for B2B SaaS prompts in early 2026. YouTube grew 2x from an already high base. Peec AI (Malte Landwehr),
  3. Reddit appears in 10.5% of all Grok chats, and 9 out of 10 of those appearances are site:reddit.com searches. Peec AI,
  4. Across seven industries, the strongest predictor of a Reddit thread being cited by AI is a title that reads like the question: short, around 7 words, and phrased as a question. Textbroker and Rankscale webinar,
  5. Reddit threads posted in the relevant subreddit take 46% of AI citations. Comparable threads take 27%. Textbroker and Rankscale webinar,
  6. 82% of AI-cited Reddit threads are text-post discussions with 10 to 50 comments and about a dozen participants. For comparable threads it is 45%. Textbroker and Rankscale webinar,
  7. In AI-cited Reddit threads, the top reply is upvoted 1.2x the post’s own score, versus 0.5x in comparable threads. It is the strongest single-thread signal. Textbroker and Rankscale webinar,
  8. The median AI-cited Reddit thread has about 9 upvotes, within a range of 5 to 15 across every industry. Textbroker and Rankscale webinar,
  9. AI-cited Reddit threads are typically 13 to 27 months old, and the oldest still-cited thread was 4.8 years old. Textbroker and Rankscale webinar,
  10. GummySearch, a site recycling Reddit content, powers 0.1% of all citations in ChatGPT. Before 8 May 2026 it accounted for less than 0.005%. Peec AI (Malte Landwehr),
  11. GummySearch was shut down at $35k MRR after failing to negotiate a Reddit API deal. Fed Folio, GummySearch founder,

Machine-translated Reddit pages

  1. 5.71% of Reddit citations across ChatGPT, Google AI Overview, Google AI Mode and Gemini pointed to machine-translated Reddit pages between March and June 2026. Peec AI, 64.77 million Reddit citations across 20 countries,
  2. ChatGPT’s translated Reddit citation share collapsed from 6.14% in April 2026 to 0.66% in May 2026 and 0.30% in early June 2026. Peec AI,
  3. ChatGPT’s translated Reddit citation share in Germany fell from 18.06% in April 2026 to 1.54% in May 2026. In Sweden it fell from 20.51% to 1.62%. Peec AI,
  4. Google AI Overview served machine-translated URLs in roughly 8% to 11% of its Reddit citations between March and June 2026. Google AI Mode served them in roughly 10% to 14%. Peec AI,
  5. Just under 1% of Gemini’s Reddit citations point to machine-translated pages in most months. Peec AI,
  6. Perplexity never cited a single machine-translated Reddit URL between 1 March and 10 June 2026. Peec AI,
  7. Translated Reddit citations stay under 1% of Reddit citations in English-speaking markets but run between 24% and 42% in non-English markets. Peec AI,
  8. Between 40% and 73% of Reddit citations in Google’s AI answers in non-English markets point to machine-translated English threads. Peec AI,
  9. In Sweden, 71.6% of Google AI Overview’s, 65.4% of Google AI Mode’s and 10.1% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  10. In Norway, 73.0% of Google AI Overview’s, 59.3% of Google AI Mode’s and 12.4% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  11. In Poland, 71.7% of Google AI Overview’s, 54.9% of Google AI Mode’s and 9.0% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  12. In Spain, 68.4% of Google AI Overview’s, 61.2% of Google AI Mode’s and 7.1% of ChatGPT’s Reddit citations point to machine-translated pages. Google AI Mode’s Spanish share hit 72% in early June 2026. Peec AI,
  13. In Italy, 65.4% of Google AI Overview’s, 53.5% of Google AI Mode’s and 9.1% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  14. In Brazil, 57.2% of Google AI Overview’s, 47.8% of Google AI Mode’s and 17.6% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  15. In Mexico, 54.5% of Google AI Overview’s, 56.6% of Google AI Mode’s and 11.5% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  16. In Germany, 52.2% of Google AI Overview’s, 40.7% of Google AI Mode’s and 7.3% of ChatGPT’s Reddit citations point to machine-translated pages. Peec AI,
  17. Gemini’s translated Reddit citation share is 6.1% in Germany and 26.1% in Mexico. Peec AI,
  18. Among translated Reddit citations served for German-market prompts, German daily life topics account for 16.9%, bags and fashion 13.0%, tech hardware 12.7%, audio and smart home 11.6%, gaming 10.7%, automotive and EVs 9.6%, and business and SaaS 9.5%. Peec AI,

Which content formats get cited

  1. 52% of AI citations are made up of just 3 content formats: listicles at 22%, articles at 17% and product pages at 14%. Peec AI and Wix, 1 million+ citations,
  2. Articles dominate informational prompts with 45% of AI citations. Peec AI and Wix,
  3. For transactional prompts, category pages get 15% of AI citations and the homepage 7%. Peec AI and Wix,
  4. The professional services industry has the highest share of self-promotional listicles in AI citations at 19%. Peec AI and Wix,
  5. Health and wellness AI citations are dominated by articles at 20%. Peec AI and Wix,
  6. 41% of commercial prompts cite listicles. Peec AI,
  7. 25% of transactional prompts cite product pages. Peec AI,
  8. Blogs and listicles make up 61% of citations. Profound,
  9. 33% of LLM citations feature comparative listicle content. Profound, pre
  10. “Best X” blog lists make up 43.8% of ChatGPT source URLs for top-of-funnel queries. Ahrefs AI Search Benchmark Report Q1
  11. Comparison tables overperform in AI search: they account for 10% to 12% of Google AI Mode citations versus 8% in classic search. Internetwarriors GEO Study
  12. Natural language URLs get 11.4% more citations. Profound,
  13. 87% of pages cited by ChatGPT have exactly one h1, versus 64% of pages ranking in Google. AirOps,
  14. 78% of pages cited by ChatGPT contain at least one list, versus 27% of pages ranking in Google. AirOps,
  15. 69% of pages cited by ChatGPT use proper h1, h2 and h3 headings, versus 24% of pages ranking in Google. AirOps,
  16. 61% of pages cited by ChatGPT have rich schema, versus 25% of pages ranking in Google. AirOps,
  17. ChatGPT pages with strong title-query overlap are cited at 2.2x the rate of low-overlap pages. AirOps,

