If your startup is not showing up in ChatGPT, the most likely explanation is that the answer did not retrieve or select a clear, accessible source about you. It may be a crawl issue, a weak explanation on your own site, too few independent mentions, or simply a different answer on that run. You cannot force a recommendation. You can make your facts easier to find, verify, and cite.
Key takeaways
- Test exact questions before changing pages. A brand-name lookup and a category recommendation are different problems.
- Let the search crawlers you want to reach access your important public pages, while keeping training controls separate.
- Make your homepage say plainly what you sell, who it is for, and how it differs.
- Earn accurate third-party mentions; repeating your own claim on ten thin pages is not corroboration.
- Treat every appearance in an AI answer as variable, not as a permanent rank.
This article is a diagnostic checklist, not a claim about ChatGPT's hidden ranking system. For a broader framework, read the GEO guide for startups. For the specific execution steps after you find a problem, use the AI citation playbook.
1. Your brand is too new to have independent context
Why it happens: A recently launched site may have a homepage and little else on the open web. When an AI search experience looks for corroboration, it may find a generic description, a social account with no details, or several similarly named products. That does not mean the startup is blacklisted. It means a user asking for category recommendations may be presented with better-documented alternatives.
The fix: Publish a concise product page with a clear problem, audience, use case, and public URL. Then pursue a few relevant, accurate mentions from communities, partner pages, product roundups, or reviewed launch profiles. Ask the publisher to describe what the product actually does rather than copy your entire homepage. A free launch platforms shortlist can help you choose plausible places; do not submit to dozens of unrelated sites merely to create noise.
You should also separate branded and unbranded tests. “What is Acme Analytics?” asks whether a system can find your entity. “Best analytics tools for a small SaaS” asks it to choose among competitors. Solving the first does not automatically solve the second.
2. A crawler or security layer cannot read the site
Why it happens: A robots.txt rule, noindex tag, login wall, or aggressive WAF can stop a retrieval system before it reaches your content. OpenAI's crawler documentation identifies OAI-SearchBot for ChatGPT search and GPTBot for model training. They are independent controls; allowing one does not require allowing the other. Perplexity similarly documents PerplexityBot for its search results. For Google AI Overviews, Google's guidance says ordinary Google Search crawl and snippet eligibility apply.
The fix: Check your public robots.txt, page-level meta robots tag, HTTP status, CDN rules, and server logs. Allow the search crawler relevant to your goal and make the main product pages publicly accessible. Do not copy a broad “allow all AI bots” snippet without understanding your training preference. A successful browser visit by you does not prove a crawler can access the same page.
3. Your homepage could describe almost any startup
Why it happens: “Transform your workflow with AI-powered insights” gives a retrieval system almost no usable facts. It is also not persuasive to a human buyer. When several tools use similar language, your product has little distinctive text that an answer can cite.
The fix: Replace the vague first screen with one sentence naming the product category, intended user, and task. Add a concrete example of what a user does and what output they get. Explain limitations or eligibility where they matter. Put the explanation in visible HTML rather than only in a video or image. A page about a task should answer that task near the top, not hide the answer behind a lead form.
You do not need to repeat the exact phrase “best tool” throughout the site. You need clear language that makes the product relevant to a question someone might actually ask.
4. Basic entity facts are missing or inconsistent
Why it happens: AI-generated answers can confuse brands with similar names. A product described one way on its site, another way in an app store, and a third way in a directory gives readers conflicting evidence. A changed domain or renamed product can make the problem worse.
The fix: Use one public brand name, canonical domain, short description, and current pricing/status across your homepage, about page, social profiles, and listings. State who operates the product when that helps establish identity. If you changed names, add an honest explanation and redirect old public URLs where appropriate. Structured data can clarify the same visible facts, but it should not invent claims that readers cannot see.
Keep a simple source-of-truth document for your team. Update it before submitting to another site. Consistency is not a secret AI trick; it prevents real users and publishers from repeating stale information.
5. You have no relevant third-party listings or references
Why it happens: Your own website is a primary source for what you offer, but an answer comparing options may need evidence beyond self-description. A credible category directory, integration partner, or editorial review can provide another path to your product. A random automated directory page with no content is much weaker evidence.
The fix: Choose a small number of places your potential customers already use. Give each a specific, accurate description and a working product URL. For a launch platform, answer comments and keep the listing current after launch. We run LaunchAF; a reviewed LaunchAF submission can be one profile in that mix, but it is neither required nor a guaranteed AI citation. Match the audience before you spend time on the form.
