When ChatGPT names your competitors and not you, the cause is almost always one of five mechanical failures: no page of yours answers the question, your page has nothing quotable, the third-party sources the AI trusts omit you, your site blocks AI crawlers, or the model answered from training data. Four of the five are fixable, and you can diagnose yours in about twenty minutes.
It's worth diagnosing, because these answers move real purchases: in G2's 2026 buyer survey, 69% chose a different vendor than planned based on AI chatbot guidance, and a third bought from a vendor they'd never heard of before the AI named it. The model doesn't dislike you, and it isn't ranking product quality. An AI answer is the output of a three-stage pipeline — reformulate, retrieve, synthesize — and you're losing at a specific stage.
Reason 1: There's no page for the model to find
The buyer asks "best payroll service for restaurants." Your competitor has a page about restaurant payroll. You have a homepage that says "Payroll, reimagined." Retrieval runs on search: if no page of yours substantively addresses the question, nothing downstream matters. Taglines don't answer questions; specific pages do. One question, one page.
Diagnosis: search Google for the buyer question. If your site is nowhere in the results, retrieval is your bottleneck. (You don't need to be #1 — the shortlist pulls from well beyond the top 10 these days — but the page has to exist and be findable.)
Reason 2: Your page exists but has nothing worth citing
You have the page, it even ranks, and the answer still skips you. The model read it and found nothing it could use: no numbers, no specifics, no verifiable claims. "We're the leading solution trusted by thousands" gives a language model nothing to build a sentence from. Your competitor's page saying "processes payroll for 2,300 restaurants, average setup 4 days, per a 2025 customer survey" hands it three ready-made facts — each arriving with your competitor's name attached.
Diagnosis: count the concrete, sourced, quotable facts on your page. If the answer is "two" and your competitor's page has fourteen, that's the whole mystery. The measured fixes are in How to get cited by ChatGPT.
Reason 3: Third-party sources decide, and you're absent from them
Watch what AI answers cite for "best X for Y" questions: very often not vendors at all, but comparison articles, roundups, Reddit threads, review sites. That's not anecdote. A University of Toronto study found AI search systematically favors earned media — third parties writing about you — over brand-owned content, far more than classic Google does. And in Semrush's study of 100M+ citations, community and reference sites dominate the most-cited domains on every engine.
If every roundup of your category omits you, the model repeating those roundups omits you too. Your competitor isn't beating you with their website; they're beating you with other people's websites.
Diagnosis: ask the question on Perplexity, which shows its sources plainly, and check each cited page for your name. The remedy is unglamorous: pitch the roundup authors, claim your review-site profiles, show up in the communities the engines keep citing.
Reason 4: You're blocking the crawlers
Surprisingly common and self-inflicted: your robots.txt or CDN blocks GPTBot or OAI-SearchBot — often a 2023-era "protect our content from AI" decision nobody revisited, or a bot-protection default. Blocked crawler, no retrieval, no citation.
Diagnosis: read yourdomain.com/robots.txt, check your CDN's bot settings, and confirm your pages render as real HTML without JavaScript. The full checklist — including which crawlers respect robots.txt and which don't — is in llms.txt & AI instructions.
Reason 5: The model answered from memory
ChatGPT runs live search on only about a third of queries; the rest come from training data, where frequency of historical mention is destiny and big old brands win by default. You can spot this case instantly: the answer has no citations.
You can't edit training data. But when retrieval fires, the field tilts toward whoever has the most citable page today — the founding GEO study measured its biggest gains for sources that weren't already on top (details). Win the cited answers you can influence now; every mention you earn is also a candidate for what future models learn.
Run the diagnosis in 20 minutes
Work top to bottom for every buyer question you lose. The first “no” is usually your bottleneck.
- Write down 5 questions your buyers actually ask.
- Ask each on ChatGPT (search enabled) and Perplexity. Note who's named and which URLs are cited.
- For each question you lose: Do you have a page answering it? (1) — Does it contain quotable facts? (2) — Do the cited third parties omit you? (3) — Is robots.txt clean? (4) — Did the answer have no citations at all? (5)
- Repeat across a few days before concluding anything — answers vary between sessions, and a single check misleads. The rigorous version is the DIY tracking protocol.
By step 3 you'll know which failure is yours — and four of the five have a fix you can start this week.
Read next: How to get cited by ChatGPT · How ChatGPT chooses its sources