Why isn't my site showing up in AI answers?

Enter a domain. This reads the live site the way an AI crawler does and tells you what is actually blocking you, one check at a time.

Reads your homepage, robots.txt, llms.txt and sitemap.xml. Free. No signup. Nothing stored.

Want the version that keeps watching?

This is a snapshot of one page. We track presence across ChatGPT, Claude, Gemini, Perplexity and AI Overviews over time, and build the content that closes the gaps.

Start a conversation

Four reasons a good site stays invisible

Ranking well in Google and being quoted in an AI answer are two different jobs. A site can do the first beautifully and fail the second completely, which is why the usual SEO report comes back clean while the answers never mention you.

1. The crawler never arrived

This is the most common cause and the least discussed. A line in robots.txt, often added years ago or inherited from a template, tells GPTBot or ClaudeBot or PerplexityBot to stay out. Everything downstream becomes irrelevant. It is worth knowing that these crawlers do different jobs: GPTBot gathers training data, while OAI-SearchBot fetches pages to answer a live question. Blocking the second removes you from ChatGPT search results today. Blocking the first affects what the model learns over time. Plenty of sites block both without ever deciding to.

2. The page is empty until JavaScript runs

Most AI crawlers take the raw HTML and leave. They do not wait for a framework to hydrate. If your copy is assembled in the browser, the crawler sees a shell: a title, a script tag, and nothing to quote. The check above counts the words present in the source itself, which is the only text a model can work with.

A model cannot cite a sentence it never received.

3. Nothing tells the model what you are

Search engines infer. Language models need the entity spelled out. Organization schema, a canonical that agrees with the title, sameAs links tying the name to profiles the model already trusts: these are the difference between a company the model can describe and a string of characters it has seen somewhere. Without them you can be indexed and still be unnameable.

4. The content has no extractable passage

Models pull passages, not pages. Long unbroken prose gives them nothing to lift. A question as a heading, a direct answer in the two sentences below it, a list where a list belongs: this is what makes a page quotable. It also happens to make it more readable, which is the rare case where the optimization and the craft point the same direction.

What this check does not tell you

Honesty about a diagnostic's limits is part of the diagnostic. This tool reads one page plus three files. It reports what it can verify and marks the rest as unverified rather than guessing, because a confident number built on a failed fetch is worse than no number.

It does not measure whether you actually appear in answers. That takes running real prompts across platforms repeatedly and watching the mention rate. This checks readiness, which is upstream of that.
It reads your homepage, not your whole site. A strong homepage and thin interior pages will score better here than it deserves.
It cannot see what other sites say about you. In most categories the answers are assembled largely from third-party sources: community threads, publisher roundups, video. Your own site is one input among many.
It does not weight llms.txt heavily. No major platform has committed to reading it. It is a cheap bet worth taking, not a fix worth celebrating.

If you want the part this cannot do, that is the work: measuring non-branded mention rate over time, per platform, and building toward the sources the models already read. That is what we do.

Common questions

In most cases it is one of four things: an AI crawler is blocked in robots.txt, the content only exists after JavaScript runs so the crawler sees an empty page, there are no structured entity signals telling the model what the company is, or the content is written as prose that offers no extractable passage. Ranking in Google is not the same job as being quoted in an answer, and a site can do the first well while failing the second completely.

GPTBot is the training crawler and OAI-SearchBot is the one that fetches pages to answer live queries. Blocking OAI-SearchBot removes you from ChatGPT search results. Blocking GPTBot keeps you out of training data but does not by itself remove you from live search answers. Many sites block both without realizing they are separate decisions.

No major AI platform has committed to reading llms.txt as a ranking or retrieval input, so treat it as a low-cost bet rather than a fix. It is cheap to publish and it forces a useful exercise: writing down what your company is in plain language. The checks in this tool that carry real weight are crawler access, server-rendered content, and structured data.

Track mention rate on non-branded prompts, separately from branded ones. Branded prompts sit near a ceiling for almost every company, so blending them into one number inflates it and hides whether anything is working. Measure the rate at which you appear in answers to questions that do not contain your name, and track citation rate separately, since being cited and being mentioned are different events.

Crawler access changes can show up within days because live-retrieval systems refetch often. Entity and structured data changes typically take weeks to propagate. Content that depends on being cited by other sites takes longest, because it waits on the sources those models already read rather than on your own publishing schedule.

No. Every check runs in your browser. The page fetches your public homepage, robots.txt, llms.txt and sitemap.xml through a public read-only proxy, scores what comes back, and forgets it when you close the tab. There is no account and no backend.