Why Your Website Is Not Showing Up in ChatGPT Search

If your website is not showing up in ChatGPT Search, start by identifying what is missing: a brand mention, a linked citation, or an answer to a specific question. Then check whether the relevant page can be discovered, fetched and read. Only after those checks should you decide whether to improve an existing page or publish a new one.

A newly published website may have few observed citations. Its age alone does not explain the result. A blocked search crawler, incomplete answer, small test panel or different search context can produce the same symptom. This guide gives you an order for investigating those possibilities.

Check these four things first when your website is not in ChatGPT search

These are the checks for OAI-SearchBot, the crawler OpenAI uses for ChatGPT search, read from OpenAI’s crawler documentation on 12 October 2026. The sections after this one go deeper.

1. Is OAI-SearchBot allowed in robots.txt? Look for a group that names OAI-SearchBot, and for a User-agent: * group that disallows the page. OpenAI says that a site that opts out of OAI-SearchBot is not shown in ChatGPT search answers. Blocking GPTBot does not do this: it controls training use, and the two tokens are set separately.
2. Does your firewall or CDN refuse OAI-SearchBot? A permissive robots.txt does not stop a bot challenge or a 403 at the edge. Check that real requests from OpenAI’s published address list, searchbot.json, reach the page. A request from your own machine with a copied user-agent string is not that test (verifying crawlers, grade A).
3. Did you change robots.txt in the last day? OpenAI says it can take about 24 hours from a robots.txt update for its systems to adjust for search results. Wait that long before you test again, or you will read the old state.
4. Is your site appearing only as a link? OpenAI says a site that has opted out of OAI-SearchBot can still appear as a navigational link. A bare link with no quoted answer is therefore what a blocked site can look like; it is also what a site that is allowed but not chosen can look like, so use checks 1 and 2 to tell them apart.

If all four are clear, the next causes are discovery, readable content and whether the page answers the question, which sections 2 to 5 below take in order.

First, define what “not showing up” means

These outcomes need different fixes:

Observation	What it establishes	Next check
The answer names your brand without a link	A mention, without a citation to your website	Is there a useful page that substantiates the answer?
The answer cites another website about your brand	Third-party coverage is being used	Compare the cited page with your own answer
A search trace includes your URL, but the answer does not cite it	Retrieval without citation, where the trace is available	Relevance, completeness and evidence in the passage
The answer neither names nor cites your website	Absence in this particular answer	Access, discovery and question coverage
Search was not used, or the request failed	No usable observation of search visibility	Repeat with a documented search-enabled context

Do not treat a single answer as a site-wide verdict. Record the exact question, date, language, country and search surface before changing anything. The consumer ChatGPT experience and a search-enabled measurement interface are different test contexts; keep their results separate.

1. Check the search crawler, separately from training access

OpenAI distinguishes OAI-SearchBot, used for search, from GPTBot, used for potential model-training data, and ChatGPT-User, used for user-initiated visits. Their controls serve different purposes. Allowing a training crawler is not a requirement for opting into search. OpenAI also notes that a blocked site can still appear as a navigational link. See the official OpenAI crawler documentation.

Check the relevant page and its dependencies, then:

1. Read the rules that apply to OAI-SearchBot in robots.txt, including wildcard rules and path-specific exclusions.
2. Inspect firewall and CDN events for blocked verified search-crawler requests. A permissive robots rule does not bypass a bot challenge.
3. Confirm the response contains the intended page rather than a login screen, consent interstitial, soft error or empty shell.
4. Review status codes, redirects and response timing for representative pages in each important template.

A request made with a copied user-agent string is a useful diagnostic, but it does not prove that a verified vendor crawler receives the same response. Use server logs and the vendor’s published verification guidance where available.

For a fuller checklist, see AI crawler access and allow-listing and crawler identities.

2. Check discovery and indexing without assuming one engine speaks for another

Make the page easy to find from your own website. Link to it from an appropriate hub or related article, include the canonical URL in your sitemap and check that redirects do not lead to a different page. Investigate conflicting canonicals or accidental noindex directives before producing more content.

Search Console can help diagnose Google indexing. It does not verify ChatGPT retrieval. A page appearing in a conventional search index and a page being selected as an answer source are separate observations.

