How to rank in ChatGPT: an evidence-first guide

By Jithin Palepu, founder of First100. Published 2026-09-27. Last updated 2026-09-27.

To rank in ChatGPT, you need four things in order. OAI-SearchBot must be allowed to crawl your site. Your pages must be indexed by the search providers ChatGPT queries, starting with Bing. Your pages must answer the question in the first paragraph. And other sites must name you. Then you measure mention rate over many runs, because a single answer tells you almost nothing.

Key takeaways

• OpenAI's crawler docs state that sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers", and robots.txt changes take about 24 hours to apply.
• OpenAI's help center says ChatGPT search sends rewritten queries to partner search providers and names Microsoft and Shopify as two of them.
• In the Princeton-led GEO paper, adding quotations raised visibility by 41.2% and adding statistics by 32.7% on its 10,000-query benchmark.
• Ahrefs found only 8% of ChatGPT in-text citations rank in Google's top 10 for the original prompt, so Google rank is neither required nor enough.
• First100 is not cited yet: our first live answer-engine check returned Postiz, Metricool and Buffer for our own buyer question, and not us.

What does "rank in ChatGPT" actually mean?

ChatGPT has no fixed results page, so "rank" means one of two things. Either your brand is named in the answer, or your page is cited as a source with a link. Both come from the same pipeline: ChatGPT rewrites the question into search queries, sends them to its own index and to partner search providers, reads the top pages, and writes an answer from them. You influence the inputs to that pipeline. You do not control the answer.

OpenAI says this plainly. Its help article on searching the web with ChatGPT states that ChatGPT "ranks search results using multiple factors" and that "placement is not guaranteed." The same page lists the one hard requirement for eligibility: allow OAI-SearchBot, and make sure your host or CDN allows OpenAI's published searchbot IP addresses.

The answers are also unstable. SparkToro's study ran 12 prompts 2,961 times across ChatGPT, Claude and Google's AI features with 600 volunteers in November and December 2025. The chance of getting the same brand list twice was under 1 in 100. The chance of the same order was about 1 in 1,000. So the useful metric is the share of runs that name you.

Which OpenAI crawler decides whether you appear?

OpenAI runs several crawlers, and only one controls ChatGPT search. OpenAI's crawler overview describes OAI-SearchBot as the bot "used to surface websites in search results in ChatGPT's search features." GPTBot is a different bot, used to crawl content for training foundation models. Blocking GPTBot keeps you out of training and leaves search alone. Blocking OAI-SearchBot removes you from search.

The docs are direct about the cost of blocking it: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." They also say it takes "~24 hours from a site's robots.txt update" for OpenAI's systems to adjust. That is fast compared with most search engines, so a wrong robots.txt is cheap to fix once you find it.

The third bot, ChatGPT-User, fetches a page when a user's action asks for it. OpenAI notes that "robots.txt rules may not apply" to it because a person started the request, and that it "is not used for crawling the web in an automatic fashion." A fourth, OAI-AdsBot, only visits pages submitted as ads.

Robots.txt is only half of access. A CDN or firewall rule can block OAI-SearchBot and never appear in robots.txt. First100's analyze_seo probes each AI crawler's user agent against your site and reports a block from either layer. Our own audit rules name Cloudflare's AI Crawl Control, Vercel's firewall and AWS WAF's Bot Control as the usual places to look.

What should a robots.txt for ChatGPT look like?

A robots.txt that keeps you in ChatGPT search names OAI-SearchBot and allows it. You can block GPTBot in the same file if you do not want your pages used for training. The two rules are independent, so the choice about training has no effect on search. A wildcard Disallow: / under User-agent: * blocks OAI-SearchBot too, unless a more specific group allows it.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Sitemap: https://www.example.com/sitemap.xml

Keep the file reachable. Our audit notes, drawn from Google's documentation, record that a robots.txt answering with a server error can make Googlebot stop crawling, so a 200 or a 404 is the safe response. We apply the same rule to every crawler, since none of them can read an allow rule from a file they cannot fetch. Test the live file after every deploy. Frameworks and CDNs sometimes rewrite it, and a staging rule that blocks everything can ship to production by accident.

