Skip to main content

Why does ChatGPT recommend your competitor instead of you?

Usually one of four diagnoses: the engine cannot parse who you are, nobody else corroborates you, your pages give it nothing quotable, or a rival simply out-structured you. Each has a different fix.

Two answer cards side by side; a beam of light lifts and highlights one while the other stays dimmed
By Lars Nyman4 min readUpdated

The four diagnoses, in the order to check them

  1. Entity mush (it cannot tell who you are): If your homepage says "AI-powered growth platform" and your competitor's says "procurement software for mid-market manufacturers", the model can classify exactly one of you. Identity failures show up as a wrong description of your company, your name attached to the wrong category, or confusion with a similarly named firm. The audit checks whether your brand name appears in your title and H1 and whether Organization schema exists for a reason: that is the minimum identity surface an engine triangulates from.

  2. No corroboration (nobody else says you exist): Engines weigh third-party evidence before recommending. Zero external mentions is the strongest poor-visibility signal in the audit rubric; a competitor with a dozen mentions across distinct, recent sources wins the recommendation even with a worse product. The symptom: you appear when someone asks about you directly, and never in "best tools for X" answers.

  3. Nothing quotable (your pages resist lifting): Models assemble answers from fragments they can quote with confidence: definitions, numbers, direct answers under question-shaped headings. Long thought-leadership essays and abstract benefit copy give the model nothing to grab. The symptom: engines cite your category content but a rival's pricing and comparison pages.

  4. Out-structured (they did the unglamorous work): Sometimes the answer is that the competitor shipped what you have not: parseable HTML without JavaScript dependence, JSON-LD that validates, dated and bylined content, an llms.txt. Across the State of AI Visibility benchmark, the median B2B site scores 75/100 and structured data averages just 49/100, the weakest layer in the field. If they are in the top decile and you are at the median, the model is not biased; it is reading the available evidence.

The engine is not choosing your competitor. It is choosing the company it can parse, verify, and quote without guessing, and that company should be you.

What it is usually not

Not a pay-to-play ranking

There is no bid that puts a brand into an organic ChatGPT recommendation today. Treating the gap as a media-buying problem misdiagnoses it.

Not a single magic keyword

Retrieval is entity-and-evidence shaped, not keyword shaped. Stuffing the phrase you want into a page does not survive the corroboration stage.

Not permanent

Answers re-assemble per query, and the evidence base refreshes continuously. Sites that fix identity and structure see engines re-describe them without any petition process.

How to find your failing stage in an afternoon

Ask the engines directly

Prompt ChatGPT, Claude, and Perplexity with your real buyer questions ("best X for Y") and with "what does [your company] do?". Wrong description means diagnosis 1; absent from category answers but described correctly means diagnosis 2 or 3.

Read your homepage with JavaScript off

If your copy is not in the served HTML, you have a stage-zero fetch problem that masquerades as all four diagnoses at once.

Run the scored version

The free AI-Readiness Audit checks the full pipeline (fetch, access, structure, identity, trust, freshness, answerability, mentions) and ranks your failures by weight, which beats guessing which of the four to fix first.

What to do next

Run your top three buyer questions through ChatGPT and Perplexity today, note who gets named, and then audit the gap rather than assuming it.

Frequently asked

Questions

Run the free AI-Readiness AuditSee the live benchmark