Ivan BuildsResearch
Original research · ChatGPT · Indexing

Google Didn't Show 18% of the Pages ChatGPT Cites, Against 6% of the Pages Its Own AI Overviews Cite (Study of 2,232 Pages)

TL;DRIf you track where ChatGPT gets its sources, my bet is: don't assume Google has them. 18% of the pages ChatGPT cited didn't show up in any of up to four Google searches for the page, against 6% of the pages Google's AI Overviews cited, and 33% of the pages ChatGPT cited for local questions. And a single site: search is no index check: it missed 24% of the pages Google's AI Overviews were citing.

I asked ChatGPT and Google the same 300 questions three times, then searched Google for every page ChatGPT cited, up to four times each, with the pages Google's own AI Overviews cited as the control.

IS
By Ivan Smiljkovic
Published 30 September 2026 · 900 ChatGPT answers and 900 Google searches, 29 September 2026 · ChatGPT and Google AI Overviews · 12 min read
Start with the short versionFindingsMethodCite
18%
of the pages ChatGPT cited didn't show up in any Google search for the page
6%
of the pages Google's AI Overviews cited didn't show up in the same searches: the method's own miss rate
48%
of ChatGPT's answers cited at least one page that no Google search found
24%
of the pages AI Overviews cited were missed by a single site: search

How I checked itA blind second reading of 80 pages no search found, by an AI model that never saw my answers, gave the same reason as my rules for 77 of them; every figure was computed in Python and again in SQL.

On this page
  1. The short version
  2. Earlier studies
  3. Also found
  4. How I did it
  5. 0118% of the pages ChatGPT cited didn't show up in any Google search for them, against 6% of the pages AI Overviews cited
  6. 02A single site: search missed 24% of the pages Google's own AI Overviews had just cited
  7. 0348% of ChatGPT's answers cited at least one page that no Google search found
  8. 04For local questions, 33% of the pages ChatGPT cited didn't show up in Google, against 15% for AI Overviews
  9. 05Most pages Google didn't show were live and on sites Google knows: only nine of 412 were on a site it showed nothing of
  10. 06ChatGPT cited the same page for the same question in every run for 9% of question and page pairs, against 29% for AI Overviews
  11. What I would test
  12. Limits
  13. Data
  14. Cite this study
  15. About the author

Ask ChatGPT for the "best dentists in Los Angeles, CA" and it cites a Yably directory page: yably.com/dentists-dental-clinics/los-angeles. I searched Google for that page four ways, from its address with site: to its full URL, and none of the searches returned it. Google knows the site: a site: search for Yably returns its other city pages. It just doesn't show this one. I wanted to know how common that is. Of the pages ChatGPT cites, how many does Google's own search not return, why, and how does that compare with the pages Google's AI Overviews cite for the same questions?

The short version

Google's search didn't return 18% of the pages ChatGPT cited (412 of 2,232), against 6% of the pages Google's AI Overviews cited for the same questions (29 of 500), after up to four searches for each page.

A snapshot shows what ChatGPT cited and what Google's search returned, not what makes a page cited, so the points below are bets to test, not rules. Under each one are the numbers behind it and the finding they come from. The full version, with the strongest objection to each bet, is in What I would test.

  1. BetTrack ChatGPT's sources on their own, not through Google.Evidence in Finding 1 and Finding 3:
    • 18% of the pages ChatGPT cited didn't show up in any Google search for them; 6% of the pages Google's AI Overviews cited didn't either.
    • 48% of ChatGPT's answers cited at least one page no Google search found.
  2. BetFor local businesses, look at the directory pages ChatGPT cites, not only the ones Google shows.Evidence in Finding 4:
    • For local service questions, 33% of the pages ChatGPT cited didn't show up in Google, against 15% for AI Overviews.
    • None of the 18 Yably pages ChatGPT cited showed up, and 19 of the 21 Best Pros in Town pages didn't.
  3. BetDon't read a single missing site: result as "not indexed".Evidence in Finding 2:
    • A single site: search missed 24% of the pages Google's own AI Overviews had just cited.
    • Each later search found more: after all of them, 6% of those pages were still missing.