Citation rates: a winner-take-most distribution

  1. 31% of URLs have a ChatGPT citation rate of 2.0 or higher, and those URLs account for 59% of all ChatGPT citations. Peec AI, 1 million+ citations,
  2. More than 9 out of 10 URLs cited by Google AI Mode have a citation rate below 1.0. Peec AI,
  3. 3% of URLs in Google AI Mode have a citation rate between 1.0 and 1.5, and they account for 12% of Google AI Mode citations. Peec AI,
  4. 64% of URLs retrieved by Perplexity are never cited in its answers. Peec AI,
  5. 6% of URLs have a Perplexity citation rate of 2.0 or higher, and they account for just under half of all Perplexity citations. Peec AI,
  6. Around 10% to 11% of URLs for informational and commercial prompts reach a citation rate of 2.0 or higher, and they account for 37% of all citations. Peec AI,
  7. Listicles make up just under one fifth of all URLs cited by ChatGPT, yet 52% of them are cited more than twice per ChatGPT answer. Peec AI,
  8. Category pages are Google AI Mode’s top-performing content type, with just under 10% reaching the high citation rate bucket. Peec AI,
  9. 85% of pages retrieved by ChatGPT never get cited. Only 15% do. AirOps,
  10. ChatGPT cites roughly 50% of the URLs it retrieves. Ahrefs AI Search Benchmark Report Q1
  11. When Ahrefs is mentioned in AI responses, a link is included only 28% of the time. Ahrefs AI Search Benchmark Report Q1

The listicle economy

  1. A rank 1 placement in a frequently AI-cited third-party listicle is associated with a 16.5 percentage point visibility lift in AI answers for B2B SaaS brands. Peec AI, nearly 200,000 AI responses across 8 engines,
  2. A rank 1 listicle placement is associated with a 13.4 percentage point visibility lift for emerging MarTech brands. Peec AI,
  3. A rank 1 listicle placement is associated with US Finance brands appearing 1.80 positions earlier in AI answers, B2B SaaS brands 1.17 positions earlier and emerging MarTech brands 0.82 positions earlier. Peec AI,
  4. A rank 1 listicle placement is associated with 0.67 additional brand mentions per AI answer for emerging MarTech brands and 0.43 for B2B SaaS brands. Peec AI,
  5. Self-promotional listicles account for about 11% of citations in AI search results for software review prompts. Peec AI, 232,000 citations,
  6. ChatGPT’s average self-promotional listicle citation rate in software review prompts is 3.6%, versus 10.3% for Google AI Mode and 10.4% for Perplexity. Peec AI,
  7. Self-promotional listicle citation rates in AI search reached 11% to 14% in December 2025 and January 2026. Peec AI,
  8. ChatGPT’s self-promotional listicle citation rate stayed between 3% and 5% from December 2025 to February 2026, and Perplexity’s between 10% and 12%. Peec AI,
  9. ChatGPT drew 7.8% of its software review citations from educational and training sites, versus 3.6% for Perplexity and 2.5% for Google AI Mode. Peec AI,
  10. Zapier accounts for about 1% of URLs cited by ChatGPT in software review prompts, but about 8% of URLs cited by other AI platforms. Peec AI,
  11. The listicle share of AI Overview sources for B2B SaaS prompts fell from 44% in October 2025 to 39% in April 2026, a constant decrease of 0.5% to 1% per month. Peec AI (Malte Landwehr),
  12. Brands ranking themselves number 1 in their own listicle get cited but not recommended by AI about 69% of the time. Lily Ray,
  13. A made-up matcha powder brand with no website and no social media presence started appearing in LLM answers after being placed at position 2 in 3 external listicles. Claneo,

Query fan-outs: what engines really search for

  1. The average word count of ChatGPT query fan-outs roughly doubled between October 2025 and January 2026, from about 6 words to about 12 words. Peec AI, 20 million+ fan-outs,
  2. Average ChatGPT query fan-out length peaked at about 16 words around calendar week 49 of 2025. Peec AI,
  3. In the first week of October 2025, ChatGPT issued between 2.4 and 2.8 query fan-outs per prompt across five analysed countries. In the final week of January 2026 it was 2.3 to 2.8. Peec AI,
  4. ChatGPT generates an average of 2.1 query fan-outs per prompt, Perplexity’s Sonar model 1.4, and Grok 6.8. Peec AI, 5 million fan-outs,
  5. “Best” is the word ChatGPT most often adds to its query fan-outs. “Reviews” is the third most frequently injected word. Peec AI,
  6. ChatGPT injects the word “best” into 24.3% of query fan-outs for advice-style prompts. Peec AI,
  7. ChatGPT adds the current year to its fan-out queries in 5.44% of prompts. Peec AI,
  8. Grok uses the site: operator in its fan-out searches in 18.3% of all Grok chats. Peec AI,
  9. 100% of Wirecutter and Consumer Reports appearances in Grok fan-outs are injected by Grok rather than present in the user prompt. Peec AI,
  10. 94% of ChatGPT’s fan-out queries for trending news prompts contain “2026”, even though only 1% of the prompts contain a direct reference to 2026. Peec AI (Malte Landwehr),
  11. Grok injects one brand into fan-out queries for 20% of US ecommerce prompts, ChatGPT for 5% and Perplexity for 2%. Peec AI (Malte Landwehr),
  12. Grok injects 2 to 5 brands into fan-out queries for 13% of US ecommerce prompts, ChatGPT for 8% and Perplexity for 0%. Peec AI (Malte Landwehr),
  13. In 13% of checked cases, ChatGPT injects brand names into shopping fan-out queries even when the original prompt mentioned no brands. Peec AI,
  14. 78 of the 183 web searches GPT-5.6 started across 75 chats about one startup used the site: operator. Peec AI (Malte Landwehr),
  15. ChatGPT retrieves 37 pages per fan-out query. AirOps,
  16. 33% of pages cited by ChatGPT appear only in the SERPs for a fan-out query, not in the SERP for the original prompt. AirOps,
  17. 95% of ChatGPT’s fan-out queries have zero monthly search volume. AirOps,
  18. ChatGPT query fan-outs at domain level overlap 39% with Google SERPs. With Bing it is below 10%. Profound,
  19. 89% of prompts generate 2 to 3 fan-out queries. Profound,
  20. ChatGPT uses 2.17 fan-out queries on average, within a range of 1 to 4. Nectiv Digital,
  21. ChatGPT fan-out queries have an average length of 5.48 words, a range of 2 to 12 words, and 77% are 5 words or longer. Nectiv Digital,
  22. The most common terms in ChatGPT fan-out queries are reviews, 2025, free, features and comparison. Nectiv Digital,

AI search still runs on the English web

  1. 43% of ChatGPT query fan-outs for non-English prompts are performed in English. Peec AI, 10 million+ searches,
  2. ChatGPT supplements non-English queries with additional English-language research in nearly 78% of cases. Peec AI,
  3. Turkish-language ChatGPT queries include English-language research 94% of the time. Spanish-language queries 66% of the time. Peec AI,
  4. No non-English language falls below a 60% rate of English-language research in ChatGPT. Peec AI,
  5. Fan-out behaviour is virtually identical across all countries and languages, showing that algorithm changes roll out globally rather than regionally. Peec AI,