For Google AI features specifically, Google says existing SEO practices apply: a page must be indexed and eligible to appear with a snippet. Eligibility does not guarantee inclusion. See Google’s guidance for AI features and your website.

Keep a short evidence log for each target URL: canonical, response status, indexing observation, internal links and last material change. This makes the next measurement interpretable.

3. Check whether the answer survives without JavaScript

Open the page source and look for the actual answer: the definition, comparison, instructions or evidence the question requires. A title and navigation menu are not enough. Then compare the initial response with the rendered page under consistent conditions.

If a product comparison appears only after a client-side request, move its essential content into server-rendered or pre-rendered HTML. Keep interactive filtering if it helps visitors, but make the useful answer available on the canonical page before interaction.

This is a resilience check, not a universal statement about every AI fetcher. Our guide to testing JavaScript-dependent content explains how to separate a rendering gap from a crawler-access gap.

4. Match the page to the exact question

Take a question you care about and identify the single page that should answer it. Read that page without relying on other tabs or internal product knowledge.

Ask whether the page provides:

• A direct answer near the relevant heading, followed by useful detail.
• The conditions and limits that would change the recommendation.
• Evidence appropriate to the claim: vendor documentation, a reproducible method, or a dated observation.
• A clear author, publisher and update history.
• Descriptive links to the next step, such as a tutorial or product capability.

For example, “How do I track AI citations?” requires definitions, a measurement method and a way to record results. A landing page saying “track your visibility” only partly answers it. Our citation tracking guide covers that workflow.

Use the source-content workflow to choose the canonical owner and substantiate its answer. Prefer expanding the existing canonical page when it already serves the intent. Publish a separate article when the reader has a distinct task that deserves its own answer. Creating several pages with the same answer makes ownership and internal linking harder to maintain.

5. Compare the sources that are being cited

Save the source URLs from the answer and read them. Identify what each contributes: a definition, current specification, comparison table, first-hand observation or clear procedure.

Then document the missing contribution on your own page. “Add an explanation of citation rate with a denominator and worked example” is an actionable change. “Make the article more AI friendly” is not.

This comparison generates hypotheses, not proof of a ranking rule. A competitor’s cited page may differ in many ways, and answers can change without a website change. Test a specific improvement and retain the original observation.

6. Measure again with the same questions

Create a small, fixed panel covering branded questions, category discovery and practical problems. Record eligible answers, failed requests, citations, mentions and the exact cited URLs. Repeat the same wording and context after a material change, while keeping a baseline history.

If you expand the panel, report the new cohort separately before merging its trend. Adding harder non-branded questions can lower an aggregate rate even when the original questions improve. How to track AI citations and brand mentions explains these denominators.

Vercatus connects those two sides of the investigation: Why shows the pages engines read and the searches they run, and Answer Share records observed citations and mentions. A content-coverage score is not evidence that an engine actually cited the page.

Common questions

How long does it take a new website to appear in ChatGPT Search?

There is no dependable site-wide waiting period you can infer from one missing citation. Confirm search access and discovery, establish a baseline and remeasure. Do not promise a citation date from the page’s age.

Does allowing GPTBot make my website appear in ChatGPT?

GPTBot and OAI-SearchBot have different purposes. Review search access independently from your training-data policy using OpenAI’s documentation linked above.

Does being indexed guarantee a citation?

No. Discovery and access make a page a possible source. The answer still has to use and attribute it for a particular question. Track those outcomes separately.

Should I rewrite every page if citation numbers are low?

The evidence behind this

• An engine can only use what its fetchers can retrieve. OpenAI states that sites opted out of OAI-SearchBot are not shown in ChatGPT search answers, though they can still appear as navigational links; Anthropic says blocking its search or user agents may reduce visibility in Claude; Google asks that crawling be allowed both in robots.txt and by any CDN or hosting infrastructure (crawler access, grade A).
• A challenge page is not your page. Cloudflare’s JavaScript challenge shows a crawler that is not exempted a challenge page instead of the content (JavaScript challenges, grade A).
• Content added by script may be missing. In a December 2024 analysis none of the major AI crawlers measured rendered JavaScript (HTML only, grade C).

First check how many questions were actually asked and which pages should answer them. Fix demonstrable access or rendering failures, then improve the highest-value answer gaps. A low result from a small panel is a reason to investigate, not evidence that the entire site needs rewriting.