Why Your Business Doesn't Show Up in ChatGPT — a Diagnostic Guide

Most advice on AI search visibility skips this step and goes straight to optimisation, which is why so much of it disappoints: it treats a firewall problem as a content problem. This guide is about working out which situation you're in. If you want to know how to structure a page so an AI system can extract an answer from it, that's a different question, covered in How to Get Cited by AI Answer Engines.

The four modes

Work through them in order. There's little point tuning content for mode 3 while mode 1 is still true.

Mode 1: The crawler is blocked

For search-based AI systems, crawler access is a necessary part of making web content available. If access is refused, nothing downstream applies.

The common mistake is blocking the wrong bot. OpenAI runs several crawlers with different documented jobs:

• OAI-SearchBot — "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features."
• GPTBot — "used to crawl content that may be used in training our generative AI foundation models."
• ChatGPT-User — used when a person asks ChatGPT something and it fetches a page. OpenAI states: "ChatGPT-User is not used to determine whether content may appear in Search."

That distinction matters. Many sites blocked GPTBot during the wave of concern about AI training — a legitimate decision. But OpenAI describes GPTBot only in training terms and nowhere says blocking it removes you from ChatGPT search. The documented search lever is OAI-SearchBot: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links."

What to check

1. 01Read yourdomain.com/robots.txt properly. Look for OAI-SearchBot, PerplexityBot and Claude-SearchBot — and note that a blanket Disallow against a wildcard user-agent catches them too.
2. 02Check your firewall and CDN. OpenAI names two separate requirements: allow the crawler in robots.txt, and "confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." Those ranges are published at openai.com/searchbot.json.
3. 03Check server logs for the crawler user-agent strings — they can provide direct evidence that a crawler request reached your infrastructure.

What this establishes: refused access is visible directly; a fetch is not proof anything was used.

What to do next: fix access and stop there — allow the crawler, allowlist the published IP ranges, then wait for a re-crawl before changing anything else, or you lose the ability to tell what worked. This is the cheapest check to perform first.

Mode 2: Accessible, but not eligible or available for retrieval

The crawler gets in; the pages still aren't eligible.

For Google's AI features the documented bar is reassuringly ordinary: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements." There is no AI-specific hurdle — "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

So what breaks eligibility is ordinary technical SEO: pages not indexed, noindex left on after a migration, snippet restrictions applied for other reasons, or a site-level opt-out someone set and forgot.

What to check

1. 01Confirm the page is indexed. If it isn't in ordinary search, it isn't eligible for the AI features built on top of search.
2. 02Check for snippet-limiting directives — nosnippet, data-nosnippet, max-snippet.
3. 03Open the generative AI performance report in Google Search Console — "As of August 31, 2026, we've rolled out these insights to all websites worldwide." If it shows nothing, Google's help text offers one explanation to rule out: your site may have been excluded from Search generative AI features.

What this establishes: eligibility, and whether Google's AI features surface links to your site — nothing about ChatGPT, and not which queries produced the impressions.

What to do next: these are ordinary indexing and crawl-hygiene problems wearing a new hat; if your diagnosis lands here, a technical SEO audit is the right response.

Mode 3: Retrieved, but not cited

You're eligible. Something else is being used instead. This is where most people start, and it should be the third thing you check.

It's also where honest advice runs out fastest. OpenAI's entire published account of how sources are selected is: "ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." No factor is named; Google names no AI-specific ranking factors either. Every confidently-listed "AI ranking factor" you've read is somebody's inference.

What to check

Ask the assistants directly, and sample properly: these systems are non-deterministic, so the same prompt can return different sources on different runs. Run each prompt three to five times, logged out, recording the prompt, date, sources cited, and whether you appeared. A single run is a weak basis for a conclusion; several runs give a directional sample, though still not a reliable citation rate.