Inside ChatGPT’s retrieval stack

  1. ChatGPT’s own retrieval index, Labrador, is not a single index but a family of vertical indexes covering web, PDF, YouTube, news, arXiv, Wikipedia, local, finance, legal, medical, shopping and images. Peec AI (Tomek Rudzki),
  2. Between 21 May and 21 July 2026, ChatGPT’s server-sent events exposed a result\_source field carrying only four values: Labrador (OpenAI’s own index), Bright, Oxylabs and SERP. Peec AI (Tomek Rudzki),
  3. On top of its own crawler, ChatGPT retrieves from at least eight outside providers. Peec AI (Tomek Rudzki),
  4. In mid-August 2026, ChatGPT’s “prefer-index-over-serp-v3” A/B test, which serves results from OpenAI’s own index instead of scraped search results, affected 8% of chats. Peec AI (Tomek Rudzki),
  5. As of 2 September 2026, ChatGPT was running at least five separate shopping ranking experiments, two of them on OpenAI’s own index. Peec AI (Tomek Rudzki),
  6. ChatGPT’s shopping ranking stack reranks a pool of 400 candidate products and narrows the field to 10 sources with a BM25 lexical component. Peec AI (Tomek Rudzki),
  7. ChatGPT’s shopping retrieval runs approximate nearest neighbor vector search at two dimensions, 12,288 and 4,096. Peec AI (Tomek Rudzki),
  8. ChatGPT had crawled 6 million pages of a one-billion-page test website by early September 2026, at a rate of 35,000 requests per hour. Peec AI (Metehan Yesilyurt),
  9. OpenAI began building its own search index in 2023, aiming to answer 80% of queries from it by the end of that year. Nick Turley (OpenAI) in US v. Google testimony,
  10. Even with full access to Google’s index data, OpenAI would need at least five years just to determine whether answering 100% of queries from its own index is achievable. Nick Turley (OpenAI) in US v. Google testimony,
  11. 58% of pages take longer than 0.8 seconds to load, while Microsoft’s Web IQ grounding platform advertises sub-165ms P95 latency. HTTP Archive Web Almanac and Microsoft, via Peec AI,
  12. The reranker BGE-reranker-large puts roughly 24% of its attention on punctuation and special tokens for a typical passage. Peec AI (Metehan Yesilyurt),
  13. The reranker ms-marco-MiniLM-L6-v2 scores a descriptive product passage at 0.0019% and a direct shortlist passage above 99.9% for the same best-tools query. Peec AI (Metehan Yesilyurt),
  14. You can forcibly refresh ChatGPT’s cache of a URL by requesting the content of the URL manually: set external web access to FALSE in the API and ChatGPT will update the cache with the new content. Profound,

API versus user interface: the same prompts, different answers

  1. Comparing the same prompts in the ChatGPT user interface and the ChatGPT API, the average brand’s visibility changes by 41%. Peec AI (Malte Landwehr),
  2. The most visible brand in the ChatGPT user interface does not make the top 3 in the ChatGPT API for the same prompts. Peec AI (Malte Landwehr),
  3. The ChatGPT API produces at most 7 fan-out queries per prompt. The ChatGPT user interface produces up to 20. Peec AI (Malte Landwehr),
  4. The share of UGC sources cited drops from 2% in the ChatGPT user interface to 0% in the ChatGPT API. Editorial sources drop from 5% to 1%. Peec AI (Malte Landwehr),
  5. Competitor domains are cited in 52% of ChatGPT API answers and in 38% of ChatGPT user interface answers. Peec AI (Malte Landwehr),
  6. Two of the top 25 source domains in the ChatGPT user interface do not appear among the top 100 sources in the ChatGPT API. Peec AI (Malte Landwehr),
  7. Listicles fall from 25% of cited pages in the ChatGPT user interface to 4% in the ChatGPT API. Comparison pages fall from 18% to 2%. Peec AI (Malte Landwehr),
  8. Product pages and how-to guides grow from 34% of cited pages in the ChatGPT user interface to 63% in the ChatGPT API. Peec AI (Malte Landwehr),
  9. ChatGPT API answers are 31% longer than ChatGPT user interface answers. Peec AI (Malte Landwehr),

How AI agents actually read your site

  1. In ChatGPT Deep Research sessions, a successful in-page find command triggers a re-read whose window contains the matched line 95% of the time. Peec AI,
  2. On pages with under 20 links, about 78% of ChatGPT Deep Research’s first read is real content. On pages with 20 to 59 links it is 55%, and on pages with 60 or more links only 33%. Peec AI,
  3. A typical ChatGPT Deep Research first read contains around 50 link markers. Peec AI,
  4. Each ChatGPT Deep Research page open returns a capped window of roughly 5,000 to 6,000 characters. Peec AI,
  5. The first open of a page returns a median of about 5,700 characters to ChatGPT Deep Research. A re-read returns about 6,200. Peec AI,
  6. A successful find command returns a median of about 435 characters to ChatGPT Deep Research. A missed find returns only about 49. Peec AI,
  7. After a find miss, ChatGPT Deep Research tries a different keyword about 60% of the time and abandons the page for a fresh Bing query roughly a quarter of the time. Peec AI,
  8. About one in five ChatGPT Deep Research re-reads is a plain continuation without a keyword trigger. Peec AI,
  9. Bing snippets shown to ChatGPT Deep Research cap at about 285 characters per result, with a median of about 250. A full Bing result list has a median of about 740 characters per search. Peec AI,
  10. ChatGPT Deep Research sets topn as high as 10 results on opening exploratory queries, uses topn=3 as its steady-state default, and drops to topn=2 when verifying a fact it already expects to confirm. Peec AI,
  11. ChatGPT Deep Research used the click command zero times across all logged sessions. Peec AI,
  12. 28% of first clicks by ChatGPT Deep Research and Agent Mode go to the position 1 ranked source per fan-out query. Peec AI (David Konitzny),
  13. The agentic CTR curve in ChatGPT declines from 28% at position 1 to 18% at position 2, 12% at position 3 and 4% at position 10. Peec AI (David Konitzny),

