How to Rank in ChatGPT Search: What Counts

If you want to know how to rank in ChatGPT, start with how it finds pages. When ChatGPT searches the web, it rewrites the user's question into one or more search queries, sends them to its search partners, reads what comes back, and cites a few of those pages in its answer. So "ranking" comes down to three things you can work on: OpenAI's search crawler is allowed to reach your site, your pages show up in the indexes ChatGPT queries, and your page is the clearest answer to the query ChatGPT writes, which is usually shorter and more specific than what the person typed.

There is no position one, and OpenAI publishes very little about how it orders sources. Below is what it does document (checked October 2026), what I think follows from that, and the advice I would ignore. If you want the broader picture of optimizing for AI answers, including the research paper behind the term, read our post on generative engine optimization, and for where it differs from ordinary search work, see GEO vs SEO. This post sticks to ChatGPT.

How does ChatGPT search pick its sources?

ChatGPT answers some questions from what the model learned in training and some by searching the web. Only the second kind links to sources, and only that kind is something you can influence this month. OpenAI's help page on searching the web with ChatGPT describes the process.

First, ChatGPT "typically rewrites your query into one or more targeted queries" and sends them to search providers. OpenAI's own example: a researcher asks about drugs that target CCR8 for cancer, and ChatGPT first searches "CCR8 immunotherapy drug development 2025", then follows up with narrower queries based on what it found. It also uses approximate location from your IP address, so "restaurants near me" becomes "top restaurants San Francisco."

Those queries do not all go to one place. The same page says ChatGPT search "sometimes partners with other search providers" and points readers to Microsoft's privacy statement, which points to Bing as one of them. OpenAI also runs its own crawler for search, OAI-SearchBot.

Then it orders what it found, and here the documentation gets thin. OpenAI's whole statement on ordering is that ChatGPT "ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." That is the extent of the official guidance. Anyone who gives you a precise list of ChatGPT ranking factors is guessing, and so am I in the section on citations below.

The practical reading: your page competes on the rewritten query, inside an index ChatGPT can reach, and the first hurdle is getting into that index at all.

The three OpenAI crawlers, and which one matters for search

OpenAI's crawler documentation lists the user agents, and they do different jobs:

The settings are independent. You can block GPTBot to keep your content out of training and still allow OAI-SearchBot so you can be cited. OpenAI says changes to robots.txt take about 24 hours to reach its search systems.

The help page adds one more requirement that gets missed: your host or CDN has to allow traffic from OpenAI's published search crawler IP addresses. Firewall bot protection can block a crawler that your robots.txt allows, and nothing in the robots file will show it.

The technical checks, in order

Do these before you touch any content. They take an afternoon, and a failure in any of them makes the rest pointless.

1. Read your robots.txt as a crawler would. Look for any group that names OAI-SearchBot. robots.txt groups do not inherit from each other: a crawler obeys only the most specific group that names it. We found our own file had a named "allow everything" group for each AI crawler, which quietly exempted those bots from every disallow in the file. The fix was one wildcard group for all crawlers. The full story is in the GEO post.
2. Fetch a page with the crawler's user agent. Run curl -s -A "OAI-SearchBot" https://yoursite.com/your-page and check that you get a 200 and that your actual body copy is in the HTML. A 403 means the firewall is blocking it. A page that is empty until JavaScript runs is a page you should not assume any crawler reads.
3. Get into Bing. Since Bing is one of the indexes ChatGPT queries, verify your site in Bing Webmaster Tools, submit your sitemap, and fix what it reports. If you publish often, set up IndexNow so Bing hears about new and changed URLs without waiting for a crawl. We send changed URLs to IndexNow automatically after every deploy.
4. Check that the page ranks in plain search for the short version of the question. Take the prompt you care about, write the 4 to 6 word search query ChatGPT would probably turn it into, and search it in Bing. If you are not on the first page or two there, ChatGPT is unlikely to find you for it either.