6. Important information exists only after JavaScript runs
Why it happens: A page that returns a nearly empty HTML shell, then fetches all meaningful text after complex client-side scripts run, can be harder for some crawlers and extractors to use. This is not a claim that every AI system fails on JavaScript. It is a practical reason to inspect what a non-interactive request receives.
The fix: Fetch the page with a simple HTTP client and inspect the returned HTML. Does it contain the product name, category, purpose, and important answer text? If not, use server rendering or static content for those essentials. Keep navigation links as ordinary anchors with href, not click handlers alone. Test again after deployment, not just in local development.
7. You have no page that directly answers the question
Why it happens: A broad homepage may be enough for a brand lookup but poor evidence for a narrow question. Someone asks “Does this tool support CSV export?” while the only answer is inside a support email or a screenshot. A search answer cannot confidently cite what is not stated on an accessible page.
The fix: Make a small set of task-oriented pages or a useful FAQ based on real customer questions. Put the direct answer in the first paragraph, then add details, constraints, and examples. Link to the relevant product or documentation page. Avoid publishing dozens of near-identical pages that swap one keyword; those are unlikely to help readers and can dilute your site. The GEO guide explains the difference between genuinely citable answers and generic copy.
8. Your brand name is ambiguous
Why it happens: A common-word product name, inconsistent spelling, or another company with the same name makes retrieval harder. Even if an AI tool recognizes the string, it may pick the wrong entity or omit the brand when it cannot tell which one a question refers to.
The fix: Pair the brand with a stable descriptor on key pages, such as “Acme, a scheduling tool for dental clinics,” and use the same canonical domain everywhere. Add an about page that clearly identifies the team and product. When you test prompts, include the domain once to see whether the system can find the right entity, then test a normal category query without the domain. That comparison tells you whether the problem is identity or selection.
9. Your information is stale, or the answer is variable
Why it happens: AI search systems do not return a fixed directory ordered the same way every time. Queries, location, account context, source availability, and the current web can change the answer. A discontinued feature, old pricing page, or outdated listing may also make a product look less relevant. One missing appearance is not proof that an optimization failed.
The fix: Update material changes on your canonical pages and the most important third-party profiles. Add a visible updated date when you actually revise information, not as decoration. Retest the same prompts over time and record whether your page was cited, merely mentioned, or omitted. Use the change log to identify obvious stale facts; do not rewrite good content after every variable answer.
A simple visibility self-test
Write five questions a real buyer might ask: a brand lookup, two category comparisons, one narrow feature question, and one question about an alternative. Ask the exact same wording in ChatGPT search, Perplexity, and Google Search where an AI Overview appears. Record the date, tool, exact prompt, whether the brand appeared, whether your domain was cited, and which competing sources were used. Repeat on another day.
The point is diagnosis, not a vanity score. If the brand lookup fails, check access and identity. If it succeeds but category questions do not, improve audience fit, useful answer pages, and independent references. If one tool cites you and another does not, avoid assuming that all AI products use the same sources or rules. The AI citation guide gives a two-week implementation plan after this baseline.
Frequently asked questions
Can I pay to make ChatGPT recommend my startup?
No legitimate submission or directory listing can guarantee an organic ChatGPT recommendation. You can pay for advertising where a product offers it, but that is a separate placement, not control over an answer. Treat anyone selling guaranteed citations with skepticism.
Does blocking GPTBot stop ChatGPT search citations?
OpenAI documents GPTBot as a training crawler and OAI-SearchBot as its search crawler. Their controls are separate, so blocking training does not by itself mean opting out of ChatGPT search. Check the current OpenAI crawler documentation and your own WAF behavior before changing rules.
How long does it take for a new startup to appear?
There is no published universal timeline or guarantee. Crawling, indexing, retrieval, and answer selection are separate steps and differ by service. Make the pages accessible, publish clear information, then monitor actual citations rather than promising yourself a deadline.
Will a LaunchAF listing make my startup show up in ChatGPT?
It can create an accurate public profile and another discovery path after approval, but it cannot force an AI answer. The free launch requires a verified badge and review, so it is not instant; keeping the badge matters after publication too. Decide whether the weekly founder audience and lasting profile are useful even without an AI mention.
Is llms.txt required?
No. Google says no special AI text file is required for AI Overviews, and there is no established guarantee that an llms.txt file changes ChatGPT or Perplexity citations. Prioritize accessible pages, clear answers, and trustworthy references first.
The bottom line
A missing ChatGPT mention is a diagnostic signal, not a verdict on your company. Run the five-question test, fix any access or identity problem, and publish answers worth citing. Then build a few relevant external profiles and monitor what actually changes, without expecting control over the generated answer.