Crawlers, rendering and technical GEO

  1. Crawlers that render JavaScript include Google, Bing, Yandex, Naver, Baidu, Apple, adidx, ClaudeBot, Meta-ExternalAgents, MistralAI-User, Bytespider and DuckAssistBot. Merj, Tech SEO Connect,
  2. Crawlers that do NOT render JavaScript include CCBot, GPTBot, OAI-SearchBot, ChatGPT-User, Claude-User, Claude-SearchBot, PerplexityBot and Perplexity-User. Merj, Tech SEO Connect,
  3. In an experiment with 8 test scenarios, none of the 5 tested AI systems (ChatGPT, Claude, Gemini, Perplexity, Google AI Mode) extracted JSON-LD schema markup during a direct live fetch, even when the information was visible nowhere else on the page. searchVIU,
  4. Gemini was the only AI system to render JavaScript during a live fetch, finding 4 of 8 hidden product prices, the best result of the 5 systems tested. searchVIU,
  5. ChatGPT found 3 of 8 hidden product prices (37.5%), extracting only prices present in visible HTML. Claude found 0 of 8. searchVIU,
  6. Perplexity found 0 of 8 hidden product prices before indexing and 1 of 8 after. Google AI Mode found 0 of 8 before indexing and 2 of 8 after, capturing only 3 of 7 products in its index. searchVIU,
  7. Hidden Microdata and hidden RDFa prices were recognised by none of the 5 tested AI systems. Visible Microdata and RDFa were read by ChatGPT and Gemini, but not by Claude, Perplexity or Google AI Mode. searchVIU,
  8. Replacing JavaScript-dependent rendering with progressive hydration, so critical content is served in HTML, lifted a client’s mention share in AI engines from 57.2% to 64.5% and its citation share from 6.8% to 7.3%. Brainlabs AI Report
  9. Common Crawl contains over 10 petabytes of data and has collected more than 300 billion web pages since 2008. Common Crawl,
  10. Over 80% of GPT-3’s training data came from a filtered version of Common Crawl. Common Crawl,
  11. Meta’s meta-webindexer crawler massively increased its crawling across Peec AI-tracked websites in 2026, with peak activity in calendar weeks 23, 26 and 32. Peec AI (Malte Landwehr),
  12. In most weeks, Meta’s meta-webindexer crawls less than Bytespider, Applebot and OAI-SearchBot. Peec AI (Malte Landwehr),
  13. Google has a limit of 400 billion indexed documents, and after 128 days without crawling a document drops from the index. OMT Summit 2026,
  14. Google crawls atom feeds easily 30x more often than XML sitemaps. Campixx 2026,
  15. White text on a white background still works as a manipulation technique for ChatGPT, Perplexity, AI Overviews and featured snippets. Claneo,
  16. google.com/goto URLs appear in 0.02% of AI Overviews sources, 0.01% in AI Mode, 0.006% in Claude and 0.004% in Perplexity. Gemini, ChatGPT, Grok and Mistral never return them. Peec AI (Malte Landwehr),

The llms.txt question

  1. llms.txt files account for 0.0001% of ChatGPT citations and 0.0007% of GPT-Search-API citations. Peec AI (Malte Landwehr),
  2. With the switch to GPT-5.6, the number of .txt and .md files cited by ChatGPT grew 300%, and the number of llms.txt and llms-full.txt files cited doubled. Peec AI (Malte Landwehr),
  3. Across 8 LLMs, .md files account for 0.013056% of all citations, .txt files 0.001053%, .json files 0.000160%, llms.txt 0.000235% and llms-full.txt 0.000008%. Peec AI (Malte Landwehr),
  4. Perplexity’s llms-full.txt file has more than 7,000 lines and over 30,000 words. Peec AI,
  5. More than 10 tools already exist that automatically create an llms.txt file. Peec AI,
  6. As of July 2025, not a single LLM-driven search and answer engine had announced support for the llms.txt protocol. Peec AI,

Local search and maps in AI

  1. Prompts containing “near me” trigger a map in ChatGPT 94% of the time. Peec AI (Malte Landwehr),
  2. City-level prompts trigger a map in ChatGPT 48% of the time, borough-level prompts 59%, neighbourhood-level prompts 74%, and prompts naming a specific location 80%. Peec AI (Malte Landwehr),
  3. ChatGPT’s map reservation partners are Yelp, OpenTable and Resy. Peec AI (Malte Landwehr),
  4. ChatGPT populates its maps with business entries from at least 7 providers, including official partners Yelp and Tripadvisor and Google Maps data via a third-party scraping provider. Peec AI (Malte Landwehr),

AI shopping and ecommerce

  1. 83% of ChatGPT carousel products are strong matches to products in Google’s top 40 organic shopping results. Only 11% are strong matches to Bing’s top shopping results. Peec AI,
  2. 60% of ChatGPT carousel product matches come from the top 10 Google Shopping organic results, and nearly 84% from the top 20. Products ranked 21 to 40 account for around 16%. Peec AI,
  3. ChatGPT product carousels usually feature 8 products. Peec AI,
  4. 136 million ChatGPT product carousels are shown to European consumers every month. Precis,
  5. 64% of the products Amazon’s Rufus AI assistant recommends do not rank in the top 10 of Amazon’s organic search results. Autopilot and The Order, The Third Shelf
  6. 74% of shoppers say they trust the product recommendations they receive from AI chat. Salesforce, Holiday Retail Predictions 2026,
  7. 45% of consumers planned to use generative AI tools for Black Friday shopping. BCG,
  8. 39% of consumers have used generative AI for online shopping, and 53% planned to in 2025. Adobe,
  9. 92% of consumers who use generative AI for shopping say it enhances their experience. Adobe,
  10. 29% of consumers have used generative AI for travel-related tasks, and 84% of them say it enhances their experience. 27% have used it for financial topics. Adobe,
  11. 50% of ecommerce searches in Google have a product grid. OMT Summit 2026,
  12. Product grid click uplifts: titles +54%, split SEO and SEA product feeds +60%, product detail page optimisation +100%, images +115%, variants and item group ID +250%. OMT Summit 2026,
  13. Optimising product attributes in the product grid lifted revenue 205%. OMT Summit 2026,
  14. Google Merchant Center product titles allow 150 characters but only the first 70 are displayed. OMT Summit 2026,
  15. One of the largest German online shops already gets 10% of its product detail page SEO traffic from ChatGPT. Campixx 2026,
  16. Over 10% of 6.8 million monitored online shops disappeared over a six-month period. ShopRank,
  17. From August 2025 to January 2026, 8% of Shopify stores and 20% of WooCommerce stores closed. As of January 2026, 76% of WooCommerce stores are dead. ShopRank,

Ads inside AI answers

  1. 60% of ChatGPT chats in the USA contained an ad on 13 August 2026, down from 73% on 22 July 2026. Peec AI (Malte Landwehr),
  2. The share of ChatGPT chats containing an ad in Brazil rose from 0% in early August 2026 to a peak of 47% around 10 August 2026. Peec AI (Malte Landwehr),
  3. Within 2 weeks of launch, ChatGPT ads in Brazil and Mexico appeared in roughly 35% to 42% of chats, matching the levels in Japan and South Korea. Peec AI (Malte Landwehr),
  4. Through July 2026, ChatGPT ads appeared in about 30% of chats in South Korea and about 22% of chats in Japan. Peec AI (Malte Landwehr),
  5. Almost 50% of sources used by LLMs can be influenced with affiliate marketing. Peec AI (Malte Landwehr),
  6. In one insurance niche, over 2% of LLM sources are paid advertorial placements. Peec AI (Malte Landwehr),
  7. The advertorial share of LLM citations in the German insurance space stayed between 2.2% and 2.7% from October 2025 to February 2026. Peec AI (Malte Landwehr),