Then check the layer above robots.txt. Cloudflare, Vercel and AWS all offer bot rules that act before your server sees the request. A request that gets a 403 from the edge never reads your allow rule. The only reliable test is to request a real page with the crawler's user agent and look at the status code, then confirm in your server logs that OAI-SearchBot visits arrive after the fix.

Does ChatGPT use Bing, and why does that matter?

Yes, in part. OpenAI's help center says "ChatGPT search sometimes partners with other search providers" and sends them rewritten, targeted queries. For how those providers process queries, it links the Microsoft privacy statement and the Shopify privacy policy. Microsoft runs Bing. So a page Bing has not indexed misses one of the sources ChatGPT can pull from, even if OAI-SearchBot can read it.

The OpenAI Publishers and Developers FAQ confirms the mix from another angle. It says OpenAI may learn the URL of a page "from a third-party search provider or by crawling other pages." Then it may show just the link and title, even if the page itself is disallowed. The same FAQ says "any public website can appear in ChatGPT search." It also notes ChatGPT adds utm_source=chatgpt.com to referral links, which is how you measure clicks later.

Google ranking is a weak proxy. Ahrefs checked 15,000 long-tail queries in early July 2025 and published the overlap between AI citations and search rankings. Only 8% of ChatGPT in-text citations ranked in Google's top 10 for the original prompt. Across ChatGPT, Gemini and Copilot combined, the figure was 12%. The explanation offered: assistants search many query variations, not your exact keyword.

How do you get indexed in Bing quickly?

Bing indexing is the cheapest step on this list. Verify your site in Bing Webmaster Tools, submit your sitemap, and use IndexNow to tell Bing about new or changed URLs the day they change. IndexNow is a single POST with your URL list and a key file hosted on your domain. It removes the wait for a crawler to notice, while the decision to index stays with Bing.

The IndexNow documentation allows up to 10,000 URLs per POST. It says that participating engines "agree that submitted URLs will be automatically shared with all other participating search engines." Its search engine list currently names seven receivers: Bing, Yandex, Seznam, Naver, Yep, Internet Archive and Amazonbot. Google is not on it.

Read the response codes carefully. The docs say a 200 "only indicates that the search engine has received your URL." A 202 means the submission was accepted while the key is still being validated. Neither means indexed.

Here is what we did on our own site on 2026-09-15. We placed an IndexNow key file in public/, then submitted our 8 key pages with First100's submit_indexnow tool. The response was 202. Bing picked those pages up. Google, which does not read IndexNow, had not yet indexed /pricing or /docs at the time.

Bing also shows AI citations directly. Its AI Performance report, announced on February 10, 2026, reports total citations, average cited pages, grounding queries and per-page citation counts. It covers Microsoft Copilot, AI summaries in Bing and select partners. It does not report ChatGPT, but grounding queries are a useful list of what AI systems search for.

The six steps to rank in ChatGPT

This is the order we would follow on a new site. Each step depends on the one before it, so a missing crawler permission wastes everything after it. Steps 1 and 2 take an afternoon. Steps 3 and 4 take weeks of writing. Step 5 takes months, and people do most of it rather than software. Step 6 never ends.

1. Allow OAI-SearchBot. Check robots.txt for a User-agent: OAI-SearchBot block or a wildcard disallow. Then check your CDN and firewall, because rules there never show in robots.txt. Wait about 24 hours after a fix.
2. Get indexed in Bing. Verify in Bing Webmaster Tools, submit the sitemap, and send changed URLs through IndexNow. Keep the sitemap to pages you want cited.
3. Write answer-first passages. Put a 40-70 word answer under the H1 and a self-contained passage under every heading. Name numbers and sources inside the sentence, and link them.
4. Publish comparison and use-case pages. Answer the buying questions ("X vs Y", "best tool for Z") on your own site, with dated facts.
5. Earn third-party mentions. Get named on lists, directories, reviews and community threads that already rank. Never automate posting.
6. Measure mention rate. Ask the same buyer questions across engines on a schedule, count how often you are named or cited, and watch utm_source=chatgpt.com referrals.

