Published
What ChatGPT and Google cite: three measured studies
Measured October 26, 28 and 29, 2026. By Corpus Plan.
Does ranking on Google get your business cited by ChatGPT?
Not in most of what we measured. On October 26, 2026, we asked Google and ChatGPT the same 16 buyer questions, one run each, and 2 of Google's 131 top-10 results were also ChatGPT sources for that question.
For 10 of those 16 questions, Google's top 10 and ChatGPT's sources shared no domain. The three studies below are counts from one run on one day each, not rates. Each finding states its date, sample and method so you can quote it on its own.
Do ChatGPT and Google's top 10 cite the same pages for "best X for a small business" questions?
Rarely, on October 26, 2026: 2 of Google's 131 top-10 results were also ChatGPT sources for the same question. ChatGPT also added words to its searches that the questions did not contain, and it cited no Reddit page while Google ranked Reddit threads for 10 of the 16.
What we measured
- Sample: 16 buyer questions we wrote, 15 of the form "best X for a small business" plus one about getting more Google reviews. No question contained a year, "pricing" or "official".
- Date: October 26, 2026, between 14:05 and 14:37 UTC.
- ChatGPT: each question asked once through DataForSEO's ChatGPT LLM Scraper, US, English, web search on. A scraper is not a signed-in user.
- Google: the same 16 questions, US desktop organic top 10, through DataForSEO's Google organic SERP endpoint.
- Pages: every cited page and Google result requested with a Chrome user agent, robots.txt requested for every host, and the cited pages that loaded rendered in headless Chromium.
What we found
- On October 26, 2026, ChatGPT's searches for 15 of 16 buyer questions added "2026" and "pricing", words that none of the 16 questions contained (one run each).
- On October 26, 2026, Google's top 10 had a Reddit thread for 10 of 16 buyer questions, at #1 for 4, while ChatGPT's answers to the same 16 questions cited no Reddit page (one run each).
- On October 26, 2026, 23 Google top-10 results across 16 buyer questions were on paths disallowed for OAI-SearchBot in the robots.txt files we fetched, on five sites: Reddit, PCMag, Facebook, Quora and Amazon. ChatGPT cited none of the five (one run each).
- On October 26, 2026, 2 of Google's 131 top-10 results for 16 buyer questions were also ChatGPT sources for that question, and for 10 of the 16 no domain was shared (one run each).
- On October 26, 2026, of 118 pages ChatGPT cited for 16 buyer questions that we compared, 9 had JSON-LD tags only after JavaScript ran (one run each).
OpenAI says ChatGPT search "rewrites your query into one or more targeted queries" (OpenAI Help Center). Google says it "can read JSON-LD data when it is dynamically injected" (Google Search Central).
What to do
Track ChatGPT apart from Google: ask ChatGPT your buyers' questions, open Sources, note which pages it lists, and repeat monthly. To start with one question, see Does my website show up in ChatGPT?.
Which sources do Google's AI Overviews and ChatGPT cite for "A vs B" software searches?
On October 28, 2026, Google's AI Overviews for 16 "A vs B" software searches cited 49 YouTube videos, at least 1 per search, and ChatGPT cited none. ChatGPT's sources for the same 16 questions were mostly the two compared vendors' own sites: 87 of 122 source entries.
What we measured
- Sample: 16 "A vs B" software pairs and 16 "[brand] pricing" searches, hand-picked.
- Date: October 28, 2026, between 14:03 and 14:25 UTC.
- Google: DataForSEO's Google organic SERP endpoint, US, English, desktop, one scraper run per search. AI Overview citations came from the same response. The 16 "A vs B" searches returned 6 to 9 organic results each; we counted the results returned, not a full page one.
- ChatGPT: the 16 "A vs B" questions only, each asked once through DataForSEO's ChatGPT LLM Scraper, US, English, web search on. A scraper is not a signed-in user. The pricing searches were run on Google only.
- Ownership: each URL classed by domain as vendor A, vendor B or neither.
What we found
- On October 28, 2026, Google's AI Overviews for 16 "A vs B" software searches cited 49 YouTube videos, at least 1 per search, while ChatGPT cited none for the same 16 questions (one scraper run each, US).
- On October 28, 2026, 87 of the 122 source entries ChatGPT listed for 16 "A vs B" software questions were on the two vendors' own sites; of Google's 121 returned organic results for the same searches, 7 were (one scraper run each, US).