What earlier studies found

  • Louise Linehan at Ahrefs found that 28.3% of the 1,000 pages ChatGPT cited most in September 2025 had no organic keywords in Ahrefs (Ahrefs). That measures whether a page ranks for anything, not whether Google has it.
  • Tomek Rudzki at Peec AI found, in ChatGPT's server-side data from 21 May to 21 July 2026, that it retrieves from its own family of indexes, "Labrador" (Peec AI). The article gives no share of cited pages missing from Google.

What's new here: each page ChatGPT cited was checked against Google itself, with repeated searches, next to the same check on the pages Google's own AI Overviews cited for the same questions, and with the reason each missing page might be missing.

Also found

  • ChatGPT and AI Overviews rarely cited the same pages. 8% of the pages ChatGPT cited were also cited by AI Overviews for the same questions.
  • ChatGPT cited a source in 98% of its answers. Google showed an AI Overview for 86% of the searches.
  • Much of what AI Overviews cited was Google's own. 1,403 of the pages AI Overviews cited were Google Maps listings, other Google pages or YouTube videos, which I counted but did not search; ChatGPT cited three.

How I did it

  • The questions 300 questions drawn at random from my earlier studies, a third each of how-to questions, SEO agency hiring searches and local service questions, each local one with its city's Google location. US English.
  • The answers Each question went to ChatGPT (the consumer app, web search on) and to Google, three times an hour apart, on 29 September 2026: 900 ChatGPT answers and 777 AI Overviews.
  • The pages Every page ChatGPT cited, and a random 500 of the pages AI Overviews cited as the control. Each was read once to see why Google might not have it: its status, where it redirects, its canonical, a noindex tag, robots.txt as Google reads it, a bot check, its publish date.
  • The searches For each page, a Google search for its address with site:. For each page that search missed, the full URL as a plain search; then the URL without https; then site: with the exact host. Each search went only to pages the ones before missed, about an hour apart, on 29 and 30 September 2026. A page counts as found when its own address, the address it names as canonical, or the address its redirects end on is among the results.
  • The sites For each page no search found, a site: search for its site: does Google show any page of it?

Finding 0118% of the pages ChatGPT cited didn't show up in any Google search for them, against 6% of the pages AI Overviews cited

412 of 2,232 pages ChatGPT cited didn't show up in any of up to four Google searches for the page (15% to 23%, allowing for chance). The same searches missed 29 of 500 of the pages Google's own AI Overviews cited for the same questions (2% to 13%, allowing for chance). AI Overviews cite pages from Google's own index, so their figure is what the method misses on pages Google certainly has.

  1. Pages ChatGPT citedn = 2,23218%
  2. Pages Google's AI Overviews citedn = 5006%
Share of cited pages that no Google search for the page returned.

Counted by citation link instead of by page, 18% of ChatGPT's citation links pointed to a page no search found, against 4% of AI Overview citation links. Pages cited only by ChatGPT didn't show up 20% of the time; the 186 pages that both ChatGPT and AI Overviews cited didn't show up 3% of the time.

Part of the gap is the method's own. Addresses with a query string (a Play Store app, a help page ending in ?hl=en) are hard to find with site:, and some pages showed up at the same address with a different query string. Leaving those out on both sides and counting same-address results as found, 16% of the pages ChatGPT cited still didn't show up, against 3% of the pages AI Overviews cited.

Finding 02A single site: search missed 24% of the pages Google's own AI Overviews had just cited

The first search, site: with the page's address, is the check people often use to see whether Google has a page. It missed 24% of the pages AI Overviews cited, pages Google certainly has, and 37% of the pages ChatGPT cited. Each later search found more of them.

  1. AI Overview pages: missed by the site: searchn = 50024%
  2. AI Overview pages: still missed after the URL searchn = 50011%
  3. AI Overview pages: missed by every searchn = 5006%
  4. ChatGPT pages: missed by the site: searchn = 2,23237%
  5. ChatGPT pages: still missed after the URL searchn = 2,23226%
  6. ChatGPT pages: missed by every searchn = 2,23218%
Share of cited pages still missing after each search, for the pages AI Overviews cited and the pages ChatGPT cited.

For the pages AI Overviews cited, the full URL as a plain search found 13% of all of them, the URL without https 3% and site: with the exact host 2%. In my pilot the same site: search, run again an hour later, came back with a different answer for some pages. A page that a single site: search doesn't return may still be in Google.

