How to Rank on ChatGPT in Search Results: Crawl Access, Bing Index, Third-Party Mentions

Quick answer

Under this method, Canlah AI treats how to rank on ChatGPT as three ordered jobs: allow OAI-SearchBot so pages can enter OpenAI’s retrieval pool, publish pages that answer one buyer question in their first 60 words, then earn mentions on the third-party pages ChatGPT already retrieves for the category. Access comes first because OpenAI’s crawler documentation states that sites opted out of OAI-SearchBot “will not be shown in ChatGPT search answers” and that a robots.txt change takes about 24 hours to register.

Canlah AI runs this work as a managed service and publishes the manual version here so any website owner can test their own site for free. Its self-audit of canlah.ai on September 7, 2026 shows which job usually decides the result: 39 of 39 branded answers named the brand, while 7 of 177 open buyer-question answers did.

Treat any single answer as directional. ChatGPT is a sampled system, so one run cannot prove a stable position or causation. Freeze a panel of eight to twelve buyer questions and repeat it by engine, market and language before deciding what to change.

Disclosure: Canlah AI publishes this guide and operates the paid monitoring service described near the end. Crawler details come from OpenAI’s own documentation, retrieval details from Peec AI’s September 4, 2026 investigation and indexing details from Bing Webmaster Tools, all reviewed on September 23, 2026 and none tested through paid accounts.

What does ranking on ChatGPT actually mean?

Ranking on ChatGPT is not one position in a list of ten links; it is four gates a page passes before its domain shows in the source rail beside an answer. A site owner controls the first two almost completely, influences the third and controls the fourth only indirectly.

• Crawl access: whether OAI-SearchBot is allowed to fetch the page in robots.txt, at the CDN and at the firewall.
• Index membership: whether the fetched page is stored in a retrieval index ChatGPT queries, including OpenAI’s own index and the third-party search providers it still calls.
• Retrieval: whether the page enters the candidate set for a buyer question, which changes by wording, location, account state and time.
• Selection: whether the model uses the page in the written answer and attributes it, rather than paraphrasing a competitor page that made the same point more plainly.

A page that was never crawled and a page that was crawled but never retrieved look identical from the outside. Separating them takes a server-log check, not another content sprint. Recommendation and citation are measured separately, in how to check if AI search recommends your brand.

Which OpenAI crawler decides whether you rank on ChatGPT?

Four OpenAI user agents touch a website and only one of them, OAI-SearchBot, decides whether the site can appear in ChatGPT search answers. OpenAI’s documentation states that each setting is independent, so blocking the training crawler does not remove a site from search, and allowing the training crawler does not add it.

OpenAI user agents and what each one decides, from OpenAI’s crawler documentation reviewed September 23, 2026.

User agent	What it decides	Obeys robots.txt	Recommended setting for a site that wants to rank
OAI-SearchBot	Whether the site can be shown in ChatGPT search answers	Yes, with about 24 hours to take effect	Allow, and allow the published IP ranges at the firewall
GPTBot	Whether the content may be used to train OpenAI foundation models	Yes	A separate commercial decision; blocking it does not affect search
ChatGPT-User	Nothing about search eligibility; fetches a page when a user’s question sends ChatGPT there	Rules may not apply, because a person initiated the fetch	Allow, so user-initiated fetches resolve
OAI-AdsBot	Whether a page submitted as a ChatGPT ad passes policy review, and when the ad is relevant	Visits only pages submitted as ads	Relevant only if you buy ChatGPT ads

Source: OpenAI crawler documentation, reviewed September 23, 2026. “Obeys robots.txt” describes OpenAI’s published behaviour, not an independently tested result on any specific site.

The practical trap is the second row. Many publishers blocked GPTBot in 2023 and 2024 to keep archives out of model training, and a few blocked every OpenAI agent with one wildcard rule, which quietly removed the site from ChatGPT search answers. Canlah AI’s policy at canlah.ai/robots.txt allows OAI-SearchBot, ChatGPT-User and GPTBot with “Allow: /” and disallows only API and file-download paths.

Does ChatGPT still rank websites from the Bing index?

ChatGPT no longer ranks websites from the Bing index alone: Peec AI’s September 4, 2026 investigation reports four result_source values in ChatGPT’s server-side events, led by OpenAI’s own Labrador index. Bing was the single gate when ChatGPT search launched, so a Bing-only plan in 2026 covers only part of the system. The field appeared between May 21 and July 21, 2026 with exactly four values: Labrador, Bright, Oxylabs and SERP.