Server-rendered HTML sits under step 1. Vercel's analysis of AI crawlers, published December 17, 2024, found that none of the major AI crawlers render JavaScript. That list includes OAI-SearchBot, ChatGPT-User and GPTBot. The same post counted 569 million GPTBot requests in a month across Vercel's network. If your main content only appears after JavaScript runs, those crawlers see an empty page.

Why do answer-first pages get cited more?

The strongest evidence here is the GEO paper, "GEO: Generative Engine Optimization", published at KDD 2024. Its authors are from Princeton University and IIT Delhi, among others. They built GEO-bench, 10,000 queries across 25 domains. They then rewrote source pages nine ways and measured how visible each page became inside generated answers. The paper's abstract reports visibility gains of "up to 40%".

The exact numbers are in the full paper on arXiv. On the test set, measured as position-adjusted word count:

• Quotation addition: +41.2%
• Statistics addition: +32.7%
• Fluency optimization: +28.7%
• Cite sources: +27.7%
• Keyword stuffing: -8.7%

The authors also ran a real-world test on Perplexity.ai. Quotation addition gained 19.5% there and statistics addition 8.6%. Keyword stuffing lost 8.8%. The pattern held: concrete, sourced content helped, and old SEO tricks hurt.

Treat those numbers with care. The GEO test used its own generative engine and Perplexity, not ChatGPT. A later NeurIPS 2025 benchmark, C-SEO Bench, found most conversational SEO rewrites "largely ineffective" and often harmful, while traditional SEO held up better. So the safe reading is modest. Pages that state facts with sources are easier to quote. That also happens to be how a careful reader wants a page written.

In practice, answer-first means three habits. Open with the answer, not a story. Start every section with a passage that makes sense if lifted out alone. And write numbers with their source in the same sentence, like "Ahrefs found 8%", so a quoted fragment keeps its attribution.

Do comparison pages and third-party mentions matter?

Yes, and they matter more than most on-page work. When someone asks ChatGPT to compare tools, it searches for comparison content. If the only comparison pages about you are on other sites, those pages decide how you are described. Your own "X vs Y" page, with dated facts, gives the answer one more accurate source to read.

Ahrefs looked at what ChatGPT cites for buying questions in its study of "best" lists. Across 750 prompts and 26,283 cited URLs in software, products and agencies, "best X" blog lists were 43.8% of cited page types. About 35% of those lists sat on low-authority domains. Of 1,100 lists checked, 79.1% had been updated in 2025.

Mentions beat links in a related study. Ahrefs measured 75,000 brands in its AI Overview brand correlation study, published May 26, 2025. Branded web mentions correlated 0.664 with AI Overview visibility. Backlinks correlated 0.218. That study covers Google's AI Overviews, not ChatGPT. It is still the best large dataset on why being named matters more than being linked.

What we did: on 2026-09-15 we published three comparison pages at /vs/typefully, /vs/hypefury and /vs/buffer. Each fact is read from the competitor's own pages on a stated date, and "unknown" stays unknown. Two days later we added use-case and integration pages. Whether they get cited is an open question we are measuring.

For mentions, a person has to do the work. Submit to directories, answer questions in communities where you are genuinely useful, and pitch "best X" list authors with facts. Automated posting or replies get accounts banned and do not build the trust these lists depend on.

What happened when we checked our own site?

We ran our own tools on our own site before writing this, and the results were not flattering. On 2026-09-15, analyze_seo gave www.firsthundred.app a health score of 72 with no critical issues. Search Console for the prior 28 days showed 8 branded queries with 11 clicks, and 7 non-branded queries with 0 clicks. Nobody found us by searching for the problem we solve.