If you need help with this layer, it is most of what our technical SEO work covers.

What makes a page get cited

OpenAI does not document this, so this section is my reading of how the system works, not a published rule set. It follows from the query rewriting: ChatGPT is looking for a page that answers a narrow, specific query, and it pulls a sentence or two from that page into a longer answer.

Answer the question in the first paragraph

If the rewritten query is "HubSpot workflow limit Starter plan", the page that says "Sales Hub Starter allows up to 50 workflows" near the top is easier to use than the page that gets there after 600 words of background.

Give each question its own page

A page that tries to cover twelve topics competes weakly for each of them. Separate pages for the separate questions buyers ask will match more rewritten queries.

Be specific, and date your facts

Names, prices, limits and the month you checked them. ChatGPT often rewrites queries with a year in them, as in OpenAI's own example, so a page that states when its facts were verified has an easier time matching and is safer to quote.

Say something the other results do not

If your page repeats the same ten tips as every other result, there is no reason to cite yours over theirs. First-hand detail, your own numbers and a clear opinion are what a model cannot get from the other nine results.

Show up on the pages that do get cited

Run your prompts and look at the citations. If they are mostly third-party roundups, review sites and forum threads rather than vendor pages, getting included in those roundups, or earning a genuine mention in those threads, is the work. Spamming forums gets you banned and helps nobody.

How to check whether ChatGPT cites you

There is no rank tracker from OpenAI, so you assemble the picture from four places. Paid AI visibility tools automate the first one, at a price.

Run the prompts yourself

Write down 10 to 20 questions a buyer would ask, run them in ChatGPT with search on, and record which sites are cited. Answers vary from run to run and by location, so repeat the same list monthly and look at the trend rather than one result.

Check referral traffic

Filter your analytics for traffic from chatgpt.com, checking both the referrer and the utm_source value, since a visit can arrive with either one. This only counts people who clicked, which will be a small fraction of the people who saw your name.

Read your server logs

Hits from OAI-SearchBot tell you OpenAI is crawling you for search. Hits from ChatGPT-User tell you someone's conversation fetched one of your pages, which is a useful signal about which pages come up in chats.

Use Bing's AI Performance report

In February 2026 Microsoft added an AI Performance report to Bing Webmaster Tools, which shows how often your pages are cited in Copilot, Bing's AI summaries and "select partner integrations", along with the grounding queries behind those citations. Microsoft does not say ChatGPT is one of those partners, so treat it as a read on how your content performs in Bing's AI layer, not as ChatGPT data. The grounding queries are still the best public example of what rewritten queries look like.

What I would skip

llms.txt as a ChatGPT tactic

We publish an llms.txt, generated from the same data file as the site so it cannot drift out of date. No AI company, OpenAI included, has documented using llms.txt to choose or rank sources. It costs us nothing to maintain, so we keep it, but I would not expect it to move anything.

Schema markup as a ranking factor

We add entity markup to every page, declaring what each page is about, because it helps disambiguation and costs little once it is built. OpenAI says nothing about structured data in its search documentation. Add it for the reasons it is useful, and do not expect citations from it.

Promises of a top spot

"Rank number one in ChatGPT in 48 hours" is a common pitch. There is no number one, the answers change between runs, and anyone selling a guaranteed placement is selling something OpenAI says does not exist.

New dates on old pages

Stamping a new "updated" date on old content is easy to spot when the facts underneath are a year stale, and it makes the page less trustworthy to quote.

Hidden instructions aimed at the model

Text on a page, or a pre-filled share link, that tells an AI to recommend you or "remember" your brand is an attempt to manipulate the model. It puts your domain on the wrong side of every company trying to keep its answers trustworthy, and a page that needs a trick to get cited has a content problem the trick does not fix.

Writing for robots

Keyword-stuffed headings and filler written "for AI" read badly to people and give a model nothing worth quoting. Write for the buyer and make the answer easy to find.

We build this into client sites as part of our generative engine optimization service.