How to rank in ChatGPT: getting recommended by AI
No trick, just an approach: how ChatGPT actually gets its answers, which crawlers matter, and what concretely improves your odds of being named. No guarantees.
The short answer: there is no ranking to climb. ChatGPT either names you or it does not, and that depends on whether you are crawlable for the right bot, whether your content is factual and quotable, and whether your brand already shows up elsewhere on the web. None of that is a guarantee, but these are the dials you actually control.
We already covered what GEO is and the difference between GEO and SEO, and untangled the terms in AEO vs GEO. This time we make it practical: what can you concretely do to improve your odds that ChatGPT names your business?
What "ranking in ChatGPT" actually means
The term "ranking in ChatGPT" is a bit misleading, borrowed from SEO where it genuinely applies. On Google you sit at position 4 or position 14, and that is measurable and stable. On ChatGPT that position does not exist. You are named in an answer, or you are not. There is no position 5 to climb toward.
That changes what "optimising" means. You are not working toward a higher position, you are improving the odds of being among the handful of names the answer includes. The distinction sounds subtle, but it decides which actions are worthwhile and which are wasted effort.
How ChatGPT actually arrives at its answer
ChatGPT pulls names from two different sources, fed by two different crawlers. OpenAI documents this itself:
| Crawler | What it does | Effect of blocking |
|---|---|---|
| GPTBot | Collects public content to train future models | Your content won't count toward model knowledge, but this does not affect live search |
| OAI-SearchBot | Fetches pages to surface in ChatGPT Search | ChatGPT can't fetch your page itself to show as a source; you can still be named indirectly, through other sources or model knowledge |
| ChatGPT-User | Fetches a page when a user explicitly requests it | That specific page fetch fails |
| OAI-AdsBot | Checks landing pages submitted as ads | Only relevant if you advertise through ChatGPT |
(Source: OpenAI's documentation on their crawlers, checked July 2026. OpenAI occasionally changes exactly how these bots behave, so check the source itself before you change your robots.txt.)
Two things stand out. First, "ranking in ChatGPT" is actually two separate
questions, model knowledge versus live search results, each with its own
user-agent you allow or block separately in robots.txt. Second, if you block
one of these bots by accident (via an overly broad Disallow: / or an
overzealous firewall rule), you have shut the door before you even started.
Step 1: make sure you are crawlable
This is the most commonly missed step, precisely because it is so boring. Check your robots.txt for rules that block AI crawlers, often included unintentionally as part of a generic "block all bots" rule. Want to show up in ChatGPT Search? Allow OAI-SearchBot at minimum. Want to also count toward what the model "knows" about you? Allow GPTBot too (keeping in mind that this only affects future training, not your visibility today).
Also check whether your site is technically crawlable: no content blocks that only appear behind a login wall, no critical information locked inside an image or video, no JavaScript-only rendering without a fallback. None of this is new if you know SEO, it is the same crawlability, just for a different kind of visitor.
Allowing both bots in robots.txt looks like this:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Allow: /
Robots.txt is a directive, not a lock on the door: it does not stop a bad actor spoofing the user-agent string. For legitimate AI crawlers it works fine, but also check your server logs or firewall/WAF settings are not blocking what robots.txt allows.
Step 2: write for quotability, not keywords
A language model does not quote a page title, it quotes a fact or an answer. Content that clearly answers a concrete question in a few sentences is easier to pick up than a page that only gets to the point after three paragraphs of throat-clearing.
What helps:
- Direct answers up front. Answer the question in the first paragraph, then elaborate. No long preamble before you get to the point.
- Structure with headings that are questions. A heading like "What does X cost" matches a user's question more directly than "Our pricing".
- FAQ blocks with short answers. Question-answer pairs are literally the format AI assistants often reply in, which makes them easy to quote.
- Concrete numbers and specific claims instead of superlatives. "For teams of 10 to 50 people" is quotable. "The best solution" is not, a model cannot verify a superlative and is therefore less likely to repeat it.
Step 3: make your brand entity unambiguous
A model needs to understand who you are before it can match you to a question. That sounds obvious, but many sites leave it implicit: who you are, who you serve, which market you're in, what sets you apart from three competitors saying the same thing. Make that explicit on your core pages, not tucked away on an "about us" page nobody reads.