Finding 0348% of ChatGPT's answers cited at least one page that no Google search found

Of 900 ChatGPT answers, 429 cited at least one page no search found. The rest cited only pages Google returned, or no page at all. Some of those answers would carry a page the method missed: on the control, the pages AI Overviews cited, the searches missed 6%.

Finding 04For local questions, 33% of the pages ChatGPT cited didn't show up in Google, against 15% for AI Overviews

Local service questions ("best dentists in Los Angeles, CA") had the largest gap. For how-to questions it was 15% against 5%, and for SEO agency questions 13% against 1%, in each case ChatGPT's pages against the pages AI Overviews cited.

  1. Local services, ChatGPTn = 54533%
  2. Local services, AI Overviewsn = 13615%
  3. How-to, ChatGPTn = 50815%
  4. How-to, AI Overviewsn = 1245%
  5. SEO agencies, ChatGPTn = 1,17913%
  6. SEO agencies, AI Overviewsn = 2401%
Share of cited pages that no Google search returned, by question type and answer engine.

The local pages ChatGPT cited that Google didn't show were often directory pages. None of the 18 Yably pages ChatGPT cited showed up, nor 19 of the 21 Best Pros in Town pages, 32 of the 41 Birdeye review pages and 13 of the 14 American Dental Association "find a dentist" search pages. The big directories mostly did show up: only eight of the 35 BBB pages ChatGPT cited were missing.

Finding 05Most pages Google didn't show were live and on sites Google knows: only nine of 412 were on a site it showed nothing of

Of the pages ChatGPT cited that no search found, 71% were live, open to Google (no noindex, not blocked in robots.txt) and not new. 24% refused my reader or showed a bot check, so their reason stays open. Only two carried a noindex tag and three were gone.

  1. Live, open to Google, not new71%
  2. Refused my reader or showed a bot check24%
  3. No answer to my request2%
  4. Published in the week before1%
  5. Gone (removed or not found)1%
  6. Server error1%
  7. Marked noindex0%
The first reason Google might not have the page, for the pages ChatGPT cited that no search found. Base: 412 cited pages.

For all but nine of those pages, a site: search for the site returned other pages of it. Google knew the site and didn't show the page.

Reading the blocked pages again only strengthens this. I read every page no search found, from either answer engine, again in a real browser, and the ones still blocked once more from a US address, with as many found pages drawn at random as a comparison. After all the reads, 379 of the 439 missing pages were live, open to Google and not new, against 391 of the 439 found ones. On the page itself, the missing pages look like the found ones.

Could they be the same pages under another address? In the browser, nine of the 439 missing pages named a different address as canonical only once their scripts ran, against three of the 439 found ones. Such pages exist, like Microsoft support articles cited in their Australian or British versions, but they are a small part of the gap.

Finding 06ChatGPT cited the same page for the same question in every run for 9% of question and page pairs, against 29% for AI Overviews

ChatGPT's sources changed from run to run more than AI Overviews' did. 72% of the question and page pairs ChatGPT cited came up in a single run. The pages no search found came back in every run about as often as the pages Google showed: 11% of their pairs against 9%.

  1. In one run only72%
  2. In two runs18%
  3. In all three runs9%
How many of the three runs cited each question and page pair, for ChatGPT. Base: 2,674 pairs.

What I would test

A snapshot shows what ChatGPT cited and what Google returned, not what makes a page cited, so each point below is something to test rather than a rule.

Test whether your own pages that ChatGPT cites are in Google. 18% of the pages ChatGPT cited didn't show up in any Google search for them. For your own site, Search Console's URL inspection says whether Google has the page, which a public search can't. The bet is that some pages earning ChatGPT citations are pages Google doesn't show, and the check would say which.

Test the directory pages behind local recommendations. For local questions, 33% of the pages ChatGPT cited didn't show up in Google. The bet is that a local business listed on directories like Yably or Birdeye is visible to ChatGPT whether or not Google shows the listing; checking which directory pages ChatGPT cites for your city and trade, and whether Google shows them, would show it.

Test before trusting a missing site: result. A single site: search missed 24% of the pages Google's own AI Overviews cited. The bet is that repeating the search, and searching the full URL, turns up pages a single site: search misses; searching the same pages both ways would show how often.