In Bing Webmaster Tools, the AI Performance report (in public preview since February 2026) shows where your pages were used as sources. One caution: its grounding queries are not what users typed — Microsoft defines them as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." Microsoft is also explicit that the data "reflects how often pages are cited, not page importance, ranking, or placement."

What to do next: this is where content work is genuinely the answer, and where ongoing organic search work matters — becoming a source is built, not configured.

Mode 4: Cited, but mis-attributed

You're being used, and it isn't helping — a wrong company name, a link to a page that has moved, or your material credited to a syndication partner. This mode is easy to miss because it can look like success in a citation count while still failing to produce a useful referral.

What to check: in your sampling runs, read the attribution, not just whether you appeared. Is the business name right, and does the link resolve? A citation pointing at a 404 or a stale redirect is a broken referral. Check too that your name and description are consistent across the places that describe you.

What this establishes: specific, fixable errors — not how often they occur, since you can't see the population of answers you're cited in. OpenAI notes that "Search results and citations can be incomplete, outdated, or incorrect," so some mis-attribution isn't fixable from your side.

What to do next: fix URLs and redirects, tighten naming consistency where you control it, then let the rest go.

What the free tools give you

Neither documents a click or traffic metric, so a citation count is not a traffic proxy.

Two pieces of common advice the evidence doesn't support

llms.txt

It's a proposal, not a standard — published by Jeremy Howard of Answer.AI on 3 September 2024. Two years on, no major platform documents reading it. Google Search Advocate John Mueller, asked on Bluesky in January 2026 whether Search uses it, answered: "to be direct, no." That is a spokesperson comment, not documentation. Ahrefs found that across 137,210 domains' server logs, 97% of published llms.txt files "received zero traffic in May 2026. Nothing fetched them at all." Google's documentation agrees: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." Publishing one is inexpensive, but there is currently no documented reason to treat it as a visibility requirement.

"Rank well in Google and you'll get cited"

A statistic still circulating holds that around 99% of AI citations come from top-ranking organic results. It traces to seoClarity research reported in August 2024, which found AI Overview citations "match one or more of the webpages from the top 10 Google organic search results 99.5% of the time" — a measure of whether at least one citation on a results page also ranked in the top ten, not what share of cited URLs rank well.

Later measurement points elsewhere. Ahrefs, analysing 4 million AI Overview URLs in March 2026, found 38% of cited pages rank in the top 10, down from 76% in July 2025. A separate Ahrefs study of 15,000 prompts in August 2025 found only 12% of links cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same prompt — an average covering those three platforms only. Perplexity was measured in the same study and reported separately, at 28.6%. That gap is the lesson: these figures measure how often cited links also appear in Google's top 10 for a sample of prompts. They are not citation rates, traffic, visibility scores or the odds of being cited, and a number from three assistants does not describe a fourth. Organic ranking and AI citation do not always move together: good rankings still help, but they are not the whole explanation.

What to do next

1. 01Check access first. robots.txt, then CDN and firewall against the published IP ranges, then server logs.
2. 02Check eligibility second. Indexing, snippet directives, opt-out flags, then the generative AI performance report.
3. 03Only then look at content — sampling properly, three to five logged-out runs per prompt.
4. 04Re-check quarterly. Both platform reports arrived within the last seven months, and the citation-versus-ranking picture shifted measurably in under a year.

If the diagnosis lands on access or eligibility, that's technical SEO work and it's solvable. If it lands on content and competitive position, that's a longer programme with no guaranteed outcome — and anyone who promises otherwise is promising what the platforms themselves won't.

We do this work through our organic search service, covering the initial technical audit and ongoing search work, and through managed SEO/AEO specialists for teams who want the capability in-house. We don't operate proprietary AI-visibility tracking software — the tools above are the platforms' own and free, and we'd rather tell you where they stop than sell you a dashboard over the top.

This guide reflects platform documentation as published on 11 September 2026. Both platform reports described here are recent and remain subject to change; we review this guide quarterly.