Labrador is OpenAI’s own index in that account, and not one index but a family of roughly twelve vertical ones covering general web, PDF, YouTube, news, arXiv, Wikipedia, local, finance, legal, medical, shopping and images. The other three values are external scraping providers. The same investigation reports that ChatGPT also calls licensed sources such as Yelp and TripAdvisor alongside Microsoft’s Web IQ grounding product. It cites Nick Turley’s antitrust testimony that OpenAI began building its own index in 2023.

Keeping a site indexable in Bing is still cheap and still worth doing. Verify the domain in Bing Webmaster Tools, submit the XML sitemap and use IndexNow, the free open protocol that notifies participating search engines the moment a page is added, updated or removed. What has changed is the weighting: in 2026 Bing indexation is an afternoon of hygiene, not the strategy. The wider evidence on which corpus feeds which surface is in does ChatGPT use Google or Bing.

How to rank on ChatGPT, step by step

Ranking on ChatGPT takes seven steps in dependency order, because content work on a page ChatGPT cannot fetch is wasted work.

Step 1: Allow OAI-SearchBot, then check nothing downstream blocks it

Open yourdomain.com/robots.txt and look for a group naming OAI-SearchBot. No group usually means the wildcard rules apply; a group with “Disallow: /” removes the site from ChatGPT search answers. Add an explicit “Allow: /” group, then wait about 24 hours.

A clean robots.txt can still sit behind a closed door. Cloudflare began blocking AI crawlers by default for new domains in July 2025, and Akamai and Fastly ship bot controls that run separately from robots.txt. Filter server logs for the OAI-SearchBot user-agent string and look for 403 or challenge responses. If no OAI-SearchBot request appears in 30 days of logs, access is the constraint.

Step 2: Confirm the pages are indexed where ChatGPT can reach them

Verify the site in Bing Webmaster Tools, submit the sitemap and check that key pages return as indexed. Add IndexNow so new and updated pages are announced rather than waited for. Keep Google Search Console coverage healthy in parallel, because two of the four result_source values in the Peec AI account route through providers that scrape Google results.

Step 3: Put the answer in the first 60 words of every section

ChatGPT retrieves and extracts passages rather than reading whole pages. Lead each section with a self-contained answer to the question in its heading, then add the detail. Canlah AI’s measurement of 92 cited pages in September 2026 found the copied sentence was almost always the first sentence of a paragraph, and about 73% of copied facts sat in the first 30% of the page. Write every number exactly: “39 of 39 branded answers on September 7, 2026” is quotable, “strong brand recognition” is not.

Step 4: Make the entity unambiguous across every source

A model unsure who you are will not shortlist you. Keep the legal name, trading name, category, address, phone number and founding year identical on the website, in Organization schema and on every directory listing. Publish a plain facts page, such as canlah.ai/ai-info, that states those facts in sentences rather than in a design.

Step 5: Earn mentions on the third-party pages ChatGPT already retrieves

Run your buyer questions in ChatGPT and list every domain in the source rail. Those pages are the shortlist ChatGPT already trusts for the category. In the Canlah AI self-audit archive of September 7, 2026, ChatGPT recorded 785 citations across 149 domains for 116 answers, and the heaviest were developers.google.com at 57, clutch.co at 30, aistudio.com.sg at 28 and oom.com.sg at 25. Two of those four are a B2B directory and a competitor’s service page, which no amount of blogging on your own domain will fix.

Accurate, disclosed coverage on those specific pages, through directory profiles, review platforms, expert quotes and category listicles, moves open-question visibility faster than another page on your own domain.

Step 6: Show recency the retrieval layer can verify

Publish a visible “last updated” date, keep dateModified in Article markup in sync with it and refresh statistics on a fixed schedule, such as every 90 days. The Peec AI account reports that ChatGPT caches pages rather than fetching them live for every answer, which makes a stale cached copy a real cost.

Step 7: Skip the levers with no evidence behind them

Some popular fixes have no confirmed effect on ChatGPT retrieval. llms.txt is one: no major answer engine has confirmed using it, so it is a cheap file to publish and not a reason to be retrieved. Buying links and spinning thin pages on one topic belong in the same group.

