How to Rank in ChatGPT: What Decides Whether It Mentions You

The short answer to how to rank in ChatGPT is that you can't, because a ChatGPT answer has no positions to rank for. What you can win is being mentioned, linked and described correctly, and how you win it depends on whether ChatGPT searched the web (which needs OpenAI's crawler to reach your site) or answered from memory (which depends on what the model learned before its training cutoff).

This guide covers both modes, the four checks you can run on your site today, and the popular tactics that OpenAI's documentation and the published studies do not support. Every OpenAI quote below was checked against OpenAI's own pages on October 16, 2026.

There is no rank in ChatGPT. Here is what you can win

OpenAI's entire public statement on how ChatGPT orders web results is two sentences: "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed" (OpenAI Help Center, "Searching the web with ChatGPT", updated August 2026). Nothing more is enumerated, so anyone selling a list of "ChatGPT ranking factors" is guessing.

Answers also change from run to run. A July 2026 survey of 45 GEO studies reports that commercial audits "reveal low source overlap, substantial run-to-run variability" (arXiv:2607.14035). One screenshot of "position 2" proves nothing. Three outcomes are real and checkable:

• Mentioned: ChatGPT names your brand when someone asks about your category.
• Cited: ChatGPT links to your page as a source. This only happens when it searched; our guide to LLM citations explains how those links are chosen.
• Described correctly: the price, founder, location and features it states are right.

The two ways ChatGPT answers about your brand

Every ChatGPT answer comes from one of two places, and they need different fixes. OpenAI says "ChatGPT may search the web automatically when your question would benefit from current information"; users can also pick Search from the tools menu. When it does not search, the answer comes from the model's training.

Memory has a date on it

As of October 2026, OpenAI's model documentation lists an "Apr 30, 2026" knowledge cutoff for its flagship GPT-6 Astra and for GPT-6.1 Sol, and "May 18, 2026" for GPT-6 Luna. The chat-latest alias, which OpenAI says points to "the latest Instant model currently used in ChatGPT", lists an Apr 20, 2026 cutoff and "will be regularly updated."

So "what ChatGPT knows" is not one date, and OpenAI does not say which model produced a given ChatGPT answer. A product launched or repriced after a cutoff does not exist in a memory answer, whatever you publish.

Search rewrites the question before it looks

When ChatGPT searches, it does not send your customer's prompt as typed. OpenAI says ChatGPT search "sometimes partners with other search providers" and, when it does, "typically rewrites your query into one or more targeted queries that it sends those providers."

The same page links Microsoft's and Shopify's privacy statements as the providers involved, and adds that saved memories can change the rewrite.

That is why "optimize for the exact prompt your buyer types" misfires. The query that actually retrieves pages is one you never see.

Step 1: allow the right OpenAI crawler

OpenAI runs separate bots for separate jobs, documented on its crawler page (undated; checked October 16, 2026). Blocking the wrong one is the most common self-inflicted error.

OpenAI states the settings are independent: a site "can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot." This robots.txt does exactly that:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Then look for a blanket rule. A User-agent: * group with Disallow: / blocks OAI-SearchBot too, unless OAI-SearchBot has its own group. OpenAI says search systems take about 24 hours to pick up a robots.txt change.

Step 2: make sure your firewall lets OpenAI in

robots.txt is half the gate. OpenAI's eligibility rule is a conjunction: "allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." Those addresses are published at https://openai.com/searchbot.json.

Run this from a terminal. It asks for your page while announcing itself as OAI-SearchBot, using the example user-agent string OpenAI publishes (the version number may change):

curl.exe -sL -o page.html -w "%{http_code}" -A "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot" https://www.example.com/pricing

On Windows, type curl.exe so PowerShell does not substitute its own alias; on Mac or Linux, use curl. This sends a GET request, as a real crawler does, saves the page to page.html and prints the status code.

A 200 is a pass only if page.html holds your real page. Open it, or search it for "challenge", "captcha" or "cf-chl": a challenge page can come back with a 200. A 403, 429, 503 or a challenge page means your firewall is turning the bot away while robots.txt says yes.

One caveat: your request does not come from OpenAI's IP ranges, so a firewall that verifies bots by IP may treat the real crawler differently. The ground truth is your CDN or server log filtered to OAI-SearchBot. If you use Cloudflare, its AI crawler controls changed on September 15, 2026; our breakdown of the Cloudflare AI crawler default change shows how to check what your zone now allows.

