Why did ChatGPT read your page but not cite it?

Published 9 November 2026. Updated 9 November 2026.

Most of the time ChatGPT never opened your page, only its title and a short snippet of it.

In Resoneo's study of 1,249 ChatGPT answers from July and August 2026, the model opened 760 of 58,000 retrieved URLs. It cited 74% of the pages it opened and 7% of the pages that only came back in its search results. Our guide to AI citations covers how ChatGPT, Claude and Grok cite in general. This page is for the site owner who found their URL among ChatGPT's sources but not in the answer: what tests and our own recordings point to, how to tell which reason applies, and what to change.

Did ChatGPT actually read your page?

Probably not. Resoneo traced ChatGPT's retrieval with free, paid and logged-out accounts in several countries, and most URLs dropped out before the model read a word of them.

What happened to the URL	URLs in Resoneo's sample
Returned by ChatGPT's searches	58,000
Promoted to the Sources panel	7,600
Cited in the answer text	5,000
Opened and read by the model	760

Resoneo also found that nine out of ten retrieved pages stay in the "More" section at the bottom of the Sources panel, one click away, and never get cited. These are one team's measurements over two months, so treat the counts as a guide to scale.

A ChatGPT conversation we recorded with Only Fan Out on 14 September 2026 has the same shape on a small scale. It held 78 fetched pages and 15 cited ones, with 7 pages in both lists. So 71 of the 78 fetched pages were left out of the answer, and 8 of the 15 cited pages never appeared in the fetched list at all. It is the same capture our post on ChatGPT's missing fan-out queries uses to show what a saved chat still holds. For how the published gap figures compare, see our page on fetched vs cited sources.

Why does ChatGPT leave out a page it fetched?

Six reasons have evidence behind them, from published tests and from conversations we recorded. None of them describes how OpenAI ranks pages inside the model.

Reason	Evidence	What to check	What to change
It only saw your search result	Resoneo: 74% of opened pages cited, 7% of pages only retrieved	Your URL is in the fetched list and nowhere in the answer	Make your title and the lines under your H1 answer the query on their own
The snippet ran out before the answer	Resoneo: the snippet in ChatGPT's own index is cut just past 200 characters and holds the H1 eight times out of ten	Your H1 and the first 150 or so characters after it	Put the answer straight under the H1
Your page came from a later search	Seer: in every turn that searched, each cited URL was in the first set of results	Which search call brought your page in	Answer the broad first question, as well as the narrow follow-ups
ChatGPT's query named other brands	Suganthan Mohanadasan: brands named in ChatGPT's own query reached the answer 68.9% of the time, brands it only fetched 2.1%	Whether your brand is in the query text	Track mentions and citations as two counts, since page edits do not put your name in the query
Another page matched the query more closely	Ahrefs: cited titles were closer to the fan-out query. AirOps: 20.1% cited when half or more of the query's words were in the title, 9.3% when under a tenth were	Your title next to the query that fetched it	Use the words of the question in your title and H1
It became a product card	Our recordings: ChatGPT shows product cards with price, rating and merchant apart from its cited sources	Whether your shop shows as a card in that turn	Count cards as a result of their own

The first two rows are one problem seen from two sides. When ChatGPT judges your page from its snippet alone, the snippet has to carry the answer. The title figures come from Ahrefs' study of 1.4 million prompts, where cited titles scored 0.602 against the prompt and 0.656 against their closest fan-out query, and uncited titles 0.484 against the prompt, and from AirOps' report on 15,000 prompts, in which 15% of retrieved pages were cited.

How much of your answer fits in ChatGPT's snippet?

Less than most pages assume. On 9 November 2026 we measured our own 13 guides against Resoneo's figure: the H1 and the visible text after it, cut at 200 characters.

Part of the first 200 characters	Our 13 guides
H1	22 to 72 characters
Published and updated line	53 to 55 characters
Room left for the answer	73 to 121 characters
Whole first sentence fits	5 of 13
Would fit without the date line	9 of 13

The line that tells readers when a page was updated used about a quarter of the window on every page. On eight of our guides the opening sentence ran past the cut, so a model reading only the snippet would see half an answer. The first sentence of this page is 90 characters long so that it ends inside the window. The measurement assumes ChatGPT's index cuts a page the way Resoneo measured it, and it says nothing about Claude or Grok.

Why are pages from ChatGPT's later searches cited less often?

Because ChatGPT seems to pick its candidates in the first round. In Seer's test of ChatGPT's search tool, every cited URL across eight turns that searched appeared in the first set of results, and none of the results that showed up only later was cited. Seer ran nine prompts on the free plan in one sitting, a small sample, as the author says.

Our own recordings show where the later searches come from. In a ChatGPT turn we recorded on 9 September 2026, ChatGPT ran 17 queries in three search calls: 7 business, 7 web and 3 image. Its second call put three business names in quotes, names its first call had turned up. A page that only matches those follow-up queries arrives after ChatGPT has already chosen what to look for.

Suganthan Mohanadasan found the same order with brands. In 21 of 27 conversations on one ChatGPT Plus account, the first search already named brands the user never typed. He calls the percentages directional, since they come from one account.

How do you find out which reason applies to your page?

Record the turn as it happens and read your page's place in it. Since 14 September 2026 ChatGPT sends the query text only while the answer streams, so a reloaded chat keeps the fetched and cited pages but not the queries that found them.

1. Write down the question your buyers ask, in their words.
2. Open ChatGPT with the Only Fan Out side panel and add your domain and brand to its watchlist. It highlights every mention across queries, fetched sources, cited sources and answers.
3. Send the question as normal. Find your URL in the fetched list and note which query and which search call brought it in.
4. Check whether your brand is in that query, and put your title next to the query's wording.
5. Read your H1 and the 150 or so characters after it. If the answer is not there, fix that first.
6. Open the page ChatGPT cited in your place and compare its opening with yours.
7. Export the turn as CSV and ask the same question a week later, since answers change from run to run.

What does not explain a missing citation?

Two common suspects do little for a page that ChatGPT already fetched. GPTBot is OpenAI's training crawler, and Resoneo found it feeds neither ChatGPT's search index nor the copies of pages the model reads. OpenAI's publisher FAQ names OAI-SearchBot as the crawler that must not be blocked for content to appear in ChatGPT.

Structured data is the second. Resoneo found JSON-LD is stripped from a page before the model reads it, and Chen and Koudas found that FAQ blocks and structured data made a difference between websites, not between pages of the same site. If your page never shows up in the fetched list at all, the problem sits earlier: check that OAI-SearchBot can crawl it, then check whether you rank for the queries ChatGPT writes.

Related questions