ChatGPT ranking factors and how to verify each one

Eight ChatGPT ranking factors can be checked without a paid tool: three are documented by OpenAI or infrastructure vendors, three were measured by Canlah AI in September 2026, one is reported by Peec AI and one is inferred.

ChatGPT ranking factors in dependency order, with the evidence behind each, reviewed September 23, 2026.

Ranking factor	What it affects	How to verify it yourself	Evidence strength
OAI-SearchBot access	Whether the site is eligible at all	robots.txt group plus OAI-SearchBot requests in server logs	Documented by OpenAI
CDN and firewall rules	Whether allowed requests actually resolve	403 and challenge responses to the OAI-SearchBot user agent	Documented by Cloudflare, Akamai and Fastly
Index membership	Whether the page can be retrieved	Bing Webmaster Tools indexation plus Google Search Console coverage	Documented by both tools; the mapping to ChatGPT is inferred
Answer-first passages	Whether a passage is extractable	First 60 words of each section read alone	Measured on 92 cited pages, September 2026
Exact figures and dates	Whether a fact is quotable	Every price, count and date stated precisely	Measured on 92 cited pages, September 2026
Entity consistency	Whether the model is confident who you are	Name, category, address and claims compared across site, schema and directories	Inferred from answer behaviour, not documented by OpenAI
Third-party mentions	Whether open buyer questions reach you	Every domain in the source rail for your frozen panel	Measured: 785 citations across 149 domains, September 7, 2026
Recency signals	Whether a cached copy stays current	Visible update date matched to dateModified	Reported by Peec AI, September 4, 2026

Source: OpenAI and vendor documentation for rows one to three; the Canlah AI citation study of September 2026 for rows four and five; the Canlah AI self-audit of September 7, 2026 for row seven; the Peec AI investigation of September 4, 2026 for row eight. “Inferred” means the factor was not confirmed by OpenAI and is read from observed answers.

How to interpret five common results

A first ChatGPT panel can be read as one of five results, and each one points to a different gate and a different fix.

Never retrieved on any question: Check Steps 1 and 2 first, because a blocked crawler and an unindexed site produce exactly this result. If OAI-SearchBot appears in the logs with 200 responses and the pages are indexed, the constraint is further down.

Retrieved only when the question names your brand: This is the most common false positive. Asking ChatGPT “what is [your company]” measures retrieval, because the question already contains the answer. Canlah AI’s September 7, 2026 self-audit produced 100.0% brand mentions on 39 branded questions and 4.0% on 177 open ones from the same site on the same day.

Retrieved but described inaccurately: Prioritise correction over more exposure. Align the site, structured data and external profiles on the same approved facts, then rerun the exact question that produced the error.

Present in the answer text with no link to your domain: ChatGPT knows the entity but takes the evidence from someone else’s page. Check which third-party domains carry that answer and whether your own page states the same fact in its first lines.

Visible in ChatGPT and absent from Google AI Overviews: Treat this as a corpus difference rather than a fault. In the same self-audit, 15 domains out of 165 appeared in both the ChatGPT and the Google AI Overviews source pools, so the two surfaces need separate panels and separate fixes.

What ranking on ChatGPT cannot promise

Ranking on ChatGPT has five limits that apply to every method in this guide, including Canlah AI’s managed service.

• It cannot guarantee that a page appears for any given question, because OpenAI controls retrieval and selection.
• It cannot prove that one content change caused a new appearance without a frozen panel and dated repeat runs. Before-and-after movement is association, not causation.
• It cannot hold a position through an index refresh, a model release or a change in the buyer’s location and account state.
• It cannot be read from one answer, because repeat runs of the same question return different shortlists.
• It cannot be bought: there is no organic placement product, and ChatGPT ads are a separate labelled surface served through OAI-AdsBot.

When to move from a manual panel to ongoing monitoring

A manual panel becomes hard to sustain above roughly ten questions on a monthly cadence, or once several engines, markets or languages are in scope. Ongoing monitoring is justified in three cases: ChatGPT research influences a real buying decision, inaccurate answers create legal or reputational risk or several teams publish facts that have to stay consistent.

Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Its GEO service is priced per AI platform: Starter from $1,500 a month (ChatGPT, 4 buyer questions), Growth from $3,000 a month (adds Google AI Overviews, 8 buyer questions) and Full customised (adds Google AI Mode and Gemini, 16 buyer questions), each re-measured monthly. It starts with a free AI visibility report within 48 hours.

