Why ChatGPT Recommends Competitors Instead of You

A recommendation answers one particular question

A roofer may fit a general question about roof replacement but not a request for emergency flat-roof repair in a town outside its service area. A restaurant may suit a quiet dinner but not a vegan lunch close to the customer's location. The need, place, timing, and preferences can change the options.

Start by confirming that you and the named competitor genuinely serve the same need, customer, and place. If the question is too broad or the competitor is a different kind of business, you do not yet have a useful comparison.

One answer is evidence, not a verdict

Run the same commercially realistic question more than once before changing your website. Preserve the exact wording, date, location, product mode, businesses named, reasons given, and visible sources. Do not keep regenerating until your company appears and save only that answer.

A recent pre-registered study illustrates why this matters. Researchers tested 96 local food-and-drink questions across ChatGPT, Claude, Gemini, and Perplexity over seven days in two Bali markets. Agreement between systems was low, and answers changed between repeated runs and paraphrased questions. The authors concluded that a single visibility check contains substantial noise.

That study is useful evidence about its tested restaurants, cafes, locations, prompts, and products. It is not proof that a dentist in Halifax or a roofer in Boston will follow the same pattern. Use its lesson about repeated measurement without borrowing its findings as universal ranking rules.

Compare the evidence in four passes

Once the omission recurs, compare your business with the competitors that recur. This is not a hunt for the longest page or the largest directory count. You are looking for information that makes one company easier to identify, understand, verify, or trust for the question being asked.

• Fit: Does each website plainly state the service, location, service area, customer, and important limitations?
• Business facts: Do the name, address or service area, hours, phone number, categories, staff, and booking details agree across important profiles and the website?
• Support: Can a customer verify licenses, qualifications, completed work, review themes, memberships, or other claims through credible sources?
• Freshness: Are old locations, departed professionals, discontinued services, incorrect hours, or closed listings still circulating?

Use visible sources without mistaking them for the whole answer

Open every page shown with the answer. Note which statements each source supports, whether the information is current, and whether the same domains appear across repeated answers. A cited page gives you a lead about that response. It does not reveal every influence on the recommendation or establish a permanent source hierarchy.

Check your own public record as carefully as the competitor's. Google explains that Business Profile information can combine details supplied by the owner with crawled websites, licensed data, user contributions, photos, reviews, and Google's interactions with a place. That documentation describes Google's local products, not ChatGPT's private selection system. It still shows why conflicting public facts are a customer problem worth correcting rather than an AI-specific trick.

Do not copy the competitor's footprint blindly

A competitor may have more reviews, more directory profiles, a clearer service page, stronger local coverage, or simply a closer location. Those differences do not prove that copying one of them will produce a recommendation. Even Google's documented local-ranking guidance uses several considerations: relevance, distance, and prominence. Distance is a reminder that some omissions are not yours to fix.

The Bali venue study found that having a website, listed price information, more reviews, and third-party web mentions were associated with whether venues entered answers in that setting. It also found no positive effect from presence in one commonly discussed point-of-interest dataset after other documentation was considered. These are associations from a specific market, not instructions to buy mentions or manufacture reviews.

Prefer the smallest change that resolves a verified gap. Correct an old address. Clarify that your clinic offers a treatment it currently leaves buried in a PDF. Add license details customers can confirm. Explain which counties your home-service team actually covers. Ask a respected trade association to fix an outdated member profile. Each action should make the public record more useful even if tomorrow's ChatGPT answer stays the same.

Turn the comparison into a short work queue

Finish the review by sorting findings into three groups. Fix inaccurate facts first because they can mislead customers. Next, clarify missing information on pages and profiles you control. Then consider outside proof only where a real customer would expect it, such as a professional register, trade association, local publication, relevant review platform, or completed-project coverage.

Give every proposed action an observed reason. ‘Correct Saturday hours because two important profiles disagree’ is useful. ‘Create 30 articles because competitors are visible’ is not supported by the comparison. Record what you changed, allow time for the relevant pages and profiles to update, and repeat the original question set under comparable conditions.

Sometimes the honest result is that the competitor is closer, better suited to the question, or better established in the public record. That is still a useful diagnosis. It keeps you from spending money on a tactic that cannot change the customer's location, the competitor's genuine reputation, or the service fit behind the answer.