Why ChatGPT names your competitor and not you

There are two routes into an answer. The first is training data: everything the model read before its cut-off. Brands that were written about often, by many sites, in consistent terms, are more likely to be “known”. The second is live retrieval: ChatGPT search rewrites the question into several web searches, fetches a set of pages, pulls out the passages that answer the question, and writes a summary with citations.

Both routes reward the same thing. A model is not judging which business is best; it is judging which answer is best supported by what it can read. A competitor mentioned on eight review sites, directories and “best in the city” lists will usually beat a better business that is only described on its own website.

That is uncomfortable, but useful. It means the gap is usually not your service or even your content quality. It is the evidence trail around your brand — and that can be built deliberately.

Your website is a self-interested source. Assistants read it, but they look for agreement elsewhere before repeating a claim like “the leading tax consultant in Indore” or “specialists in ACL rehab”. When three or four independent pages say something similar, the claim becomes safe to repeat.

In our experience on audits, the brand that gets named is rarely the strongest business in the market. It is the most corroborated one: the clinic listed on every directory with the same description, the agency quoted in two trade articles, the software tool that appears in every comparison post. Brands that lose usually have a strong site and almost nothing else.

The work here is closer to PR than SEO: getting listed, reviewed, quoted and compared in the places your buyers — and the models — already read.

• Industry and local directories with a full, consistent profile
• Review platforms your buyers actually use, with recent reviews
• “Best [service] in [city]” and comparison articles by third parties
• Trade press, association member lists and event speaker pages
• Podcast show notes, interviews and partner pages
• Community threads where your name comes up unprompted

Small inconsistencies add up. “[Name] Dental Care” on the website, “[Name] Dental Clinic” on Google, an old address on a directory from before the move, and a service list that mentions implants in one place and not another. To a person these are obviously the same practice. To a system stitching facts together from different pages, they are three slightly different entities.

Write one canonical description: legal and trading name, one-line description of what you do, primary category, address or service area, phone, founding year and main services. Use it word for word on your site, your Google Business Profile, LinkedIn, and every directory. Add Organization or LocalBusiness schema with sameAs links pointing to those profiles so the connection is explicit.

Compare “We help businesses unlock their potential” with “We are a physiotherapy clinic in Baner, Pune, specialising in rehab for runners and ACL injuries.” The second sentence answers a real question on its own. It can be lifted into an answer without any surrounding context and still be accurate.

Pages that tend to get cited are the ones built around a specific question: service pages that say exactly what is included, pricing or cost explainers, honest comparison pages, and FAQs written in your buyers’ words. Put the direct answer first under each heading, then the detail. If an assistant has to read six paragraphs to work out what you do, it will quote someone who said it in one.

OpenAI documents separate crawlers for separate jobs: OAI-SearchBot for surfacing sites in ChatGPT search, GPTBot for collecting training data, and ChatGPT-User for pages a user asks it to visit. You can block training while still allowing search — but many sites block all three by accident with a blanket rule, or because a security service now blocks AI crawlers by default.

Rendering matters too. If your key content only appears after JavaScript runs, or sits inside tabs, images or PDFs, a crawler may fetch the page and still find nothing to quote. A quick test: view the page source, or fetch it with a plain request, and check that the answer text is there in the HTML.

Week one: confirm crawlers can reach and read your important pages, and rewrite the first lines of your homepage and main service pages so they state plainly what you do, for whom and where. Week two: publish your canonical business description and update every profile and directory to match it.

From then on, corroboration is a steady programme rather than a project. Each month, aim for a few new independent mentions in places that matter to your buyers: a directory listing completed properly, a guest article, a client case study on a partner’s site, a comparison list you have earned a place on. Track the same set of buying prompts monthly so you can see whether the pattern shifts.