AI referral traffic: smaller, but far more valuable

  1. From July 2024 to February 2025, traffic from AI search increased 1,200% for banks, 1,200% for retail and 1,700% for travel. Adobe,
  2. Compared to other traffic sources, generative AI traffic has 8% higher engagement, 12% more page visits and a 23% lower bounce rate. Adobe,
  3. Conversion rates to ecommerce websites by channel: ChatGPT 11%, direct 10%, paid search 9%, SEO 5%, email 5%, display 4%, social 4%. SimilarWeb,
  4. Conversion rate in the US is 6.93% for ChatGPT traffic versus 5.44% for Google traffic. Growth Memo,
  5. The average length of a search on Google is 3.2 words. On ChatGPT it is 80. Growth Memo,
  6. The value of a click from an LLM is 4x higher than a click from organic search. Precis B2Breakfast 2026,
  7. Visitors coming from AI search spend 3x longer on the website. Lureon Report
  8. Webflow has a 6x higher conversion rate from ChatGPT than from Google. Graphite,
  9. LLM traffic for some clients converts at up to 20x the rate of typical SEO traffic. Peec AI via Search Engine Journal,
  10. Clients grow their LLM traffic by 100% every 2 to 3 months. Peec AI via Search Engine Journal,
  11. ChatGPT traffic is 0.5% to 2% of total website visits for most brands. Peec AI (Malte Landwehr),
  12. ChatGPT sends 12 million clicks per month to websites in Germany, and Idealo captured 2% of them. SimilarWeb,
  13. Idealo was the most visible ecommerce destination in ChatGPT in the first half of 2025, ahead of Amazon and eBay. SimilarWeb,
  14. When an AI assistant recommends a brand, users are within the following week 182% more likely to Google the brand, 117% more likely to visit its website and 185% more likely to view its product on a retailer’s site. Kevin White / Scrunch AI,

What public companies report about AI traffic

  1. Auto Trader reports 9 million monthly unique visitors, of which 80% arrive direct, 13% via organic search, 4% via paid and less than 1% via AI chatbots. Auto Trader,
  2. Expedia reports traffic from new AI experiences is less than 1.5% of its total traffic. Expedia,
  3. CarGurus reports AI traffic is a low double-digit percentage of its total traffic and growing quickly. CarGurus,
  4. CarGurus reports AI-driven traffic submits leads at nearly 50% higher rates than traditional SEO traffic. CarGurus,
  5. LegalZoom reports traffic from AI platforms converts at higher rates and shows a higher propensity to upgrade subscriptions than its other traffic sources. LegalZoom,
  6. Etsy reports agentic traffic in Q4 2025 was 15x higher than the prior year, though it remains a small portion of overall traffic and gross merchandise sales. Etsy,
  7. Shopify reports orders from AI search are up 15x since January 2025, albeit on a small base. Shopify,
  8. eBay reports AI-driven traffic is very small but high-intent and converts at a high rate. eBay,
  9. Airbnb reports traffic from chatbots converts at a higher rate than traffic from Google. Airbnb,
  10. Tripadvisor expects mid-to-high-teens percentage revenue declines, largely driven by SEO traffic headwinds. Tripadvisor,
  11. Angi Inc. reports SEO currently generates around 7% of its service request lead revenue. Angi Inc.,
  12. CoStar Group reports brand, direct, SEM, display, social, email, SEO and AEO traffic remain 99.5% of its top-of-funnel sources, leaving AI and GEO at under 1%. CoStar Group,
  13. News Corp reports Realtor.com properties received 29% of total visits to all real estate portals, triple the visits of Homes.com and double those of Redfin. News Corp,
  14. Pinterest reports 85% of its users access the platform directly via its mobile app, and 100% of users are logged in. Pinterest,

Revenue and attribution

  1. Peec AI’s B2B SaaS and services clients often report 10% to 20% of revenue coming from AI search. Peec AI (Malte Landwehr),
  2. The highest AI search revenue share reported by a Peec AI client is 30%. Peec AI (Malte Landwehr),
  3. B2B SaaS and tech companies typically attribute 4% to 10% of revenue to AI search. Ethan Smith / Graphite,
  4. A study of Amsive client data found 4% to 5% of revenue comes from AI search. Lily Ray / Amsive,
  5. About 10% of brand discovery already happens via LLMs. Peec AI (Malte Landwehr),
  6. For B2B SaaS businesses, self-reported attribution regularly shows 20% of leads influenced by LLMs, while click-based tools show close to zero. Peec AI (Malte Landwehr),
  7. One online shop generates 2 million euros in revenue every month from clicks coming out of ChatGPT. Peec AI (Malte Landwehr),
  8. Flow Agency receives 2.4% of its traffic from ChatGPT but 20% of its leads. Flow Agency,
  9. According to HubSpot, Flow Agency gets 0% of leads from LLMs. According to a form survey, it is 22%. Flow Agency,
  10. In 2025, 34% of qualified B2B leads originated from AI-powered platforms like ChatGPT and Perplexity. Lureon Report
  11. AI search is the fastest-growing channel for B2B lead generation. Lureon Report
  12. 66% of fintech CEOs report client growth from AI search. IDC via Lureon Report

