How to get cited by ChatGPT comes down to one uncomfortable fact: it reads far more pages than it quotes. Across the prompts we track, ChatGPT's search mode fetches roughly six or seven candidate pages for every one it names in an answer. Getting found is table stakes. Getting quoted — landing in the citation card the user actually clicks — is the real contest, and most guides skip it.
This breakdown comes from our own citation logs at MaxAEO, where we watch which URLs ChatGPT fetches versus which it credits, day after day, across ChatGPT, Gemini, Perplexity, Copilot, and Grok. ChatGPT is the engine most teams optimize for and the one wrapped in the most guesswork. Below is what its source selection actually rewards — its own index, Bing, freshness — and the specific moves that flip a page from retrieved to cited.
What "getting cited by ChatGPT" actually means
Getting cited by ChatGPT means your URL appears as a named, clickable source in an answer generated with web browsing on — a live citation, not a brand mention reconstructed from training data. These are two different systems. In its default mode, ChatGPT writes from memory, and any "source" it names may be pattern-reconstructed rather than fetched. When search fires, it retrieves live pages and attaches real citations.
That distinction matters because you can only earn the second kind. Training-data mentions lag your overall web footprint by months; live citations are won page by page, prompt by prompt. This is the core of answer engine optimization: shaping content so a retrieval system can lift a clean, correct passage and attribute it to you. Everything below targets the live-search path — the surface you can measure and move.
How ChatGPT chooses sources: the three gates
ChatGPT's source selection runs as three sequential gates — retrieval, selection, and attribution — and a page has to clear all three to appear as a citation. Most content clears the first and dies at the second. Treat them as separate problems with separate fixes.
| Gate | The question ChatGPT is answering | What you actually control |
|---|---|---|
| Retrieval | Can I find and read this page right now? | Crawler access, presence in Bing + OpenAI's index, render-free text, freshness |
| Selection | Is this the cleanest source to quote for this claim? | Answer-first passages, extractable structure, on-page numbers |
| Attribution | Whose URL goes on the citation card? | Third-party consensus, entity clarity, being the origin of the fact |
The reason so many well-optimized pages never get cited is that classic SEO only trains you for the retrieval gate. Ranking in a search index gets you into the candidate pool. It says nothing about whether your prose is quotable or whether ChatGPT decides a competitor is the safer name to attach. The next sections take the gates one at a time.
The retrieval–citation gap: why being found isn't being quoted
In our citation logs, ChatGPT quotes only a small fraction of what it fetches — roughly one in six to seven retrieved pages earns a citation. The other ~85% are read and discarded. That gap is where your work lives.
We watched a B2B observability vendor get fetched constantly for prompts like "best observability tools for Kubernetes" — ChatGPT pulled their pricing page and docs — yet the citation card went to a third-party roundup and a Reddit thread. Retrieval was never their problem. Selectability was. Their pages answered slowly, hid key numbers behind JavaScript, and stated claims no other domain repeated.
The takeaway: if you're already indexed and still not cited, stop chasing more crawl coverage. The bottleneck is almost always gate two or gate three. Measuring the two failure modes separately — fetched-but-not-cited versus never-fetched — is the fastest way to know which problem you actually have.
Where ChatGPT gets its sources: its own index, Bing, and live scrapers
ChatGPT search doesn't pull from one index. It blends OpenAI's own crawl, Bing's results, and commercial live-scraping vendors, then routes different query types down different pipes. Public teardowns of ChatGPT's search traffic have surfaced distinct source paths — a Bing pipe, licensed-publisher feeds, and third-party scrapers leaned on heavily for shopping, finance, and local queries.
Three things follow. First, crawler access is non-negotiable: allow OAI-SearchBot (live search) and GPTBot, per OpenAI's crawler documentation, and don't wall off content in your robots.txt by accident. Second, Bing still matters — being absent from Bing's index removes you from a major retrieval path, so verify coverage in Bing Webmaster Tools. We break the current split down further in our analysis of how much of ChatGPT search still runs on Bing.
Third, the pipe that serves your query decides which signals win. A shopping prompt routed through a live scraper rewards fresh, cleanly structured product data; a definitional prompt leans on licensed and authority sources. For the full engine-by-engine map, see which search index powers each AI engine.
What ChatGPT quotes: the traits of a cited passage
Once a page is in the candidate pool, ChatGPT favors passages it can lift verbatim: a direct answer stated early, backed by a specific number, in plain HTML text. Structure beats prose polish. The model isn't reading for elegance — it's scanning for a self-contained sentence that resolves the query cleanly enough to quote.
Here's what our citation logs show separates cited pages from ignored ones:
| Trait of the page | What we observed in ChatGPT citations |
|---|---|
| Answer stated in the first ~60 words of a section | Cited markedly more often than pages that bury the answer |
| An original number, benchmark, or price on the page | Strong lift — ChatGPT prefers the page that states the figure |
| Clean, render-free text | Reliably fetched; JS-gated facts pushed ChatGPT to third parties |
| Content near the top of the page | Openings are quoted far more than deep-page content |
| A claim corroborated on other domains | Far likelier to win the visible citation card |
The pattern to internalize: write the answer, then the argument — not the reverse. Open each section with a definition or verdict a model can excerpt, then support it. One practical corollary: ChatGPT's extraction leans on visible HTML structure — headings, lists, tables — more than on JSON-LD schema, so a clean answer paragraph beats marked-up-but-buried prose. For the page-level signals that win citations across AI answers, see the page signals that get you quoted in AI Overviews.
Answer-first is the single highest-use change
If you fix one thing, make it this. A section that opens "X is …" or "The best option for Y is …" gives ChatGPT a ready-made pull quote. A section that opens with backstory forces the model to synthesize — and when it synthesizes across sources, it often attributes the point to a cleaner-written competitor instead of you.
