ChatGPT SEO: how to rank in ChatGPT's answers.

ChatGPT SEO is access plus quotability plus reputation. Access, because a page OpenAI's search crawler cannot fetch cannot be cited by ChatGPT Search — that one is documented. Quotability, because the engine lifts passages rather than ranking pages. Reputation, because most of what it cites about a brand sits on somebody else's domain. Working order: open the crawler gate, get indexed fast, publish pages that answer one buyer question directly, state your facts in structured data, earn the third-party record, then re-measure on a cadence — because the answers drift from one run to the next.

The gate: which crawler actually decides this

Before any content tactic, one rule is published rather than inferred. OpenAI runs separate crawlers for separate jobs, and only one of them governs whether you can appear in a live answer:

• OAI-SearchBot supports ChatGPT Search — the surface that answers with links. If your robots.txt, CDN bot rules, or firewall block it, you are out of that surface. This is the one to keep open.
• GPTBot is primarily training-data collection. Blocking it is a legitimate choice about whether your content trains future models, and it does not remove you from ChatGPT Search.
• The same split exists at Anthropic — Claude-SearchBot for search appearance, ClaudeBot for training, Claude-User for pages a person asked about directly.

The reason this leads the playbook: the blocking is usually accidental. CDN bot protection ships with AI crawlers denied by default, and copied-around robots.txt snippets tend to disallow every AI user-agent they have heard of. The result is a business that objected to training and quietly removed itself from search. Check your rules bot by bot, and if you are invisible without knowing why, the six-reason diagnostic starts with exactly this.

The overlap: where its citations actually come from

With the gate open, the next question is where ChatGPT's citations concentrate. The most useful measurement anyone has published is a count taken from outside the system:

Be precise about what that licenses. OpenAI documents OAI-SearchBot for ChatGPT Search and does not publish which index or indexes sit behind it; ChatGPT Search may work with more than one search provider. So “get into Bing and you get into ChatGPT” is a claim nobody can support. What the number does support is a strong, cheap proxy: if Bing has never indexed your page, your page sits outside the population ChatGPT's citations are drawn from, and you would be relying on the far thinner Google overlap.

It also explains a statistic that otherwise reads as a mystery: an SEO firm's own 2026 visibility audit (AI Search Engineers) found 77% of businesses ranking on Google page 1 invisible in ChatGPT. Read that as a vendor's dataset rather than a census — but the direction squares with the citation counts. Google ranking is not the same asset as AI citation. Three moves close the gap:

1. Confirm you exist in Bing. Search site:yourdomain.com on Bing itself and check whether the pages you would want quoted — service pages, comparison pages, FAQ pages — are in the index at all, not just your homepage.
2. Claim the console nobody claims. Verify your site in Bing Webmaster Tools, submit your sitemap, and use its index inspection to see what was kept versus skipped. It is free, and it reports on a half of your visibility Google Search Console cannot see. The full walkthrough lives in the Bing and Copilot guide.
3. Stop waiting on crawls.IndexNow, the open submission protocol Bing accepts, lets you tell the index a URL is new or changed within minutes of publishing — you host a small key file on your domain, then ping each URL as it ships. Submission is immediate; whether and when to index remains Bing's call.

The playbook: three jobs, in working order

Access gets you eligible. These three jobs decide whether the engine uses you. A typical live answer shows only a handful of clickable citations — vendor analyses by AirOps and ZipTie put it around three to six — so every job below is about surviving a hard cut.

1. Shape pages like the answers they should become

ChatGPT does not rank your page; it quotes it, or it does not. Pages win quotes by answering one specific buyer question directly, in the first paragraph, under a heading phrased the way buyers ask. Comparison pages, honest pricing pages, and FAQ pages get quoted constantly because their structure matches the question. A homepage that opens with a slogan gives the engine nothing liftable. The test for any page: could a stranger paste one paragraph of it into a chat as a complete answer? If not, rewrite the opening until they could.

2. State the facts in structured data

Structured data is how you stop machines from guessing. Mark up your organization or local business with the correct specific subtype, your service area, offerings, and FAQ content — and keep the markup identical to what the visible page says, because mismatches read as spam. Be clear on what this does and does not do: no engine publishes markup as a requirement for AI citation, and Google states plainly that no special markup is needed to appear in its AI surfaces. What schema buys you is unambiguity — your name, category, prices, hours, and services arrive machine-readable instead of inferred — plus the classic rich results it has always driven. The full AEO playbook covers which types matter most by business shape.

3. Earn mentions where the engine actually looks

Your own site is the minority of the story. Octolens, a citation tracker, analyzed its own dataset and found roughly 91% of the sources AI engines cite about a brand are third-party pages — review platforms, directories, comparison articles, and community threads, Reddit above all. One vendor's dataset, so hold the exact figure loosely; the shape of it matches what anyone sees reading answers by hand. This is the slow half of the playbook: reviews accumulate, listings propagate, and community reputation builds over months, not sprints. It also feeds the memory pipeline — the third-party record is what the next model version trains on. Start it now precisely because it cannot be rushed later.

The drift: your rank is a rerun, not a ladder

Classic SEO trained everyone to think in positions: you climb to #3, you hold #3. ChatGPT does not work that way. Ask the same buying question twice and the answer regenerates — different wording, sometimes different names, even minutes apart. Underneath the surface variance, the sources themselves churn:

The practical consequences for how you measure:

• Fix the prompt set. Write down the 10–50 buyer questions that matter and re-ask the same set, in the same phrasing, on a schedule. Changing the questions each check makes the data unreadable.
• Count appearances, not positions.The useful metrics are rates across runs — how often you are mentioned, how often a source about you is cited, how often you are the first name — not where you “rank” in any single answer.
• Give changes weeks, then judge. In Semrush-reported practitioner testing, newly indexed pages that earned a ChatGPT citation took a median of ~6.8 days to get the first one, and about 42% of tested pages were cited within 30 days — an observed pattern, not a promise, since many pages are never cited at all. Check before that window and you will conclude nothing worked; check once and never again and you will miss the churn taking it back.

This is also the point where people start shopping for software, so one comparison note: most AI-visibility tools stop at the number. If you are weighing that decision, our side-by-side with Profound lays out which parts of this playbook a tracker covers and which parts stay on your desk either way.

One more reason not to over-index on any single engine's answers: they disagree with each other. The same question put to Perplexity or Google's AI Overviews routinely returns a different set of names, because each runs on a different retrieval stack. Measuring ChatGPT alone tells you about ChatGPT alone.

Disclosure: the product behind this playbook

AEO Action builds the product that runs this exact playbook as software: it scans ChatGPT through the live consumer interface daily — the same surface your buyers use, not a weekly API poll — alongside up to five other engines, and it audits your robots.txt crawler by crawler so a blocked OAI-SearchBot shows up as a finding instead of a mystery. Its Bing Index Engine submits every landing page you approve via IndexNow within minutes, with the honest caveat printed above: submission is ours, the indexing decision is Bing's. And because the drift makes one-time checks unreliable, every approved landing page is registered against its target question and re-checked on each following scan for up to 120 days — so when a page earns its first citation, you get a dated receipt: shipped date A, first cited date B. That whole loop is what we sell as a done-for-you AEO service; the playbook above is true whoever runs it.