What actually moves visibility: tests and case studies

  1. Adding statistics and quotations to content lifted AI visibility roughly 30% to 40%. Princeton GEO Study via Brainlabs AI Report
  2. Rewriting vague headings such as “About X” into direct-answer headings such as “How to fix X” lifts the average citation rate 41%, from 29% to 41%. Brainlabs AI Report
  3. Correcting factual errors in existing content grew citations steadily within 2 weeks of fixing an inaccurate AI response. Brainlabs AI Report
  4. Updating website metadata grew LLM-referred new users 787% and LLM referral traffic 174% for a North American nutritional beverage brand. Brainlabs AI Report
  5. Updating website metadata grew AI Overview citations 54%, LLM citations 33%, Google impressions 34% and Google clicks 10% for a North American nutritional beverage brand. Brainlabs AI Report
  6. Ramp increased LLM visibility 78% by updating existing pages for zero-volume and low-volume queries. George Bonaci / Ramp,
  7. Ramp’s narrow micro-pages for niche product use cases achieved only a citation rate of about 2%. George Bonaci / Ramp,
  8. Ramp increased AI visibility in a test cohort by 24.6% by rebuilding content around jobs-to-be-done instead of keywords. George Bonaci / Ramp,
  9. Ramp’s programmatic pages built on proprietary spend data increased sessions 149% between June and August 2025. George Bonaci / Ramp,
  10. In an Omio A/B test, adding brand USP modules to pages lifted LLM referral traffic 18% without harming Google organic traffic, and the change was rolled out. Omio via Tom Critchlow,
  11. A second Omio A/B test variant was not shipped because it would have cut Google organic sessions 6.5%, Omio’s biggest organic channel. Omio via Tom Critchlow,
  12. Angling ecommerce shop fischen.ch increased its AI citations 35% after publishing 100 community-generated content pieces across 79 shop URLs. Guru webinar,
  13. After fischen.ch published community expert content, the number of its category pages cited in AI answers grew 104%. Guru webinar,
  14. Ski touring brand Dynafit roughly tripled AI source usage of its online shop after embedding community expert content on product detail pages. Guru webinar,
  15. Dynafit’s brand mentions in AI answers rose from a 76.8% baseline by about 15 percentage points after embedding community expert content. Guru webinar,
  16. Duckfund achieved a 16% ChatGPT visibility lift in 90 days across commercial real estate financing queries. Peec AI,
  17. Lula became the number 1 most-cited domain on ChatGPT for South African SME funding prompts and gained a 3% share of voice boost in AI search. Peec AI,
  18. Mint Position’s GEO service line drove 20% year-over-year revenue growth for the agency, and 70% of its clients actively use their Peec AI workspace login. Peec AI,
  19. Merge drove a 7x increase in demo requests from AI search and a 10x boost in LLM search traffic. Peec AI,
  20. Merge sees 6x higher conversion rates from AI search compared to other channels. Peec AI,
  21. Merge’s LLM-optimised article on best unified API platforms garnered 2,600 LLM mentions in a single month. Peec AI,
  22. Momentum boosted its AI search visibility 10x across all major LLMs, doubling visibility within one month and seeing a 2x increase in session visits from AI search. Peec AI,
  23. Momentum optimised 100 articles for AI visibility, each targeting a specific prompt. Peec AI,
  24. Planeco Building increased SEO and AI search leads 5x, and grew its citation share from 50% to over 130%. Peec AI,
  25. Heyflow’s AI search visibility increased from 10% to 40%. Peec AI,
  26. Radyant boosts AI search visibility across more than 50 startups and scaleups. Peec AI,
  27. One website grew its ChatGPT citations by more than 1,000% within a few weeks. Peec AI (Malte Landwehr),
  28. Ranking-style prompts lift brand mentions by about 20% on average, and keyword-style prompts produce up to 25% more brand mentions than conversational prompts. Peec AI (Malte Landwehr) via Search Engine Journal,
  29. 88% to 92% of human prompt pairs sharing the same intent sit above 0.50 cosine similarity, and about 95% sit above 0.40. Peec AI (Malte Landwehr) via Search Engine Journal,
  30. Brand mention probability averages 4.9% and drops 2.4 percentage points when prompt similarity falls into the 0.35 to 0.39 bin. Peec AI (Malte Landwehr) via Search Engine Journal,
  31. 2% to 5% of customers leave a review unprompted, skewed to 1 star. With active asking, 5% to 20% leave reviews, skewed to 5 stars. Campixx 2026,
  32. Sixt lifted its TrustPilot score from 1.3 to 4.2 by actively asking customers for reviews. Campixx 2026,

B2B buying behaviour

  1. 51% of B2B buyers start their research with AI search instead of Google, up from 29% in 2025. Precis B2Breakfast 2026,
  2. 60% of B2B buyers use AI tools during the purchasing process. Precis B2Breakfast 2026,
  3. The average B2B purchasing process now takes 3 months instead of 5. Precis B2Breakfast 2026,
  4. 1 in 5 dollars spent in B2B is influenced by AI Overviews. Precis B2Breakfast 2026,
  5. 82% of B2B buyers end up purchasing a product or service that was on their day 1 list. Precis B2Breakfast 2026,
  6. 41% of buyers trust AI search results over ads. Lureon Report
  7. Prompts from B2B buyers are 15 to 23 words long. Lureon Report
  8. Generative AI chatbots are the number 1 source influencing shortlists for B2B SaaS buyers. G2 Buyer Behavior Report
  9. 16% of Surfer’s registered users coming via Google search are SEOs, general managers or business owners. Via LLMs it is 21%. Surfer, BrightonSEO
  10. Via Google search, content writers find Surfer. Via LLMs it is owners, CEOs and general managers. Surfer, BrightonSEO

Consumer behaviour in AI-assisted research

  1. 50% of Europeans researching banking or insurance products now begin their research by using AI tools. BCG B2C Banking and Insurance Survey 2026,
  2. 53% of consumers researching banking or insurance products in six European markets use AI somewhere in their journey. BCG B2C Banking and Insurance Survey 2026,
  3. 89% of consumers who use AI to research banking or insurance products start that research directly in an LLM. BCG B2C Banking and Insurance Survey 2026,
  4. 72% of consumers who use AI to research banking or insurance products have specific providers named inside an AI-generated answer. BCG B2C Banking and Insurance Survey 2026,
  5. 60% of consumers who see providers named in an AI answer while researching banking or insurance are shown a provider they had not previously considered. BCG B2C Banking and Insurance Survey 2026,
  6. 47% of consumers who use AI to research banking or insurance products have their shortlist changed by AI answers. BCG B2C Banking and Insurance Survey 2026,
  7. 23% of consumers who use AI to research banking or insurance products changed their final provider choice as a direct result of AI, and 32% say AI influenced their choice. BCG B2C Banking and Insurance Survey 2026,
  8. 87% of consumers who use AI to research banking or insurance products say AI makes exploring options faster, and 91% would use AI tools again. BCG B2C Banking and Insurance Survey 2026,
  9. 46% of consumers who use AI to research banking or insurance products use AI four or more times for the same product. BCG B2C Banking and Insurance Survey 2026,
  10. A Spanish digital insurer founded in 2021 became the first in its sector to launch a native app inside ChatGPT in early 2026, returning real-time personalised home insurance quotes without the user leaving the chat. BCG,
  11. 37% of consumers start searches with AI rather than Google, but 85% still cross-reference through traditional search before converting. Eight Oh Two AI and Search Behavior study
  12. 58% of users used an answer engine in a recent product research or shopping journey within the last week. Profound,
  13. 80% of buyers make at least half of their decisions with answer engines. Profound,
  14. Relevance in the buying process for US internet users: during discovery and ideation AI scores 35% versus search at 14%. SimilarWeb Generative AI Brand Visibility Index
  15. During research and comparison, AI scores 30% versus search at 20%. During the decision what to buy, AI scores 31% versus 15%. For where to buy and price comparison, AI scores 24% versus 22%. SimilarWeb Generative AI Brand Visibility Index
  16. Consumer trust in purchase recommendations: brand websites 10%, search engines 11%, social media 15%, generative AI 18%, physical stores 19%. eMarketer via Accenture Consumer Pulse,
  17. More U.S. adults say AI chatbots help rather than hurt their productivity (30% versus 5%), how informed they are (28% versus 5%), their creativity (21% versus 11%) and their happiness (8% versus 5%). Pew Research Center,
  18. 31% of U.S. adults predict AI will negatively impact them personally over the next 20 years, while 23% predict a positive personal impact. Pew Research Center,
  19. 40% of U.S. adults predict AI will have a negative impact on society over the next 20 years, while 16% predict a positive impact. Pew Research Center,
  20. 54% of New York Times readers prefer AI-written content over human-written content, based on 86,000 people who did not know which content was AI-written. Kevin Roose / New York Times,
  21. Expert readers are 8x more likely to choose an AI’s writing over a human expert’s writing, and evaluated the general writing quality of AI almost twice as good. MIT and Columbia,