Who gets the citation: winning attribution over third parties
When several pages support the same fact, ChatGPT tends to cite the source it treats as the origin or the independent verifier — which is why brands so often watch a review site or a Reddit thread get the card for a claim about their own product. Owning the fact isn't enough; the model has to see it confirmed somewhere it didn't come from you.
This is the gate money can't shortcut and PR quietly wins. In the observability example, the fix wasn't a better landing page — it was earning two analyst mentions and a comparison in a credible roundup. Within about three weeks, their own domain started appearing in the citation card for branded prompts, because the claim now recurred across independent domains. That corroboration is what moves you from mentioned to cited.
Two moves compound here: get your facts repeated on sources ChatGPT already trusts — including earning coverage in the news articles AI quotes — and get into the lists it pulls for recommendations, covered in getting into the "best tools" listicles AI quotes. Strong third-party consensus is also the backbone of durable AI reputation management: it shapes how every engine describes you, not just whether it links you.
Freshness: how recency changes what ChatGPT quotes — and when it doesn't
Freshness is a strong citation signal, but only for query types where the answer decays. This is the nuance most guides flatten into "update your content." In our logs, pages refreshed within ~30 days were heavily over-represented in citations for time-sensitive prompts — pricing, "best of" comparisons, "latest," anything with a moving answer. For evergreen definitional prompts, recency barely moved placement.
So don't churn your cornerstone explainers for a date bump — it won't help, and it can hurt if edits degrade the answer. Do keep comparison pages, pricing, and roundups genuinely current, and signal it by updating the visible copy and the numbers, not just a timestamp. If you publish time-sensitive facts, fast indexing matters: a stale index means ChatGPT quotes last quarter's figure from someone else.
One more wrinkle: source selection is cohort-gated. The same URL can arrive through different pipes for different accounts, tiers, and regions. A single manual check tells you what one cohort saw on one day — not the truth about your visibility, which is exactly why spot-checking from your own logged-in account is so misleading.
A get-cited playbook: step by step
To get cited by ChatGPT, work the three gates in order — retrieval first, then selectability, then attribution — because fixing prose is wasted effort if you're not being fetched, and fixing crawl access is wasted if your page can't be quoted. Run this sequence:
- Confirm access. Allow
OAI-SearchBotandGPTBotinrobots.txt; make sure key facts render in server-side HTML, not JavaScript. - Confirm indexation. Verify the page is in Bing's index and submit updates — IndexNow speeds this up.
- Front-load the answer. Open every section with a 40–60 word answer a model can lift verbatim.
- Put a number on the page. State an original stat, benchmark, or price in text — be the source that asserts the fact.
- Structure for extraction. Question-style headings, short paragraphs, one comparison table, clean lists — visible HTML the model can parse without schema.
- Earn corroboration. Get the same claim repeated on independent, trusted domains and into the roundups ChatGPT quotes.
- Refresh what decays. Keep time-sensitive pages current with real edits, not timestamp cosmetics.
- Measure by cohort, over time. Track citations continuously across accounts and regions — never from one manual check.
The order is the point. Teams that jump to step 6 (link building) while failing step 3 (answer-first) spend money earning authority for pages ChatGPT still won't quote.
How to measure whether it's working
You can't improve what you spot-check. Because source selection shifts by cohort, tier, geography, and week, the only reliable read is continuous AI search monitoring across engines and accounts — a manual query is a single sample of a moving system. Track two numbers, not one: how often ChatGPT fetches you and how often it cites you. The gap tells you which gate to fix.
From there, the metric that survives a budget review is AI share of voice — your slice of citations for a prompt set versus named competitors, trended over time. That's what turns "we appeared in ChatGPT once" into a defensible line on a dashboard. It's the job an AI visibility tool does: LLM brand tracking that logs every citation, flags when a competitor takes your slot, and points you to the specific page to fix to get recommended by ChatGPT more often.
The same discipline extends past ChatGPT. Retrieval logic differs by engine — see how Perplexity picks sources, how Claude searches the web, and how citations play out in the second and third reply of a conversation, not just the first answer. Optimize for the mechanics, verify with data, and let the citation logs — not vibes — tell you what's working.
Frequently asked questions
Does ChatGPT use Bing to find sources?
Partly. ChatGPT search blends OpenAI's own index, Bing's results, and third-party live scrapers, routing different query types down different pipes. Bing remains a major retrieval path, so being absent from Bing's index removes you from consideration for many prompts — but Bing rank alone doesn't guarantee a citation.
Why does ChatGPT cite a competitor or a review site instead of my own page?
Because attribution favors corroboration. When a claim about your product also appears on independent, trusted domains, ChatGPT often cites the verifier rather than you. Fix it by earning third-party coverage and roundup mentions so the same fact recurs across sources you don't control.
How long does it take to get cited by ChatGPT after publishing?
It varies by gate. Retrieval can happen within days of indexation, but attribution — winning the citation card over established sources — usually takes weeks. In our tracking, meaningful placement gains for competitive prompts typically showed up around the three-to-four-week mark after both the page and its corroboration were in place.
Can I force ChatGPT to cite my site?
No — you can only make your page the easiest correct choice. There's no submission that guarantees a citation. What works is clearing all three gates: crawlable and indexed, answer-first and extractable, and corroborated on independent domains. Force isn't available; selectability is.
How do I know if ChatGPT is citing my brand?
Through continuous monitoring, not manual checks. Source selection is cohort-gated, so one logged-in query reflects a single cohort on a single day. Track brand mentions in ChatGPT across accounts, regions, and time to see real citation rates, share of voice, and which pages win or lose the slot.