Engines assemble your competitor set from co-occurrence in third-party content, so you are grouped with whoever gets listed alongside you, not with whoever you actually lose deals to.
Your sales team knows the real set. The engine knows the published one, and those two lists overlap less than anyone expects.
This is a measurable gap rather than a philosophical one. Ask an engine for alternatives to your product, then ask your head of sales for their last ten competitive losses, and compare the two lists.
How does an engine decide who your competitors are?
By reading pages where you appear next to other names, and there is no other mechanism available to it. Nobody publishes a register of who competes with whom, so the set is inferred from adjacency.
Three kinds of page do most of the work.
| Source | Why it groups you | How accurate it is |
|---|---|---|
| Comparison and alternatives pages | They exist specifically to list you beside named others. | Often stale, and frequently written by one of the vendors listed. |
| Review platform categories | Fixed category taxonomies place you in a bucket with everyone else in it. | Accurate about the category, blunt about who you actually meet in deals. |
| Community threads | People name the options they genuinely weighed, with reasons. | The most accurate of the three, and the least controllable. |
The third row is worth taking seriously. Google’s February 2024 announcement confirms it has access to Reddit’s Data API and will “display, train on, and otherwise use” that content. When somebody in a thread writes that they evaluated you against two named alternatives, that sentence is doing more to define your set than your own comparison page is.
Why is the resulting set usually wrong?
Because co-occurrence and competition are different relationships, and only one of them gets written down.
- What happens
- A roundup written two years ago still lists you beside a product you no longer meet, or one that has since moved market.
- Why it persists
- Nobody updates a listicle that still gets traffic.
- What happens
- A review platform taxonomy files you with the largest players in a broad category rather than with your actual segment.
- Why it persists
- The taxonomy is fixed and your segment is not one of its options.
- What happens
- The thing you actually lose to is a spreadsheet, an incumbent suite, or doing nothing. None of those appear in alternatives roundups.
- Why it persists
- Nobody writes “alternatives to not bothering.”
None of these are engine failures. Given only published adjacency, the set an engine produces is a reasonable inference from the evidence available. The evidence is just not the same thing as your pipeline.
How do you find out which set you are in?
Four steps, and the first two take an afternoon.
Running it five times matters more here than for most checks. OpenAI’s advanced usage guide notes that “Chat Completions are non-deterministic by default (which means model outputs may differ from request to request),” and a competitor list is exactly the kind of output that reshuffles between runs.
Microsoft’s AI Performance report in Bing Webmaster Tools helps with step three, since it reports “citation counts for specific URLs from your site” and the grounding queries behind them. That tells you which of your own pages are competing for the same questions.
What does a wrong set actually cost you?
Three things, in ascending order of expense.
You get compared on the wrong axis. Grouped with enterprise suites, you look expensive and thin. Grouped with free tools, you look overpriced. The comparison frame decides the verdict before any feature is discussed.
You lose the questions you should win. If your genuine differentiator answers a constraint nobody in your published set addresses, buyers asking about that constraint never reach the comparison where you win.
And you plan against the wrong competitors. A team that reads its AI competitor set as market intelligence will build a roadmap answering products it rarely meets, which is expensive in a way that takes two quarters to notice.
The set an engine returns is a map of who has been written about beside you. Treating it as a map of your market is the error.
How do you change it?
By changing the adjacency, which is slow, public and mostly not on your own site.
Get listed where the right set already appears. If your real competitors are named together on a roundup you are absent from, being added to it does more than a page of your own, because that page is already the evidence engines read.
Publish the comparison you actually want. A page comparing you to the products you genuinely meet creates adjacency where none existed. It works, it is slow, and it works better when the page is honest about where the other product wins, because an obviously partisan page gets less use as evidence.
Correct stale groupings at the source. A two-year-old roundup listing you beside a product you no longer meet is a request rather than a project, and most publishers will update a factual error, as covered in third party sources ai citations audit.
And fix the category label if that is the real cause. Being filed in the wrong bucket puts you beside the wrong names by construction, which is the mechanism in why doesnt ai put my product in the right category.
Should you ever accept the set you are given?
Sometimes, and it is worth considering before spending a quarter fighting it. Being grouped with larger, better-known products is not automatically bad: it puts you in answers to questions that have far more volume than your own name does.
The test is whether the comparison flatters or damages you on the axis buyers actually decide on. Listed beside an enterprise suite, a cheaper and simpler product often reads as the sensible option rather than the lesser one, and the adjacency is doing free work.
Where it damages you is when the grouping implies a capability you lack. Filed with products that all carry a compliance certification you do not have, every answer invites a comparison you lose on a single line. That case is worth the effort to escape; the flattering case is worth leaving alone.
What can you not change?
The unwritten competitor. If you mostly lose to inertia, no amount of adjacency work will make an engine name “doing nothing” as an alternative, and you should stop expecting your visibility numbers to reflect that part of your pipeline.
You can measure around it, though. Track prompts describing the problem rather than the category, the ones a buyer types before they know a category exists. Those are where a product competes against inaction, and they are constrained situational questions rather than vendor comparisons, which is the distinction in branded vs unbranded prompts ai visibility.
Recheck the set quarterly rather than continuously. Adjacency moves on the timescale of people publishing pages, so watching it weekly produces noise, and the run-to-run variation will look like change when nothing has moved.