Brand perception, bias and hallucination

  1. Across 904 real brands tested, AI models describe companies with positively-valenced names more favorably. A brand name that is 11% more positive leads to a 0.8% more positive description of the company by AI assistants. Peec AI (Jan Ehrlinspiel),
  2. Asked directly about Costco, 0% of LLM answers mention a pharmacy. Asked for supermarkets known for pharmacies, 80% of answers contain Costco. Peec AI (Malte Landwehr),
  3. ChatGPT picks the wrong domain in up to 99% of cases for startups whose name is shared with another company. Peec AI (Malte Landwehr),
  4. 5% of GPT-5.6’s site: searches about one startup ask for a parked domain instead of the company’s real domain. For other startups the rate is 6%, 7%, 7% and 16%. Peec AI (Malte Landwehr),
  5. On 11 September 2026, 0% of the claims LLMs made about whether Peec AI can track Naver AI Briefing were correct. Three days after a single Reddit post answered the question, 83% were correct. Peec AI (Malte Landwehr),
  6. 74% of the pricing sources cited for a brand are the brand’s own pricing page, 23% are the brand’s secondary pricing content such as help center articles and press releases, and 3% are competitor pricing pages. Peec AI (Malte Landwehr),
  7. Gemini and Perplexity repeat seeded misinformation in 37% to 39% of answers. ChatGPT-4 and ChatGPT-5 do so in under 7%. Ahrefs AI Search Benchmark Report Q1
  8. 75% of negative Sitejabber reviews of Revolut come from accounts with only one review. Peec AI,
  9. Asked repeatedly for a random number between 1 and 100, LLMs answer 42, 47, 57 or 73 in 66% of cases. Peec AI (Malte Landwehr),
  10. Asked for a random number between 1 and 100, Mistral answers 47 or 73 in 93% of cases, Claude answers 47 in 92%, ChatGPT answers 30, 47 or 57 in 88%, Perplexity answers 42 or 57 in 87%, Copilot answers 47 in 83%, Google AI Overview answers 42 or 73 in 78%, Gemini answers 42 in 75%, Grok answers 73, 42 or 57 in 75%, and Google AI Mode answers 42 in 52%. Peec AI (Malte Landwehr),
  11. Across more than 15,000 chats in 10 languages from 18 countries, LLMs predict Spain will win the FIFA World Cup. AI Overviews predict it with 88% confidence, Copilot with only 39%. Peec AI (Malte Landwehr),
  12. Asked in Spanish from a Spanish IP, ChatGPT picks Spain to win the World Cup in 86% of cases. The all-country average is 61%. Peec AI (Malte Landwehr),
  13. ChatGPT is the most consistent LLM across countries with 94% answer agreement. AI Overviews are the least consistent at 29%. Peec AI (Malte Landwehr),
  14. LLMs use web search in 62% of World Cup prediction chats, and Grok regularly cites X posts with 0 likes from accounts under 10,000 followers. Peec AI (Malte Landwehr),
  15. ChatGPT shows 90% fewer follow-up questions since the GPT-5.5 update of 24 June 2026. Peec AI (David Konitzny),
  16. ChatGPT follow-up questions dropped 50% in late May 2026 but recovered to previous levels by mid June. Peec AI (David Konitzny),
  17. A website marketed as “built for AI citations” accounts for 1 of the last 1.5 million citations Peec AI tracked for the prompts that site targets, less than 0.0001%. Peec AI (Malte Landwehr),

The risk of AI-generated content

  1. 36% of brands featured in one AI content generation tool’s success stories showed massive visibility and traffic drops in Google. Peec AI,
  2. 3 out of 4 brands featured on another AI content generation platform’s website have suffered significant Google visibility losses. Peec AI,
  3. 57% of the flagship case-study customers of a popular AI content tool saw their Google rankings tank. 22% show a perfect Mount AI shape. Malte Landwehr,
  4. 100% of pages removed by Google’s March 2024 spam update had AI-generated posts, and half of those sites had 80% to 90% of their content generated by AI. Originality.AI,
  5. Google tested AI content detectors on 500 million web pages in research published in 2020 and 2021, and employs 16,000 human contractors on its Search Quality Raters team. Google, via Peec AI,

Publishers and the traffic recession

  1. Organic search traffic declined 42% from Q1 2023 to Q4 2025, based on 64 publisher sites generating over 1 billion clicks per quarter. Define Media Group,
  2. Organic search traffic to publishers declined noticeably in Q2 2024 when AI Overviews launched and again in Q2 2025 when AI Overviews expanded. Define Media Group,
  3. Google Discover is now driving more traffic to publishers than organic search. Define Media Group,
  4. Organic traffic to news sites has dropped 26% since Google launched AI Overviews. SimilarWeb,
  5. Between January 2024 and May 2025, news-related prompts in ChatGPT rose 212%, while equivalent Google searches declined 5%. SimilarWeb,
  6. Clicks from ChatGPT to US news websites increased 25x from H1 2024 to H1 2025, from 1 million to 25 million. SimilarWeb,
  7. IAC and Dotdash Meredith report a 50% drop in Google search referrals over two years. IAC,
  8. People Inc., formerly Dotdash Meredith, reports Google Search declined from 50% to 28% of its sessions and now represents 18% of its digital revenue. People Inc.,

