# ChatGPT Search Volume vs Google: The 20% Is Counted in Fetches *OpenAI says ChatGPT runs a fifth of Google's search volume. Its own definition makes each search a web fetch of publisher content. Nobody is pricing the fetch yet.* Published: 2026-11-07 | Read time: 5 min read | Author: Smalk AI Research | Source: https://www.smalk.ai/blog/chatgpt-search-volume-vs-google-fetch-economy --- OpenAI's claim that ChatGPT search volume is 20% of Google's is less a market-share headline than a supply-chain disclosure. By OpenAI's own definition, a search is any prompt that sends ChatGPT to the web, so the figure counts a trillion-scale stream of publisher pages fetched to build answers. The question for CMOs and publishers is not whether the number is inflated. It is who gets paid for the fetch. - OpenAI's Adam Fry said on November 3, 2026 that ChatGPT's search-like queries run at roughly one-fifth of Google's volume, on 1.2 billion weekly active users. - OpenAI counts a search whenever a prompt sends ChatGPT to the web, so the metric is a count of answers built from fetched pages, not of results pages served. - On Ahrefs' October 2026 panel, ChatGPT drove 0.36% of site visits against Google's 24.74%: a fifth of the questions, roughly a seventieth of the referrals. - For brands, engine share is the wrong fight, because answers on every engine are sourced from the same publisher pool. For publishers, fetch volume is inventory nobody has priced. ## What a ChatGPT search actually counts A ChatGPT search, in OpenAI's accounting, is any prompt that makes the model retrieve live web content before it answers; writing, coding and definitions are excluded. That makes it a different unit from a Google query, which is a request for a results page the user then navigates. One unit opens a door to the open web. The other consumes the open web on the user's behalf. The distinction decides the economics. A query that returns ten links creates ten chances for a click and a page with ad slots. A query that fetches ten pages creates one answer, and the slots on those ten pages never render. ## The query economy and the fetch economy, side by side In the open-web search economy, the query sat at the top of a funnel that paid everyone downstream: Google sold the slot, the user clicked, the publisher sold the impression, the brand bought both. In the agentic-web economy, the fetch replaces the click as the moment content is consumed, and nothing downstream of it bills. The engine still sells a slot beside the answer. The publisher whose page built the answer sells nothing. What carries over is intent. Fry's point that people now bring ChatGPT the questions they once typed into a search box is credible, and Google's own October 28 call said AI Mode queries run three times longer than classic ones. Intent did not disappear. The billing event attached to it did. ## What Forbes reported, and what we verified Forbes' November 5 piece by Gabriel Alin Zainescu reports Fry's figure fairly and flags what OpenAI left out: the Google baseline it used, and whether follow-ups inside one conversation count as several searches. We verified the independent estimates Forbes cites. Ahrefs modelled ChatGPT at about 12% of Google's volume in February 2026; seoClarity modelled about 7.5%, or roughly 375 billion queries a year, on October 28. Applied to Google's last disclosed figure of more than 5 trillion searches a year, OpenAI's 20% implies over 1 trillion web-fetching prompts annually. One number Forbes does not carry matters most. Ahrefs' October 2026 traffic panel across 118,671 sites, reported by Search Engine Journal, put Google at 24.74% of visits and ChatGPT at 0.36%. Set against the 20% claim, ChatGPT is a fifth of Google's questions and about one-seventieth of its referrals. The gap between those two ratios is the fetch economy, measured. ## Why the argument over the denominator misses the market ### The strongest case that the pool is expanding, not changing hands Google reported its fastest query growth in five years on October 28, and Search and other revenue still grew 12% to $63.35 billion. Search Engine Land and Ahrefs both note that OpenAI's definition flatters the comparison, and OpenAI has not published the data behind it. On this reading the total pool of questions is growing, Google is not shrinking, and the 20% is a vanity ratio marketers can safely discount until the methodology appears. ### Why even 375 billion fetches is a market with no price Take the lowest estimate and the point stands. Whether ChatGPT runs 375 billion or a trillion web-fetching prompts a year, every