What this study does not show

A search is not Search Console. Google may have a page and not show it to any search; my searches missed 6% of the pages Google's own AI Overviews cited, so read ChatGPT's figure next to that, not alone. "Didn't show up" here means that none of up to four searches returned the page at the address ChatGPT cited, the address it names as canonical, or the address its redirects end on.

One day of questions, US English, and the kinds of questions my earlier studies used, not a sample of what people ask. ChatGPT through the consumer app with web search on, collected through a third-party service; AI Overviews as a US desktop search shows them. The three runs measure run-to-run change through the same collection path.

The searches ran on 29 and 30 September 2026. The first search for each page ran within minutes of the answers, and each later one about an hour after the one before, except once: one stage ran about ten hours after the one before it, when the run paused overnight. A page Google added in between would count as found.

Not every page could be read, so some reasons stay open. My reader ran from Europe, and 24% of the missing pages ChatGPT cited refused it or showed a bot check. The reasons in Finding 5 are from that first read, as planned; the read in a browser and the last one from a US address are extra: the browser read 33 of the blocked missing pages, and the US address 55 more.

Some of this was decided after the data came in, and each step is logged in the study's plan: the robustness figure in Finding 1, the later reads, and the fixes the blind check found in my page reader (a noindex header written in an odd form, and a "checked as of" date taken for a publish date). The fixes moved a few pages between the reasons in Finding 5 and changed no found or missing figure. The blind check agreed with my reasons on 77 of 80 pages before the fixes.

YouTube videos and Google's own pages, such as Maps listings, were counted but not searched, on both sides: a site: search can't pick out one video by its address.

Nothing here is causal. Being missing from Google's search is an observation about a page, not a judgement of it, and this study can't show whether a page's absence from Google makes ChatGPT more or less likely to cite it.

Data

900 ChatGPT questions and 900 Google searches on 29 September 2026. ChatGPT's 900 answers made 3,664 citation links to 2,235 pages on 925 sites; 2,232 of those pages were searched, the rest being YouTube videos or Google's own pages. AI Overviews cited 3,478 pages, of which 500 drawn at random were searched. Every figure has a numerator and a denominator in the study's metrics file, every search is kept as the raw response and every page read with its HTML. Running it cost $13.24.

Cite this study

Ivan Smiljkovic (2026). Google Didn't Show 18% of the Pages ChatGPT Cites, Against 6% of the Pages Its Own AI Overviews Cite (Study of 2,232 Pages). Ivan Builds. https://ivanbuilds.com/chatgpt-pages-not-in-google.html

Every section has its own link: hover a heading and use the # to copy it.

About the author

IS
Ivan Smiljkovic
Independent research on search and AI answers

I study how Google and AI assistants choose what to cite, using public data I collect myself, and I publish the method and limits with every study.

More research

More research

Original research
83% of Google AI Overviews for How-To Questions Cite a YouTube Video (Study of 1,777 AI Overviews)
I searched Google with 600 real how-to questions, three times each, and traced every YouTube video its AI Overviews cited, down to the second they linked to. Then I compared the cited moment with the rest of the same video, the cited videos with the videos YouTube itself ranks for the same question, and the line at the cited second with the lines around it.
Original research
66% of the Pages ChatGPT Cites When Recommending SEO Agencies Are Written by SEO Agencies (Study of 3,000 AI Searches)
I asked ChatGPT and Google the same 300 SEO agency-hiring questions five times, then traced every page they cited and every agency they named. Most of the evidence is written by SEO agencies.
Original research
82% of the Pages ChatGPT Cites When Recommending a Local Business Are Directories and Review Sites (Study of 2,112 AI Searches)
I asked ChatGPT and Google for local roofers, plumbers, HVAC contractors, pest control companies, electricians, dentists, auto repair shops and landscapers in 22 US cities, three runs an hour apart, and traced every page they cited and every business they recommended. On ChatGPT, the cited pages are mostly directories and review sites, and the businesses' own websites barely appear. Google AI Overviews lean the other way.
Original research
64% of the Passages Google's AI Overviews Quote Start at a Heading or Right Below One, Against 31% of Other Passages on the Same Page (Study of 3,224 Quotes)
I searched Google with 900 questions, read the web pages its AI Overviews cited, and found where on each page the source card's words sit. Then I compared each quoted passage with up to twelve other passages of the same length on the same page.