How to Rank in ChatGPT: What Gets You Recommended

There is no ranking, there are three routes to a mention

When someone asks ChatGPT a question, a brand can end up in the answer in one of three ways. Each has a different level of control for you, which is why a single "ranking" tip is misleading.

Most of the practical work is on the second route, which the next page covers in detail: how to rank higher in ChatGPT search. This page deals with the larger question of what to build around it.

What OpenAI documents, and what it does not

OpenAI's crawler documentation names four user agents. Three matter for organic visibility. OAI-SearchBot is used to surface websites in ChatGPT's search features, and robots.txt applies to it. Sites that opt out do not appear in ChatGPT search answers, though they can still appear as plain navigational links. GPTBot collects content that may be used to train OpenAI's foundation models, and disallowing it signals that your content should not be used for that purpose. ChatGPT-User handles user-initiated actions, such as visiting a page to answer a question, and OpenAI says robots.txt rules may not apply to it. The fourth, OAI-AdsBot, only checks landing pages submitted as ChatGPT ads.

The settings are independent. A site can allow the search bot and block the training bot, or the reverse. OpenAI also publishes IP address lists for each agent.

In the documents we could open, OpenAI publishes no ranking formula, no list of factors and no promise of placement. Anyone who claims to know the exact formula is inferring it from observation at best.

What research suggests helps

None of the evidence is specific to ChatGPT search, so treat it as a direction.

• Specific, sourced content. The 2023 Princeton-led GEO paper found that adding quotations, statistics and cited sources increased how much of a generated answer relied on a page, and that keyword stuffing did worse than no change. The experiments used a 2023 model and two engines, so the result is a hypothesis for ChatGPT today.
• Visibility is not the same as rank. Ahrefs found in 2026 that only 37.9% of URLs cited by Google AI Overviews also ranked in Google's top 10. That is a Google figure, and it shows that engines can cite pages beyond the top results. It says nothing direct about what ChatGPT does.
• Sound SEO is the safe base. Google's guide says its generative features rest on core ranking systems. For ChatGPT, OpenAI has not said the same, so we treat clean, indexable, helpful pages as a precaution and not a proven mechanism.

A 90-day plan that stays inside the evidence

1. Weeks 1 to 2: access. Read your robots.txt and your CDN or security rules. Confirm OAI-SearchBot is allowed. Decide separately about GPTBot.
2. Weeks 2 to 6: five pages. Pick the five pages closest to revenue. Rewrite each so every section opens with the answer, includes a specific figure or condition, and names its source.
3. Weeks 4 to 8: facts elsewhere. Make your name, services, location and pricing consistent on your profiles and directories. Earn genuine mentions from customers and partners. Google's guide warns that manufactured mentions are less useful than they look.
4. Ongoing: test. Run the test below every month and keep the log.

Which five pages to rewrite first

Do not start with the homepage. Score your pages out of 10 on each of these, then rewrite the five highest.

For a Limassol accounting or dental practice, that usually means a pricing or fees page, the two most common service pages and an emergency, urgent or "how to book" page. For a software company, it is often the integration and security pages. The pattern is the same: pick pages where a specific answer is easy to give and valuable to the person reading it.

Prompts a local business should include in its test

Customers use a mix of languages and styles, so write prompts the way they talk. These are illustrative.

• "Best [your service] in Limassol for [type of customer]"
• "How much does [your service] cost in Cyprus"
• "[Your service] near [neighbourhood] open on Saturdays"
• "Is [your business name] reliable"
• The same questions in Greek, if your customers use it. We have not measured Greek-language answers, so treat the results as a first look.

The last two types matter most. Reputation prompts reveal what an assistant says when it is asked about you by name, and Greek prompts show whether the language you serve is covered at all.

Run your own test

Because you cannot see inside ChatGPT, the only fair evidence is sampling. The protocol below is the one we would use. It is a method for you to run, not a set of results.

1. Write 15 prompts a buyer might use: five broad (category and city), five comparison (you versus a competitor type), five problem-led.
2. Run each prompt three times in a fresh chat, with the same wording and the same search setting each time.
3. Record, for each answer: were you named, were you linked, in what position, and was what it said about you accurate.
4. Calculate your mention rate. For example, being named in 9 of 45 answers is 20%.
5. Repeat monthly.

The arithmetic matters. With 45 answers, one extra mention moves the rate by about 2.2 points, so a change of 2 or 3 points is noise. Do not report a trend until it holds for two or three months.

What to ignore

• Prompts that "teach" ChatGPT to remember you. They do not change the model for other users.
• Hidden text aimed at assistants. It is manipulation, it risks your standing with search engines, and we advise against it.
• Rank scores from tools. Some tools publish a visibility score. These are the tool's own samples, useful for trends, and not OpenAI's data.
• Promises of a date. No source we found supports a timeline.

For the full evidence map see the AI search optimization guide, and if you want the audit done for you, our AI search optimization service begins with exactly this test. Start this week by writing the 15 prompts.