This also helps beyond your own site. The more consistently your name, positioning, and category show up everywhere (your site, reviews, directories, news articles), the easier it is for a model to match you to the right question.
Step 4: build visibility beyond your own site
Part of what a model "knows" does not come from your page, it comes from how often and how consistently you are mentioned elsewhere: reviews, comparison sites, news articles, forums. That is not something you fix with a text edit, it is work you were already doing for brand building and PR, now with an extra reason to take it seriously.
This is exactly why a single keyword-stuffed page delivers little. A model forms its picture of your brand from multiple signals at once, not from one optimised page.
An example: from vague copy to quotable copy
An indicative example of the same piece of copy, written two ways.
| Vague (SEO habit) | Quotable | |
|---|---|---|
| Heading | "Why choose our accounting software" | "Accounting software for freelancers with under 20 invoices a month" |
| Opening line | "We've been helping businesses with smart solutions for years." | "Our software connects automatically to three major banks and processes VAT filings inside the platform itself." |
| Claim | "The best choice for small businesses." | "From €12 a month, no annual contract, with a free 14-day trial." |
The right-hand column is not necessarily more persuasive to a human reader, but it is easier for a model to pick up: it is specific, checkable, and free of superlatives.
What does not work
A few things that are not worth the effort:
- Keyword stuffing. A model matches on meaning, not on how many times a term repeats. Excessive repetition makes text less quotable, not more.
- Fake reviews or inflated claims. A model can weigh conflicting signals; inconsistent or overblown claims can backfire on how credible you come across. Aside from that, it simply is not honest to your customers.
- Blocking bots and then complaining about visibility. See step 1: an overzealous firewall rule is the quietest way to make yourself invisible.
- Optimising based on a single test. Asking ChatGPT one question and concluding you do or don't "work" says little; the answer varies by phrasing, by session, and by model update.
No guarantees: why "ranking" stays the wrong metaphor
We want to be upfront about this, partly because it is core to our own positioning: nobody can guarantee you a spot in a ChatGPT answer, us included. AI answers are not a static index, they shift with every model update and can turn out completely differently depending on how a question is phrased. What you can do is address the factors that improve your odds: crawlability, quotability, entity clarity, and mentions elsewhere. The rest is up to the model.
How to measure whether it is working
Because there is no fixed position to track, measuring here works differently than in SEO. You cannot check your ranking every morning, you need to periodically ask the same buying questions and see whether and how you are named.
A few things to watch:
- Whether you are named, and for which phrasings of the question you are and are not.
- Which assistant answers (ChatGPT, Gemini, Claude, and Perplexity can answer the same question differently).
- With what sentiment, and against which competitors.
- How stable that is across multiple runs, since a single measurement can be a fluke.
That is exactly what a GEO audit is for: not a one-off sample, but a repeatable measurement across multiple assistants and question variants, so you can see whether your visibility shifts after a change, instead of guessing.
Frequently asked questions
How do I get ChatGPT to recommend my business? Make sure OpenAI's OAI-SearchBot can crawl your site, write content that answers a concrete question clearly and factually, make your brand entity unambiguous (who you are, for whom), and build mentions elsewhere on the web. That improves your odds; there is no guarantee.
What is the difference between GPTBot and OAI-SearchBot? GPTBot collects content to train OpenAI's models; block it and your content will not count toward future training. OAI-SearchBot fetches pages to show in ChatGPT Search; block it and ChatGPT can't fetch your page itself to show as a source (you can still be named indirectly). They are two separate user-agents you allow or block independently in robots.txt.
Can I guarantee that ChatGPT will mention me in an answer? No. AI answers are not a fixed ranking and shift by question, by model, and by update. You can improve your odds by being crawlable, factual, and quotable, but anyone promising a guaranteed spot is selling something that does not exist.
Should I block GPTBot or allow it? That is a trade-off, not a technical necessity. Allowing it can grow your visibility in model knowledge over the long run; blocking it keeps your content out of the training material. For visibility in ChatGPT Search, OAI-SearchBot is the switch that matters, not GPTBot.