Somebody on your team asked ChatGPT for the best tool in your category, and it named your competitor. Maybe it named three competitors. You read the answer twice, checked that it wasn’t a fluke, and now there’s a Slack thread with a screenshot in it and a certain amount of feeling.
Here’s the thing to hold onto before anyone starts theorizing: the engine isn’t expressing an opinion about your product. It has never used your product. It has never used theirs either. What you’re looking at is a retrieval outcome, and retrieval outcomes have causes you can find. In my experience the cause is almost always two or three specific pages, and you can usually identify them in under an hour.
That’s the whole post: the answer to “why them and not us” is knowable and boring, and the diagnosis is a procedure, not a mystery.
What the recommendation actually is
When an engine answers “what’s the best X,” it isn’t ranking products. It’s assembling a summary of what its sources say, weighted by which sources it trusts and how often they agree. The mechanics differ engine by engine, but the shape is constant: some small set of comparison pages, review profiles, community threads, and articles taught the engine that your competitor is the name you say when someone asks that question.
Your competitor didn’t win a quality bake-off. They won a coverage bake-off, mostly by existing in the right places, and often without knowing they were doing it.
That reframe matters because it changes the question from “how do we make the AI like us” (not actionable, slightly deranged) to “which pages taught it this answer” (a research task with an end).
The twenty-minute diagnosis
Run this before you commission any strategy work. You need a chat window and a notes file.
First, ask the question the way your buyer asks it, on the engines your buyers actually use. Not your marketing phrasing, the buyer’s phrasing. Run it three times per engine, because these answers are non-deterministic and one run is noise. Note who gets named and in what order.
Second, ask the engine to show its work. On the engines that cite live sources, the citations are sitting right there in the answer. Follow every one. On ChatGPT, ask a follow-up in plain language: “what sources would you point to for that recommendation?” You’ll get an imperfect but useful list. What you’re building is a short inventory of the pages that write this answer.
Third, read those pages the way the engine reads them. Open each source and look for your competitor’s name, then look for yours. You’ll typically find one of three situations. Your competitor is present and you’re absent. You’re both present but their entry is fuller, fresher, or more specific. Or you’re present and described wrongly, in the wrong category or with stale facts. Each of those is a different fix, and now you know which one you have.
That’s the diagnosis. I’ve written up the fuller version of this source audit if you want the complete procedure, but the twenty-minute version usually finds the culprits. When I run this for companies, the answer is rarely a diffuse cloud of coverage. It’s a comparison post from last year, a review profile with forty reviews against your nine, and one Reddit thread that comes up for everything.
One more thing worth noting while you’re in there: check whether the engines even agree. Run the same question on ChatGPT, Perplexity, and Google’s AI answers and you’ll often find your competitor owns one engine while the others are up for grabs, because each engine leans on different sources. An answer that feels like “the AI recommends them” is frequently one engine’s habit, not a consensus, and the engines where nobody has taught a firm answer yet are the ones you can move fastest.
The three usual culprits
The comparison page is the most common. Somewhere there’s a “best tools for X” roundup or a head-to-head that the engines lean on hard for your category. Your competitor is in it and you aren’t, or you’re in it with a two-line entry written by someone who spent four minutes on you. Engines love these pages because they’re structured exactly like the answer the user wants. Comparison queries have their own dynamics, and being absent from the page that answers them is close to being absent from the answer.
The review gap is second. Review sites are corroboration machines: dozens of independent-ish voices agreeing that a product exists, works, and is used by companies like the asker’s. The review layer feeds AI answers more than most vendors expect, and a competitor with ten times your review volume reads, to a machine, like a company ten times as established.
The community thread is third, and it’s the one founders find most annoying. One well-upvoted Reddit thread where three people recommend your competitor can echo through AI answers for a year, because engines treat community consensus as the closest thing to ground truth they can retrieve. Nobody in that thread was lying. Your users just weren’t in the room.
What I got wrong
The first time a client came to me with the screenshot and the feeling, I skipped the diagnosis, because the answer seemed obvious. The competitor was bigger, older, and better funded. I told the client, more or less, that the engine was reflecting market reality and the fix was to grow. Sympathetic, plausible, and wrong.
When we finally did the source work, the recommendation wasn’t coming from the competitor’s size at all. It was coming from two places: a single category roundup that half the engines cited, where my client simply wasn’t listed, and the competitor’s own comparison pages, which were the only head-to-head content in the category and which the engines quoted almost verbatim. The market-reality story I’d told was unfalsifiable and useless. The actual causes had URLs.
Getting the client added to the roundup took one email and three weeks. Publishing an honest comparison page took a fortnight more. The answers didn’t flip overnight and they didn’t flip completely, but within two months my client was appearing alongside the competitor in the same answers that had excluded them. The lesson I kept: “they’re bigger” is a description, not a diagnosis. If you haven’t traced the actual sources, you don’t know why you’re losing, and you’ll aim the fix at the wrong thing.
What to do with what you find
Match the fix to the situation you found, because they’re different jobs.
If you’re absent from the sources, the work is presence: get into the roundups honestly, build the review base, show up where your category gets discussed. This is outreach and patience, not content volume.
If you’re present but thin, the work is depth: fuller entries, fresher facts, a real comparison page of your own that gives the engines something specific to lift.
If you’re present but wrong, the work is correction, and that’s a source-layer fix with its own playbook: correct the facts where they’re published, and put a clean, dated, quotable version of the truth on your own site.
And in every case, keep the distinction between being cited and being recommended in view. The first milestone isn’t the engine saying you’re the best. It’s the engine knowing you belong in the answer at all.
The screenshot in your Slack thread feels like a verdict. It’s a symptom, it has a cause, the cause has a URL, and URLs can be changed. If you want help finding yours and changing them, that’s the work we do and the method we use.