- On October 28, 2026, for 3 of 16 "[brand] pricing" Google searches, none of the 7-9 organic results returned was on the brand's own site (one scraper run each, US).
- On October 28, 2026, for 3 of 16 "[brand] pricing" Google searches set to US English, an organic result on the brand's site had /es/ or /es-la/ in its pricing URL and a Spanish title and snippet (one scraper run each).
What to do
Ask ChatGPT "[your product] vs [your rival]", open Sources, and read every page of yours it lists, then search the same pair on Google and count the videos in the AI Overview's sources.
What do software sites' robots.txt files tell AI bots?
On October 29, 2026, 44 of 50 hand-picked software sites served us a robots.txt, and 10 of those named any of 8 AI bots we picked. In 4, a group for one of those bots had "Allow: /" and no Disallow, while the "*" group had 15-105 distinct Disallow paths.
What we measured
- Sample: 50 software sites, hand-picked before any fetch, not random. 44 served a robots.txt and 42 served a homepage; the other requests failed and were not counted.
- Date: October 29, 2026, one run, 14:02 to 14:06 UTC.
- Fetch: from our server in Ashburn, Virginia, United States, with a Chrome desktop user agent, redirects followed. A site may serve real crawlers a different file.
- The 8 AI bots we picked: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User.
- Parsing: robots.txt groups split as RFC 9309 describes; homepage descriptions read from raw HTML with no JavaScript run.
What we found
- On October 29, 2026, 44 of 50 hand-picked software sites served us a robots.txt, and 10 named any of 8 AI bots we picked (one run).
- On October 29, 2026, in 4 of 44 robots.txt files served by hand-picked software sites, a group for one named AI bot had "Allow: /" and no Disallow, while the "*" group had 15-105 distinct Disallow paths (one run).
- On October 29, 2026, of 44 robots.txt files served by hand-picked software sites, 8 named ClaudeBot, 4 named Claude-SearchBot and 3 named Claude-User (one run).
- On October 29, 2026, 7 of 44 robots.txt files served by hand-picked software sites had a Content-Signal line: 5 said ai-train=yes and 2 said ai-train=no (one run).
- On October 29, 2026, 3 of 44 robots.txt files served by hand-picked software sites had Disallows for a user-requested fetcher (ChatGPT-User, Claude-User or Perplexity-User) (one run).
- On October 29, 2026, 42 of 50 hand-picked software sites served us their homepage as raw HTML; 16 had both a meta description and an Organization markup description, and ignoring case and punctuation, 4 matched and 12 differed (one run).
Anthropic lists "three robots": ClaudeBot for training, Claude-SearchBot for search and Claude-User for user requests (Anthropic). On ChatGPT-User, OpenAI says "robots.txt rules may not apply" (OpenAI). Content-Signal is a stated preference; we did not test which bots follow it.
What to do
Before you give an AI bot its own group in robots.txt, read that vendor's robots.txt documentation and name each bot you mean. To see which AI crawlers your file allows today, run the AI crawler robots.txt checker.
How were these studies run?
Each study ran once, on one day, with queries or sites we picked by hand before collecting any data.
- October 26, 2026: 16 buyer questions, ChatGPT and Google, one run each, US, through DataForSEO.
- October 28, 2026: 16 "A vs B" searches on Google and ChatGPT and 16 "[brand] pricing" searches on Google, one run each, US, through DataForSEO.
- October 29, 2026: robots.txt and homepage of 50 hand-picked software sites, one run, from a server in Ashburn, Virginia, United States.
- ChatGPT was reached through DataForSEO's ChatGPT LLM Scraper with web search on, not as a signed-in user.
- Every count was recounted from the saved raw responses.
What can't these numbers tell you?
These are counts from one run, not rates. They do not show how often any engine behaves this way.
- One run per question or site, on one day, and not re-measured since. ChatGPT answers change between runs.
- One engine per comparison: ChatGPT and Google, nothing else.
- Hand-picked questions, pairs and sites in software categories. Not a random sample.
- A scraper, not a signed-in user, and for robots.txt a browser user agent from one server. Real crawlers may get different files.
- We did not test why any page was or was not cited, or which bots follow any robots.txt rule.
Check your own site
Start a free check of your site from the Corpus Plan homepage.
Free preview with no account · free account for the full report · up to 20 pages per audit · 3 free audits per connection per day.