When a buyer asks ChatGPT for the best tool, platform, or product in your category and hears a competitor's name instead of yours, the engine is not expressing a preference. It is reporting the evidence it could find. An answer engine assembles a shortlist from the sources it can retrieve and verify for that specific question, and a brand that never makes the shortlist is missing from that evidence in one of a small number of specific, fixable ways.
We know because we ran the diagnosis on ourselves. Across four recommendation and visibility prompts in our own category, 1,728 production observations returned 0 AirPulse mentions and 0 citations of our own domain, and the matching Gemini expansion was no kinder. The patterns below are the ones that explain results like that, ours included.
The five failure patterns
Almost every competitor-gap case reduces to one or more of these, and each has a different fix.
- Weak category language. Your pages describe your product in your own vocabulary, not the buyer's. If the prompt says “AI visibility platform” and your homepage says “answer-engine intelligence suite”, the engine may never connect the two. The fix is stating plainly what the product is, in the words the failing prompts use.
- Missing third-party corroboration. The engine can read your claims but finds nobody else repeating them. Recommendations lean on comparison pages, review directories, editorial coverage, and community threads; a brand that exists only on its own domain reads as unverified.
- Inaccessible pages. Key pages render only with JavaScript, block crawlers, load slowly, or bury the product facts. If retrieval fails, nothing else matters: the engine cannot recommend what it cannot read.
- Inconsistent product facts. Your pricing page, docs, review profiles and old blog posts disagree about what the product does or costs. Conflicting evidence makes an engine hedge, and a hedging engine names a competitor with a cleaner story.
- Sources that mention competitors but not you. The pages the engine already cites for your category simply do not include you. This is a distribution gap, not a content gap, and no amount of on-site rewriting closes it.
The diagnostic decision tree
Diagnose before rewriting anything. Run your real buyer questions repeatedly and match what you observe to the starting pattern:
| Do this | If you observe | Start here |
|---|---|---|
| Ask the engine your buyer's question, five times or more | You never appear in any run | Start at pattern 1: your pages likely never enter the candidate set. Check retrieval and category language first. |
| You appear sometimes, competitors always | Mention is unstable | Start at pattern 4: your evidence is thinner or less consistent than theirs. Compare cited sources side by side. |
| You are named but never recommended | The engine hedges on you | Start at pattern 2: third-party corroboration is missing. The engine can describe you but cannot vouch for you. |
| You are recommended with wrong facts | Stale or conflicting sources | Start at pattern 5: inconsistent product facts. Reconcile pricing, features and positioning across every source. |
| Ask the engine to cite sources | Your domain is never cited | Start at pattern 3: retrieval. If the engine cannot read your pages, nothing downstream matters. |
One run proves nothing in either direction. The same prompt can name you today and skip you tomorrow, so repeat each prompt at least five times before believing a pattern, and evaluate each engine separately.
Prompt fit: are you even answering the question being asked?
Engines answer the buyer's question, not your positioning. A brand that sells to “enterprise revenue teams” but is asked about “sales tools for small agencies” loses on fit, and no optimization fixes a prompt you genuinely do not serve. Write down the 10 to 20 questions your buyers actually ask, in their words, and check each one honestly: is your product a correct answer? The ones where it is and the engine still skips you are your working list.
The source and citation gap
For each failing prompt, ask the engine to cite its sources, and read them. Three questions matter: which domains carry the answer, whether those domains mention you at all, and whether your own domain is ever cited. In our own diagnostic the competitor recommendations were carried by category comparison pages and community threads where we were absent, which no on-site change would have fixed. That reading tells you whether your 30 days go into your pages, other people's pages, or both.
Crawl and retrieval checks
Before investing in copy, verify the engine can read you: your key pages return their content without JavaScript, robots rules allow the AI crawlers you care about, structured data validates, and the product facts appear in the first screenful of text rather than behind tabs and accordions. These checks take an afternoon and regularly explain the whole gap, especially for JavaScript-heavy sites.
Evidence quality: what makes a page quotable
Engines prefer sources that state checkable facts: who the product is for, what it does, what it costs, and how it differs, with numbers and dates. A page that reads like a brochure gives the model nothing to quote. The test is simple: could a stranger answer the buyer's question using only your page, in two sentences, without marketing language? If not, neither can the engine.
Engine differences
The same question gets different shortlists on different engines because each retrieves from a different source mix: ChatGPT leans on comparison content and community discussion, Google's AI surfaces stay close to what already ranks, and Perplexity favors pages it can cite line by line. Diagnose and measure per engine; a blended score hides exactly the gap you are trying to close.
The 30-day fix plan
| Window | Work |
|---|---|
| Days 1 to 5 | Diagnose. Run your 10 to 20 buyer questions at least five times each on ChatGPT, and record who is named, who is recommended, and which sources are cited. Do not change anything yet. |
| Days 6 to 10 | Fix retrieval. Verify your key pages render without JavaScript, are indexable, load fast, and carry accurate structured data. A page the engine cannot read is invisible regardless of its copy. |
| Days 11 to 18 | Fix category language. Rewrite your product and comparison pages to state plainly what the product is, who it is for, and how it differs, using the words buyers use in the failing prompts. |
| Days 19 to 25 | Fix corroboration. Identify the sources cited alongside your competitors and earn accurate presence there: comparison pages, review profiles, directories, and the community threads the engine reads. |
| Days 26 to 30 | Re-measure. Re-run the same prompts on the same engines and compare against the baseline. Judge direction only; stability takes repeated runs over the following weeks. |
Two honest limits. First, causality is hard to claim: engines change under you, so use repeated runs and matched controls before crediting your own fixes. Second, some gaps close slowly; if your absence comes from missing citations on sources the engines trust, earning that presence takes longer than 30 days, and the plan above starts it rather than finishes it. Watch for citation decay after you do win a spot: a recommendation gained is not a recommendation kept.