A competitor shows up in AI answers and you don't: what to actually compare

Seeing a competitor's name in a generated answer, and not your own, immediately leads to the wrong question: "what are they doing that we're not?" The right question comes first: what exactly happened in that answer, and how stable is it. Many differences observed once are gone the next day, and chasing them means working on noise.

What follows helps you separate what you actually see from what you imagine, and choose what's worth checking.

First of all: reproduce the answer

A generated answer isn't a position in a list. It depends on the question, the context, who's asking, and the moment. Before drawing any conclusions, reconstruct the case and write it down.

• The exact question, copied word for word. Rephrasing it changes the answer, so the phrasing is part of the data.
• The moment and the channel: which system, in which mode, with or without web search turned on.
• Whether there was a conversation before. Earlier questions influence the one that follows, and a test done in a clean conversation isn't the same thing.
• The full text of the answer, saved. A week from now it won't reproduce identically, and memory alone isn't enough to discuss it.

Then you repeat it: the same question multiple times, on different days, and with two or three rephrasings a customer would use. If the competitor shows up once out of five times, the conclusion isn't "they cite them," it's "they sometimes cite them." These are two different situations, and they lead to different decisions.

Look at the sources, not just the answer

The answer is the result; the sources are the observable part. When the system lists them, they're the only concrete trace you can reason about.

• What type of sources they are: the competitor's own site, industry directories, articles, reviews, pages from associations or bodies.
• How many are the competitor's own and how many are third-party. A company cited only through its own site and one cited by independent sources are in two different situations.
• What exactly they say: often the phrase picked up is one specific line — a service, an area, a qualification — that doesn't appear in that form on your site.
• Whether they're up to date and correct. It happens that an answer cites outdated information or attributes things to a company that it doesn't actually do.

Comparing what you can actually observe

A useful comparison deals with public facts, not assumptions about the criteria the system uses to choose.

The last row is the most important. Nobody, outside of whoever builds those systems, can say why one company was chosen over another: anyone who claims to know is selling a certainty they don't have.

A three-column sheet: facts, hypotheses, missing data

The simplest way to avoid fooling yourself is to keep three things separate within a single document.

1. Facts: what you can see and show to someone else. The question asked, the answer received, the sources cited, what each site declares.
2. Hypotheses: the possible explanations, written as such. "They're cited because they state their area explicitly" is a hypothesis, not a confirmed cause.
3. Missing data: what you can't know. Listing it helps prevent a lack of information from turning, over time, into a conviction.

The document doesn't close the discussion: it opens it in an orderly way. Reread a few weeks later, it shows which hypotheses held up and which were just impressions of the moment.

Choosing one check at a time

A hypothesis leads to an action, and an action leads to a check. One at a time, otherwise the result can't be attributed to anything.

• The action needs to be made concrete: making the service area explicit, describing a service using the words of the query, publishing verifiable references of work done.
• The check is decided beforehand: which questions will be retested, how often, and for how long before forming a judgment.
• The comparison is with yourself over time, not with the competitor: their situation can change for reasons that have nothing to do with you.
• You also need a measure of the result that actually matters: if citations increase but inquiries don't, that's an interesting data point and nothing more.

When there's nothing to fix

• If the competitor is much better known and is cited by independent sources that have been writing about them for years, the gap isn't closed with a better-written page: you work on public evidence, and it isn't quick work.
• If the question that generates that answer isn't one customers actually come from, the citation isn't worth much.
• If the answers vary a lot from one test to another, there's nothing to explain yet: there's more to observe over a longer time.
• If the information about you is scarce or scattered, it isn't a competitive-comparison problem: it's a foundation that needs fixing regardless, independent of who gets cited.

What this guide doesn't cover

This guide covers a competitive diagnosis of what can be observed, keeping correlations separate from causes: no element of this analysis should be presented as a certain factor in selection. How to measure your own company's citations over time, with a repeatable method, is a separate piece of work. And which evidence makes content more credible remains a broader question, one that holds even with no competitor to compare against.