The biggest fix was the sitemap. It listed 525 URLs, including 500 programmatic /find-customers/ pages. Google had indexed 7 of 30 sampled URLs and skipped /pricing and /docs. We trimmed the sitemap to 26 URLs, kept the 500 pages live but out of the sitemap, and submitted the 8 key pages through IndexNow (202). The same day we added Organization sameAs data, a shorter pricing description and question-shaped docs headings.

Then we asked an answer engine directly. First100's ai_answers tool sends a brand's buyer questions to ChatGPT, Claude, Gemini and Perplexity through DataForSEO, with web search on. For each answer it records which brands are named and which domains are cited. Our first live answer was from Claude, for "tools to schedule posts from Claude Code". It named Postiz, Metricool and Buffer. It did not name First100.

That result is the honest baseline. We have not yet been cited for our own core question. It also shows the cost of measuring properly. One answer cost $0.049 through DataForSEO, about $0.05 each. Three questions across four engines is about $0.60 per round. Repeat that round many times before you trust a trend, because SparkToro's data says single answers rarely repeat.

How should you measure ChatGPT visibility?

Measure a rate, not a rank. Pick three to ten questions a buyer would really ask. Run each several times per engine on a fixed schedule. Record whether you are named, whether your domain is cited, and which competitors appear. Track the share of answers that name you over time. One run proves nothing, because SparkToro found the same list appears under 1 time in 100.

Three data sources cover most of it:

• Answer sampling. Ask the questions yourself, or use a tool. ai_answers stores up to 3 buyer questions per brand and reruns them every two weeks on paid plans.
• Referral traffic. Filter analytics for utm_source=chatgpt.com, which OpenAI's FAQ says ChatGPT adds to referral URLs. First100's ai_referrals counts visits from AI engines if you add its snippet to your site.
• Bing AI Performance. Free, and it shows grounding queries and cited pages for Copilot and Bing's AI summaries.

Keep a dated log of what you changed. If mention rate moves, you want to know which change came before it. Expect slow movement. Our internal research estimates a measurable AI mention rate at around 90 days, and compounding mentions nearer 180. Those are planning assumptions, not a promise.

FAQ

Does blocking GPTBot remove my site from ChatGPT?

No. OpenAI's crawler docs say GPTBot collects content that may be used for training. ChatGPT search uses OAI-SearchBot. You can block GPTBot and still appear in ChatGPT search answers, as long as OAI-SearchBot is allowed in robots.txt and at your CDN.

How long until ChatGPT sees a robots.txt change?

OpenAI says about 24 hours from a robots.txt update for its systems to adjust to OAI-SearchBot rules. Getting cited after that still depends on indexing, content and mentions, which take weeks or months.

Do I need to rank on Google to get cited by ChatGPT?

No. Ahrefs found only 8% of ChatGPT in-text citations ranked in Google's top 10 for the original prompt. ChatGPT searches many query variations and uses partner providers such as Microsoft's Bing, so Bing indexing and clear answers matter more than a Google position.

Does llms.txt help me rank in ChatGPT?

Not on current evidence. OpenAI's publisher guidance asks for OAI-SearchBot access and does not mention llms.txt. An Ahrefs study of 137,210 domains found 97% of llms.txt files got zero requests in May 2026. Spend the time on crawler access and content first.

How do I get cited by ChatGPT as a new site?

Make every page crawlable in raw HTML, get indexed in Bing, and answer buyer questions in the first paragraph with sourced numbers. Then get named on lists and directories that ChatGPT already cites. Ahrefs found 35% of cited "best" lists were on low-authority domains, so a new site has a real opening.

Check your AI crawler access today

Start with step 1, because nothing else counts without it. Connect First100 to your agent with the quickstart, or add it to ChatGPT with the ChatGPT setup guide. Then ask your agent to run analyze_seo on your domain. It reports whether OAI-SearchBot, Claude-SearchBot and PerplexityBot can reach your pages, at both robots.txt and the CDN. The tools page lists submit_indexnow and ai_answers for steps 2 and 6.