Canlah AI’s main limitation for this job is evidence depth. Its September 7, 2026 self-audit sampled OpenAI and Gemini at the API layer rather than the consumer ChatGPT interface, so those figures illustrate the branded trap rather than a consumer-surface ranking rate.

Do not choose a tool or an agency from this guide alone. Ask each shortlisted vendor to rerun your own frozen ChatGPT panel in front of you, with prompts, dates, answers and cited URLs visible, and compare the result against your own spreadsheet before you sign.

Frequently asked questions

How do I rank on ChatGPT?

Ranking on ChatGPT starts with allowing OAI-SearchBot in robots.txt and at the firewall, because OpenAI states that an opted-out site will not be shown in ChatGPT search answers. Next, publish pages that answer one buyer question in their first 60 words. Then earn accurate mentions on the directories, review sites and listicles ChatGPT already retrieves for your category, which is usually the binding constraint.

Can I pay to rank higher in ChatGPT search results?

No. There is no product that places a website inside an organic ChatGPT answer, and any vendor guaranteeing a position is promising something it does not control. ChatGPT ads are a separate labelled surface whose landing pages are reviewed by OAI-AdsBot, and they do not change organic retrieval.

Does ChatGPT rank websites from the Bing index?

ChatGPT uses Bing as one input among several in 2026 rather than the whole pipeline. Peec AI’s September 4, 2026 investigation reports OpenAI’s own index family alongside three external providers, licensed sources and Microsoft’s Web IQ. Keep Bing indexation and IndexNow working as cheap hygiene, and do not build the plan on Bing alone.

How long does it take to rank on ChatGPT after fixing crawler access?

OpenAI states that a robots.txt change takes about 24 hours to register, so an access fix can show in the next panel round. Content and third-party coverage changes take longer and vary by question. Track the same frozen panel monthly and report appearance frequency rather than a single screenshot.

Should I block GPTBot to protect my content from training?

Blocking GPTBot is a separate decision from ChatGPT search, and OpenAI documents the two settings as independent. Blocking GPTBot signals that your content should not train foundation models; it does not remove the site from ChatGPT search answers. The mistake to avoid is a wildcard rule that blocks every OpenAI agent, because it also removes OAI-SearchBot.

Does Canlah AI track ChatGPT rankings?

Yes. Canlah AI tracks ChatGPT visibility as prompt-level appearance frequency rather than a numbered position, because ChatGPT returns no ranking number. Its retainers run a fixed pool of buyer intents, re-measured monthly, and report ranges with timestamped evidence. Confirm the buyer-intent count and the engine list in the proposal: Starter covers ChatGPT, Growth adds Google AI Overviews and Full adds Google AI Mode and Gemini.

Method and sources

This guide was written on September 23, 2026. Crawler behaviour was taken from OpenAI’s published documentation; retrieval architecture from a third-party investigation reported here as that vendor’s finding rather than as confirmed OpenAI behaviour; indexing mechanics from Bing Webmaster Tools. No paid accounts or vendor briefings were used. The Canlah AI figures are API-layer samples of OpenAI and Gemini from September 7, 2026: a single-day sample of one brand in one category, not a stable visibility rate and not proof of causation.

• OpenAI crawler documentation. The four user-agent roles, independent settings and the 24-hour robots.txt note.
• Peec AI, ChatGPT built its own search index, September 4, 2026. The result_source field, the Labrador index family, licensed sources and Web IQ.
• Bing Webmaster Tools IndexNow. Protocol for announcing added, updated and removed pages.
• canlah.ai robots.txt. Worked example of explicit allow rules.
• canlah.ai/ai-info. Worked example of a plain facts page.
• Canlah AI pricing. Per-platform GEO service prices, the free AI visibility report, buyer-question counts, monthly cadence and AI platforms by tier.
• Canlah AI self-audit evidence archive, anonymised, September 7, 2026. Mention counts, 785 ChatGPT citations across 149 domains and the 15 of 165 domain overlap with Google AI Overviews.
• Canlah AI citation study, September 2026. Sixty buyer queries across five categories, 526 citations and 92 cited pages; used for the first-sentence and first-30% findings.

See where your brand stands in AI answers. Run the free AI visibility audit →

Related articles

• Why is my business not showing up in ChatGPT?
• How to check if AI search recommends your brand
• How AI engines choose which sources to cite
• How to get cited by Perplexity AI