one of them, by OpenAI's own definition, consumes publisher content to produce a recommendation, and none of them pays the page it consumed. Google's slower yield on AI Mode shows the same thing from the incumbent's side: the answer earns less per query than the results page, so the engines will not fund the source out of their own margin. An expanding pool does not help a publisher whose share of the fetch is unpriced. It just makes the unpriced volume larger. ## What this means for brands and for publishers ### For CMOs and media buyers: stop planning around engine share The 20% invites a question about whether to move budget from Google to OpenAI. That is the wrong question. ChatGPT, AI Mode, Perplexity and Claude build their answers from a largely shared set of publisher pages, and a brand present in those pages is present in every engine's answer regardless of which engine wins the query. Treat the sourced answer, not the engine, as the planning unit; track which publisher pages engines cite for your category; and budget a test line for native ads that live on those pages rather than only in the slot beside the answer. ### For publishers: price the fetch, not the referral A fifth of Google's question volume now reaches your content through a channel that sends back one-seventieth of the traffic. Reporting that as a referral decline misreads it. It is a fetch surge, and fetch volume is inventory. Log which engines fetch which pages, treat cited pages as a sellable unit, and sell presence in them to brands, because the engines have already shown they will pay a royalty at best, never a media price. ## Three signals to watch before Google's Q4 results First, OpenAI's methodology: naming the Google baseline and the conversation-counting rule would turn a claim into a benchmark, and OpenAI's moves toward its own crawler and index suggest it wants to own that benchmark. Second, ChatGPT Ads scale: a $1 billion run rate reached around 200 days after launch, more than 100,000 advertisers by October 21, and Sponsored Agents in US testing since September 16 mean demand is already monetising the answer surface while supply is not. Third, Alphabet's Q4 call: whether AI Mode yield narrows the gap between query growth and revenue growth will tell publishers how long the engines can afford to leave the source unpaid. ## Conclusion Hold on to this: the 20% is OpenAI telling the market, in its own definition, that a trillion-scale stream of publisher pages is being fetched to build answers nobody pays the source for. Smalk AI exists for that gap: a Generative Engine Advertising network that places native ads for AI agents on the publisher pages every engine fetches, so brands buy presence in the answer wherever it is formed and publishers earn a media price for being the source. What to watch next: whether OpenAI publishes its methodology before Alphabet's Q4 call, because the first party to standardise the fetch as a unit will set its price. ## FAQ ### Does ChatGPT really handle 20% of Google's search volume? OpenAI says so, but the figure is self-reported and the company has not named the Google baseline or explained how multi-turn conversations are counted. Independent models land lower: Ahrefs estimated about 12% in February 2026 and seoClarity about 7.5% in October 2026. All three figures depend on what counts as a search. ### How is a ChatGPT search different from a Google search? A Google search returns a results page the user navigates, which creates clicks and ad impressions on publisher sites. A ChatGPT search, as OpenAI defines it, is a prompt that makes the model fetch web pages and synthesise one answer, so the content is consumed without a click. The first unit funds publishers; the second does not. ### Should media buyers shift budget from Google Ads to ChatGPT Ads? Not as a response to the 20% figure. Engine share is the wrong planning unit, because every AI engine sources its answers from a largely shared pool of publisher pages. The durable move is to buy presence in the pages engines cite for your category, which reaches the answer on every engine, and to test ad slots beside the answer as a secondary line. ### Why do publishers get so little traffic from ChatGPT if it handles so many searches? Because the answer is consumed inside ChatGPT. Ahrefs' October 2026 panel of 118,671 sites recorded ChatGPT at 0.36% of visits against Google's 24.74%, even as OpenAI claims a fifth of Google's query volume. The content is fetched and used; the user rarely leaves. That makes the fetch, not the referral, the unit publishers should measure and price.