To rank in ChatGPT, let OAI-SearchBot crawl your site, get your pages indexed in Bing as well as Google, publish pages that answer buyers' questions with specific, sourced facts, and earn mentions on the sites ChatGPT already cites. Then measure with a fixed set of prompts, run repeatedly. Nobody can guarantee you a placement.
How does ChatGPT pick its sources?
ChatGPT answers in one of two ways. Either it replies from what it learned in training, with no links and possibly out-of-date facts, or it searches the web, reads a set of pages, and writes an answer with citations. Only the second gives you a link, and only the second responds to changes you make this month.
Almost every “rank in ChatGPT” project is really about that second route. It’s the part of our AI SEO and ChatGPT visibility work that clients ask about most, and it works more like search than most guides admit.
When does it search?
It decides for itself. OpenAI’s documentation for its web search tool says the model can “choose to search the web or not based on the content of the input prompt”. In our testing the product behaves the same way: questions about local businesses, recent events, or “which company should I use for X” usually trigger a search, while general explanations often don’t. Users can also switch search on themselves.
Where do the results come from?
OpenAI is open about one piece and vague about the rest.
The clear piece is its crawler. In OpenAI’s words, “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.”
The vaguer piece is everything else. OpenAI’s help page on ChatGPT search mentions third-party search providers and content from partners. Microsoft announced in May 2023 that it was bringing Bing to ChatGPT as the default search experience. How the mix looks today isn’t published. For a business, the practical conclusion is simple: be crawlable by OAI-SearchBot and well indexed in Bing, not only Google.
Retrieved isn’t the same as cited
OpenAI’s web search docs separate inline citations, which show “only the most relevant references”, from the full list of sources the model looked at. The ChatGPT interface shows the same split: a few links in the text, more behind the sources list. Your page can be read and still not earn the visible link. That gap is where content quality matters.
There’s no position one
The same prompt returns different answers on different runs, for different users, with or without memory, from different places. “Ranking” in ChatGPT is really a probability: how often you’re named across many runs of the questions that matter to you. Keep that in mind when anyone shows you a single screenshot.
Step 1: Make sure OAI-SearchBot can reach you
Open yoursite.com/robots.txt and look for OpenAI’s user agents. If you want ChatGPT search to show your pages but don’t want your content used for model training, this is the pattern:
# Allow ChatGPT search, keep content out of model training
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /OpenAI says each setting is independent of the others, and that robots.txt changes take about 24 hours to reach its systems. Our robots.txt generator writes these rules for you if you’d rather not edit the file by hand.
Then check the layer most people forget. CDNs and firewalls often have a bot-protection setting that challenges or blocks AI crawlers regardless of what robots.txt says. OpenAI publishes the IP ranges OAI-SearchBot uses (linked from its crawler page) so you can allow them explicitly. We’ve found this setting switched on in sites whose owners had never heard of it.
A third agent, ChatGPT-User, fetches pages when a person asks ChatGPT to open a link. OpenAI notes that because these requests are user-initiated, “robots.txt rules may not apply.”
Step 2: Get indexed in Bing
Most businesses we audit treat Bing as an afterthought, and for ChatGPT that’s a mistake. Here’s the routine:
- Verify your site in Bing Webmaster Tools. You can import a verified Google Search Console property to save time.
- Submit your XML sitemap.
- Turn on IndexNow, so Bing hears about new and updated pages as soon as you publish. Many CMS plugins and CDNs support it.
- Search Bing for your key pages and confirm they’re there.
While you’re in there, open the AI Performance report. It counts citations in Microsoft Copilot and Bing’s AI summaries, not ChatGPT, so treat it as a related signal rather than your ChatGPT number. Bing’s own advice in the launch post for getting cited more is worth reading: keep content fresh, use clear headings, tables and FAQ sections, and back claims with examples, data and cited sources.
Step 3: Publish pages that answer what buyers ask
People don’t type keywords into ChatGPT. They type the whole problem.
A search might be “accounting firm dubai”. The ChatGPT version is closer to: “Which accounting firms in Dubai handle VAT registration for e-commerce start-ups and can work in Arabic?” To be named in that answer, some crawlable page on your site needs to say plainly that you do VAT registration, that you work with e-commerce start-ups, that you’re in Dubai, and that you work in Arabic. Surprisingly often, at least one of those facts lives only in a PDF, an image or the founder’s head.
What we do on client pages:
- Put the deciding facts in text: services, who they’re for, locations served, languages, how the engagement works.
- Lead with the answer: a question-shaped heading, then two or three sentences that answer it, then the detail.
