Seeing a weaker competitor mentioned can feel arbitrary. Sometimes it is. AI-generated answers are variable and may not include every qualified provider. But repeated exclusion across a carefully tested group of prompts normally exposes a real evidence gap.
The useful question is not ‘why does ChatGPT hate us?’ It is ‘what evidence was available for this answer that was unavailable or unclear for us?’
Your competitor answers the precise question
A broad homepage about ‘digital transformation’ is a poor match for ‘Manchester agency that helps local firms appear in ChatGPT’. A competitor with a focused page explaining the service, location, process and price gives the system a cleaner candidate.
Map the wording of missed prompts to your pages. If no page owns the intent, create one. If a page exists but hides the answer beneath slogans, rewrite the opening and headings before adding more content.
Your entity is ambiguous
Inconsistent names, old domains, duplicate profiles and conflicting addresses make it harder to connect facts to one organisation. A competitor may simply have a cleaner trail across its website, business listings, professional profiles and press references.
Audit every high-trust profile. Remove obsolete links where possible, align descriptions and connect the current domain. Do not create fake profiles to manufacture consensus.
They have independent evidence
A service claim on your own page is necessary but self-published. Reviews, customer stories, partner pages, regulatory records and respected publications can verify the relationship between a business and the work it claims to perform.
Compare sources, not backlink totals. Ten real customer reviews describing the relevant service can be more informative than hundreds of unrelated directory links.
Their evidence is fresher or easier to retrieve
An excellent page can still be absent if it is new, orphaned, blocked, rendered poorly or missing from internal navigation. Check indexing, crawl status, canonical tags, sitemap inclusion and whether the core answer appears in the rendered HTML.
Freshness should be honest. Update a date when the substance changes, not every week through automation. A changed date on an unchanged article does not make the evidence better.
The prompt itself favours them
Location, budget, sector and previous conversational context can alter a recommendation. Test neutral variants and record the conditions. One personalised chat is not a market-wide result.
Separate visibility problems from conversion problems. If you appear but users do not click or enquire, the issue is likely the offer, proof or landing page rather than citation coverage.
The takeaway
Build a competitor evidence matrix: prompt, named competitor, cited source, matching page, third-party proof and your corresponding gap. Fix repeated gaps first instead of copying competitors’ wording.
Questions people ask
No. Results can vary with wording, context, location, product experience and time. Use repeated neutral tests and record the conditions.
No. Copying creates an inferior substitute. Identify the information gap, then answer it with your own expertise, evidence and examples.
Relevant third-party mentions can help corroborate a business, but raw backlink volume is not a reliable shortcut. Relevance and credibility matter.
Start with the prompts that carry buying intent, inspect the sources used, then check whether you have an equivalent page and independent evidence.
Treat the answer as less diagnosable. Repeat the test in search-enabled experiences and compare it with conventional search results and trusted market sources.