The GEO industry itself

  1. AEO and GEO is the number 1 strategic priority for CMOs in 2026, and 32% of CMOs listed it as their top priority. Conductor, State of AEO/GEO in CMO Investment Report
  2. In 2025, enterprises allocated 12% of their digital budgets to AEO and GEO. Conductor, State of AEO/GEO in CMO Investment Report
  3. 94% of companies plan to increase their GEO budget in 2026. Conductor, State of AEO/GEO in CMO Investment Report
  4. 64% of companies plan to upskill existing team members in GEO, and 29% plan to hire for AEO-specific roles. Conductor, State of AEO/GEO in CMO Investment Report
  5. 35% of B2B marketers listed GEO as their top priority, overtaking traditional SEO at 29%. Lureon Report
  6. 90% of CMOs globally agree that generative AI is already reshaping how consumers discover and evaluate brands. BCG CMO Survey 2026,
  7. 61% of senior banking and insurance marketing leaders say AI-driven searches already have a large or very large impact on customer acquisition. BCG B2B Banking and Insurance Survey 2026,
  8. 61% of banking and insurance marketing leaders rate their own GEO and AEO maturity as Established, Advanced or Leading, but only 37% have a defined framework for measuring it. BCG B2B Banking and Insurance Survey 2026,
  9. 73% of banking and insurance marketing leaders expect GEO and AEO to become a higher priority over the next 12 months, and 88% are already researching how AI is changing customer decision-making. BCG B2B Banking and Insurance Survey 2026,
  10. GEO and AEO is owned by Digital Marketing in 31% of banking and insurance organisations and by SEO or Content in 28%. Only 15% run a cross-functional squad. BCG B2B Banking and Insurance Survey 2026,
  11. GEO and AEO is discussed at marketing level in 79% of banking and insurance organisations and at executive level in 77%, but by Analytics in only 39%, Tech in 26% and Legal or Compliance in 25%. BCG B2B Banking and Insurance Survey 2026,
  12. GEO and AEO already accounts for around 10% of first-touch entry channels in digitally originated banking and insurance acquisition journeys. BCG B2B Banking and Insurance Survey 2026,
  13. 84% of Germany’s top 100 SEO agencies now offer GEO services, 30% list GEO as one of their focus areas, and 7% primarily offer GEO. iBusiness,
  14. An SEO agency with 30 employees pays 5,000 to 6,000 euros per month in pure LLM token costs. A small agency pays under 1,000 euros. Campixx 2026,
  15. 357 teams signed up for Peec AI’s $10,000 MCP workflow contest, and 67 submitted an entry. Peec AI,
  16. The energy used for a Gemini text prompt has dropped 33x in only 12 months, and the carbon footprint 44x. Google,

Peec AI by the numbers

  1. Peec AI grew from $0 to more than $4M ARR in 10 months and reached $10M ARR 16 months after launch. Peec AI,
  2. Peec AI grew from 25 people and $5M ARR to 70 people and $10M ARR in 5 months. Peec AI,
  3. Peec AI is a top 2% company globally by ARR growth. Peec AI,
  4. Peec AI raised a $21 million Series A led by Singular, bringing total funding to $29 million. Peec AI,
  5. Peec AI onboarded more than 1,300 brands and agencies between February 2025 and November 2025, and adds 300+ customers per month. Peec AI,
  6. Peec AI passed 2,000 paying customers within 14 months and helps more than 2,500 companies understand how they are represented in ChatGPT, Gemini and other engines. Peec AI,
  7. Peec AI has analysed more than 400 million LLM chats across more than 3,000 customers. Peec AI (Malte Landwehr),

What these numbers mean

The click has stopped being the unit of value. ChatGPT is roughly 99% zero-click and Google AI Mode about 95%. Web analytics misses more than 95% of the impact of AI search. If your reporting still counts sessions, it is measuring the small residue of the channel and calling it the channel. The number to track is whether you appear inside the answer, not how many people leave it.

Answer engines have learnable preferences, and they are not subtle. Reddit is the most-cited domain on every major engine. Wikipedia leads inside ChatGPT, YouTube inside both of Google’s AI surfaces, X inside Grok. Listicles and comparison pages get cited far out of proportion to how many of them exist. Engines rewrite your prompt into sub-queries before they search, inject words the user never typed, and research on the English-language web even when asked in another language. None of that shows up in a rank tracker, and all of it is measurable.

Citations follow a winner-take-most distribution. A large share of URLs never get cited at all, and a small fraction of them take most of the citations. That is bad news if you are starting from nothing and good news if you are willing to do the work, because the same asymmetry means a modest number of well-placed third-party mentions moves the needle more than a large volume of your own pages.

The traffic that does arrive is worth more than what it replaced. ChatGPT converts at 11% on ecommerce sites against 5% for SEO, a click from a language model has been valued at 4x an organic one, and visitors from AI search spend longer on site. 51% of B2B buyers now start their research in AI search rather than Google, and 82% of them buy something that was on their shortlist on day one. Being absent from that first list is the expensive failure, not losing a ranking.

What follows from all four is the same thing. AI search is its own channel, not a modifier on the old one, and it needs its own measurement: citations rather than rankings, placements in the third-party sources engines trust, pages structured for a machine reader, and every engine looked at separately rather than averaged. It also needs looking at often. ChatGPT reversed its treatment of machine-translated Reddit pages inside a single month, and Google went from almost no external links inside AI Overviews to 26.2% of them in nine days. A quarterly report would have missed both.

Frequently asked questions

How many people use AI search?

About half of US adults (49%) said they use AI chatbots in 2026, up from 33% in 2024. Worldwide, close to 4 billion people use a large language model every month, and monthly sessions across AI tools are now 56% the size of search worldwide and 34% the size of search in the US.

What percentage of AI searches end without a click?

ChatGPT’s zero-click rate is roughly 99% and Google AI Mode’s is about 95%. Where 35% to 40% of Google searches lead to a click, only 1% to 5% of search-like prompts in ChatGPT and Google AI Mode do.

Which websites do AI search engines cite most?

Reddit, across every major engine. The ten most-cited domains are Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium and TechRadar in that order. Wikipedia leads inside ChatGPT specifically, and YouTube inside both of Google’s AI surfaces.

How much do Google AI Overviews reduce click-through rate?

Across more than 15 studies the measured effect ranges from +19% to -85%, averaging -40% with a median of -37%. Ahrefs measured a 58% drop in clicks to number-one-ranked content, and 19.4% at position 10.

Does AI referral traffic convert better than search traffic?

On the measurements available, yes. ChatGPT traffic converts at 11% on ecommerce sites against 9% for paid search and 5% for SEO, and a click from a language model has been valued at 4x a click from organic search. The volume is the catch: ChatGPT traffic is 0.5% to 2% of total website visits for most brands.

What actually works to improve visibility in AI search?

Adding statistics and quotations to content lifted AI visibility roughly 30% to 40% in the original Princeton research. Rewriting vague headings into direct-answer headings lifted the average citation rate from 29% to 41%. Ramp increased visibility 78% by updating existing pages for zero-volume and low-volume queries, and Omio lifted referral traffic from language models 18% in an A/B test by adding brand USP modules to its pages.

Methodology

This page is maintained by Malte Landwehr, who has worked in SEO for over 20 years and is CPO and CMO of Peec AI. Every statistic on it was checked against its original source in September 2026.

The 521 entries draw on more than 100 named sources. Roughly a third come from Peec AI’s own research and from my own analyses of AI search data, collected across ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Microsoft Copilot, Claude and Grok. The rest come from primary sources only: peer-reviewed and preprint research, official company announcements and earnings calls, court filings, independent research organisations and vendor studies that publish their method. Nothing here is taken from another statistics roundup.

Every entry states the period the data covers as well as when the source published it, and 364 of the 521 figures were published in 2026. Where a number is an estimate rather than a measurement, the entry says so. Where two sources disagree, both are listed rather than averaged: the samples, engines, markets and definitions behind them differ enough that a single blended figure would be less true than either. Nothing on this page is here without a source you can open.