- Be specific and cite things: named standards, real figures, quotes from named people. The research behind the GEO study’s tested methods found citations, quotations and statistics helped most, and keyword stuffing hurt.
- Write the comparison pages: “X vs Y” and “is X right for a company like mine” are common prompts. If you don’t publish a fair answer, ChatGPT will build one from someone else’s page.
Step 4: Get mentioned on the pages ChatGPT already cites
This step moves the needle most for “who should I hire” prompts, and it’s the one most guides skip.
Run your buyer prompts (see the testing method below) and export every cited URL. Group them: directories, review platforms, “best of” articles, trade publications, forums, competitor pages. Each group needs a different approach:
- Directories and review platforms: get listed, with the same name, services and contact details you use everywhere else.
- “Best of” articles: contact the author with accurate information and something useful to add. Some will update; many won’t.
- Trade publications: contribute expertise they’ll actually publish, such as data you hold or a practitioner’s view on a change in your sector.
- Forums: answer questions honestly under your own name. Undisclosed promotion backfires quickly.
One tactic we don’t use: publishing a “best companies for X” list on your own site with yourself at number one. Assistants do sometimes cite these lists. A buyer who notices also stops trusting everything else you say.
Step 5: Make your business easy to identify
When ChatGPT describes you, it’s stitching together your site, your Google Business Profile, LinkedIn, directories and press. If those disagree about your name, your services or where you operate, you get described vaguely or wrongly.
Pick one canonical version of each fact and make every profile match it. Add Organization schema with sameAs links to your official profiles (our free Organization schema tool builds the JSON-LD). In the Gulf, decide how your Arabic and English names relate and use them consistently, since a model can’t infer that two spellings belong to one company.
If ChatGPT states something wrong about you, such as an old address or a service you dropped, look at the sources it cites for that answer. The error usually traces back to one outdated listing.
Step 6 (online stores): look at product feeds
If you sell products, OpenAI runs a merchant programme built on its Agentic Commerce Protocol, where merchants share a structured product feed with ChatGPT. Applications start at chatgpt.com/merchants. It’s separate from everything above and worth a look if shopping questions matter to you.
How do you test your visibility in ChatGPT?
Write 20 to 40 questions your buyers would actually ask, run each one at least three times in a clean session with web search on, and log whether you’re named, whether you’re cited, which competitors appear and which sources ChatGPT used. Repeat the same set every month. The trend across runs is your ranking.
The details matter, so here’s the method we use:
- Write four kinds of prompt. Category (“which firms in Riyadh do X”), problem (“how do I fix Y”), comparison (“A vs B for a company like mine”) and reputation (“is [your company] reliable?”). The reputation prompt is the one most people forget, and it’s often where the wrong descriptions turn up.
- Cover your languages. For Gulf markets, run the Arabic versions too. Answers in Arabic can draw on completely different sources, which is why it’s standard in our AI SEO work for Dubai companies.
- Clean the session. Use a Temporary Chat or a logged-out window so memory and earlier conversations don’t skew the result, and make sure web search is on.
- Run each prompt three times or more. One answer is an anecdote.
- Log every run. A spreadsheet is enough:
| Prompt | Run | Named? | Cited link? | Competitors named | Sources cited |
|---|---|---|---|---|---|
| Which firms in Riyadh do X? | 1 | Yes, third of four | No | A, B, C | Directory, trade article |
| Which firms in Riyadh do X? | 2 | No | No | A, B | Directory, competitor blog |
After a few months you’ll see which sources keep coming up, which competitors own which prompts, and whether your changes moved anything.
What won’t help
Buying ads and expecting citations. ChatGPT ads exist (OpenAI’s crawler page lists OAI-AdsBot, which checks pages “submitted as ads on ChatGPT”), but an ad is a paid slot, not a mention in the answer.
An llms.txt file on its own. OpenAI’s crawler documentation covers robots.txt rules, not llms.txt, as the way to control ChatGPT search access. The file is harmless and cheap; our llms.txt explainer covers what it actually does.
Hidden instructions. White-on-white text telling AI to recommend you is a modern version of hidden keywords. It risks the domain for a gain nobody has shown.
Ignoring Google. ChatGPT isn’t the only assistant your buyers use, and the groundwork overlaps heavily. Our comparison of GEO, SEO and AEO shows how much of this is shared.
How long does it take?
Crawler changes reach OpenAI in about a day. New and updated pages get picked up as fast as Bing and OAI-SearchBot recrawl them, often within days or weeks. Mentions on third-party sites take months to earn. What a model learned in training only changes when OpenAI ships a new model, which you can’t schedule. Report those timelines separately and you’ll avoid deciding a project failed because the slowest part hasn’t moved yet.