Step 3: put your facts in the HTML, not behind JavaScript

Vercel measured OpenAI's and Anthropic's crawlers on its network: "ChatGPT and Claude crawlers do fetch JavaScript files (ChatGPT: 11.50%, Claude: 23.84% of requests)" but "they don't execute them. They can't read client-side rendered content" (Vercel, December 17, 2024). OpenAI's crawler page says nothing about rendering, so treat that finding as current until OpenAI says otherwise.

The check takes a minute. Open your pricing page, choose "View page source" (not Inspect), and search for your price. If it is missing from the source, a crawler that does not run JavaScript never sees it.

Step 4: fix the facts ChatGPT repeats

A search-on answer is assembled from pages, yours and other people's. OpenAI warns that "Search results and citations can be incomplete, outdated, or incorrect." The part you control is whether the pages it reads agree with each other.

• Your own site first. If the pricing page, an old blog post and the footer state three different prices, any of them can be quoted. One version of every fact.
• Then third-party pages. Directories, review sites, comparison posts and old press releases that rank for your name. Ask for corrections where they are wrong.
• Accept what you cannot patch. A wrong fact in a memory answer stays until a newer model ships. The fixable surface is the set of pages a search-on answer reads.

Our guide to fixing AI hallucinations about your brand walks through the correction process in detail.

What does not get you mentioned in ChatGPT

Most "how to rank on ChatGPT" advice is borrowed from Google SEO or invented. Here is what the evidence says about the common tactics.

Do brand mentions on other sites help?

Several top-ranking guides say brand mentions drive ChatGPT recommendations. The honest version: big brands are cited far more, and nobody has shown that earning mentions causes it.

The largest published dataset, AI Search Arena conversations from March to May 2025, found that the top 20 news outlets took 67.3% of OpenAI models' citations to news sources (arXiv:2507.05301). That is an association with brand size. No study controls for size, because the available "authority" metrics are themselves built from size.

Being described accurately on pages that already rank for your category is worth doing for human buyers anyway. Paying for thin "best of" placements to move an AI answer is a bet with no published evidence behind it.

How to check what ChatGPT says about your brand

1. Turn memory off or log out. OpenAI says saved memories can change how ChatGPT rewrites a search query, so your own account is a biased test.
2. Ask category and brand questions. "What is [brand]?", "How much does [brand] cost?", "Best [category] for [use case]".
3. Note which mode answered. Citations or a Sources button mean it searched. No citations suggests memory but does not prove it.
4. Force search. Use the refresh control and choose "Search the web" to see the search-on answer next to the memory one.
5. Repeat and compare. Run each question several times and check every stated fact against your own records.

For a fuller test plan, see what ChatGPT says about your brand. To see how ChatGPT compares with the other engines, our guides cover Perplexity SEO, how to rank in AI Overviews and the cross-platform mechanics of LLM brand visibility.

99Visibility's ChatGPT visibility audit queries ChatGPT through OpenAI's API with web search on, labels each answer as searched or from memory, and flags facts that contradict the ones you enter. Its access check requests your page as OAI-SearchBot, GPTBot and ChatGPT-User and flags firewall blocks.

Frequently asked questions

How do I get mentioned in ChatGPT?

For answers that use search, allow OAI-SearchBot in robots.txt, confirm your CDN lets OpenAI's published IP ranges through, and make sure your key facts are in the page's HTML. For answers from memory, nothing you publish today shows up until a newer model is trained. Keep your facts consistent across your own site and the third-party pages that describe you.

Does ChatGPT use Bing to search?

OpenAI says ChatGPT search "sometimes partners with other search providers" and links Microsoft's and Shopify's privacy statements. It does not say how often, or how partner results are combined with its own crawl. Staying indexed in Bing is reasonable hygiene, but no published measurement links it to ChatGPT mentions.

Can you manipulate ChatGPT into recommending a brand that does not exist?

In controlled tests, sometimes. A June 2026 study fed manipulated web evidence to AI search agents in a controlled framework and measured attack success rates from 0.0% to 31.4% across 13 model backends (arXiv:2606.16821); it did not test ChatGPT's live product. The practical risk for you is the reverse: a third-party page that misdescribes your brand can be repeated as fact.

How do I measure brand visibility in ChatGPT?

Record whether ChatGPT mentions you, whether it cites you, and whether its facts are right, separately for search-on and search-off answers. Repeat each question several times, because answers vary. Never report a position: there is no ground truth for one.

Is SEO dead now with AI?

No. For ChatGPT, the documented requirement is crawler access, which is technical SEO under another name. For Google's AI features, Google says the same SEO best practices apply and there are no special optimizations.

To see what ChatGPT and four other AI platforms say about your brand, and which facts they get wrong, run a free 99Visibility audit.