Can I see which sources ChatGPT looked at but didn't cite?

OpenAI exposes two different lists when you use web search through its API. One is the citations: the URLs the model attached to sentences in its answer. The other is the sources field, which OpenAI describes as the complete list of URLs the model consulted. OpenAI's documentation also says the number of sources is often greater than the number of citations.

That gap is the set you're asking about: pages ChatGPT read and then left out of the answer. For background on the two terms, see the definition of retrieved sources and the cited subset.

The ChatGPT app does not show that gap. It shows the links it cites, and nothing else. As of October 2026, no consumer AI surface exposes whether your page was retrieved for a real user's query. So the only way to see retrieved-but-not-cited URLs is to run the query yourself through the API and log both fields. The playbook for logging retrieved vs cited sources walks through that setup.

OpenAI is not alone. Anthropic and Perplexity also return the full retrieval set separately from what they cite, but each counts citations differently.

Two practical notes. Perplexity's Agent API makes web search opt-in, so a request without it returns an answer with no retrieval at all. And because the citation units differ, a "cited" count on one engine is not comparable to a cited count on another. Compare the retrieved-to-cited gap within one engine, not across them.

AI answers pass through four stages: what the model already knows (parametric), whether a crawler can fetch your page (access), whether your page enters the model's context for this query (retrieval), and which context pages get used and cited (generation). Most visibility problems look identical from the outside: you're not in the answer. The sources field tells you which stage failed.

• Not retrieved at all. The problem is likely upstream. Check whether robots.txt allows OAI-SearchBot, whether your CDN or firewall lets it through, and whether key content exists in the HTML before JavaScript runs. These are the levers with the strongest evidence, and they're in your control.
• Retrieved but not cited. Access worked for that query. The model had your page and chose other sources. That shifts the question from "can it reach me" to "what did it prefer", which is the subject of why crawlers visit but AI never cites you.
• Retrieved and cited. You're in. Check that what the answer says about you is correct.

This split also stops you wasting work. If you're already being retrieved, another robots.txt review won't change the outcome. If you're never retrieved, rewriting copy won't either.

Retrieved-not-cited is a signal, not proof that your page was ignored. Four limits apply.

Citation is not the same as use

Mechanistic research on how models cite (arXiv 2606.28358) repaired over 90% of missed citations by amplifying specific internal components, without changing the sources. Citation emission is partly separate from whether a source shaped the answer. Your page may have influenced an answer that never named you, a case covered in whether AI can use your content without citing you.

Position and relevance drive selection

C-SEO Bench (NeurIPS 2025) found a document's position in the model's context mattered far more than content rewrites, and most rewriting methods were ineffective or harmful. A 252,000-trial factorial study (arXiv 2605.25517) found topical relevance and list position were the biggest drivers of being cited first. You can't set your position in ChatGPT's context.

The API is not the app

ChatGPT search decides per query whether to search at all, and it rewrites your prompt into one or more targeted queries, according to OpenAI's help page. With memory on, saved memories can shape that rewrite. Your API test is a controlled sample, not a replay of a user session.

No published baseline

Nobody has published a typical ratio of retrieved to cited URLs. Any number you see is your own measurement, not a benchmark.

1. Pick a small set of buyer questions and run each through the API with web search on. Log both the sources and citations fields, with the date.
2. Sort each result into one of the three outcomes above: not retrieved, retrieved only, or cited.
3. For "not retrieved", check access first. Confirm OAI-SearchBot is allowed in robots.txt, then confirm your firewall doesn't return a 403. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
4. For "retrieved only", look at who was cited instead. If engines cite third-party pages, being named on those pages is associated with appearing in the answer. Also check that your page states specific facts (a price, a number, a one-line definition) in fetchable HTML, which gives the engine a sentence it can quote.
5. Don't chase citations with rewrites. The evidence on citation-oriented rewriting runs from null to negative.
6. Repeat on a schedule. Retrieval varies by query and over time, so one run is a snapshot.

For where ChatGPT's cited pages tend to come from, see where ChatGPT citations come from.

Does the ChatGPT app show sources it read but didn't cite? No. As of October 2026, the ChatGPT app shows the links it cites. It does not expose the wider set of pages it consulted. That list only appears in OpenAI's API, in the sources field returned by the web search tool. If you want to see retrieved-but-not-cited pages, you have to run the query through the API and log the result yourself.

Does OpenAI offer a report of how often ChatGPT retrieves my site? Not as a publisher dashboard. OpenAI's documented ranking disclosure says ChatGPT ranks results using multiple factors and placement is not guaranteed. The sources field only covers queries you run yourself through the API. For what OpenAI does and doesn't report, see [whether OpenAI offers a citation report](/questions/openai-chatgpt-citation-report).

If my page is retrieved but not cited, should I rewrite it? Not as a citation tactic. C-SEO Bench found most conversational SEO rewriting methods were largely ineffective and often harmful. What has some support is stating specific facts plainly in fetchable HTML, so the engine has a sentence it can quote. Beyond that, check which third-party pages were cited instead, since being named there is associated with appearing in the answer.

Does blocking GPTBot affect whether ChatGPT retrieves my page? No. OpenAI states its crawler settings are independent. GPTBot governs training use. OAI-SearchBot governs whether your site can appear in ChatGPT search answers. If your page never shows up in the sources field, check OAI-SearchBot permission in robots.txt and your CDN or firewall, not GPTBot. ChatGPT-User handles user-initiated fetches and robots.txt may not apply to it.

Can I compare retrieved-to-cited gaps between ChatGPT and Perplexity? Only loosely. Each engine counts citations differently: Anthropic cites short text spans, Perplexity cites whole sources by index, and OpenAI attaches URLs to answer text. A gap on one engine has a different meaning on another. Track the gap within each engine over time, and read cross-engine differences as different systems behaving differently, not as one score.