Engines answer category questions using the category words the web uses, not the ones your positioning deck uses. Inventing a category name reliably removes you from the queries buyers actually ask.
This is not an argument against category creation. It is an argument that the invented name is a second name, and the first one still has to exist somewhere retrievable.
Category creation is good advice that acquires a new cost in AI search. The cost is specific, measurable, and almost always discovered late, because the metric that would show it is the one nobody was tracking.
What does an engine do with a category question?
It works from the words in the question, and those words come from the buyer rather than from you. Google’s May 2026 post on AI Mode reports that “the average AI Mode search is triple the length of a traditional Search query,” and describes people “asking the questions that are really on their mind.”
Longer, more conversational questions are worse for invented vocabulary, not better. A three-word search might happen to contain your category name. A twenty-word question describing a situation almost never does, because the buyer is describing a problem rather than naming a solution class.
So the retrieval step runs on the buyer’s words. If your pages describe you only in your words, there is nothing for those queries to reach.
Why does an invented name remove you specifically?
Because it breaks the association at both stages, and for different reasons at each.
The buyer asks about “tools that do X.” Your pages say you are a “revenue intelligence orchestration layer.” No page of yours contains the phrase being searched.
The model learned which brands belong to the common category from sources that use the common words. A name only you use appears in one place, so it forms no association.
The structural version of the same point is visible in how categories get stored. Wikidata’s introduction describes statements that “consist of a property and a value,” attaching an item to other items. A category is a value that other things also point at. A category with exactly one member is not a category, it is a description, and it connects you to nothing.
A category nobody else uses has no other members, so it cannot put you on a shortlist. Shortlists are made of categories with more than one member.
How do you find out which words the web actually uses?
Stop guessing and read the record. Three sources, in descending order of reliability.
Microsoft’s AI Performance report is the strongest of the three because it is a record rather than a sample. It covers Microsoft surfaces only, which is a real limit, and within that limit it tells you the exact phrases that pulled your pages into an answer.
Does this mean category creation is a mistake?
No, and the post would be wrong if it said so. Creating a category is how a company escapes being compared on features it does not want to compete on, and that logic has not changed.
What has changed is the cost of doing it exclusively. In a search world, an invented name cost you some head-term traffic and you accepted the trade. In an answer world it costs you membership of the list, because the list is assembled from a category label and yours has one member.
The failure is not the new name. The failure is retiring the old one from everywhere it was written down, which usually happens in a rebrand and is usually invisible for two quarters.
How do you carry both names?
Deliberately, and with the common one doing the retrieval work.
| Surface | Which name | Why |
|---|---|---|
| Homepage and positioning | Your name, prominently | This is what it is for, and an engine reading it learns the association exists. |
| Comparison and alternatives pages | The common name | These pages exist to be retrieved on the buyer’s words. Your vocabulary defeats their purpose. |
| Docs and integration pages | The common name | The most retrievable pages you own, usually written in plain terms already. |
| Third-party profiles | The common name, plus yours | Category fields on review platforms are picked from a fixed list. Fighting that list removes you from it. |
| Structured data | Both, as name and alternateName | Gives a machine the equivalence explicitly rather than making it infer one. |
The one sentence worth writing somewhere retrievable is the bridge: your invented name stated as an instance of the common one, in plain words, on a page an engine can fetch. That single sentence does more than a rebrand, because it is the thing a model can lift and reuse.
None of this requires special markup. Google’s documentation on AI features states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and that a page “must be indexed and eligible to be shown in Google Search with a snippet.” The lift is editorial rather than technical, which is also why it tends to sit unowned, as covered in who owns AI visibility in a company.
What if the common category is wrong about you?
Sometimes it genuinely is, and the honest answer is that you carry the wrong label anyway for a while. A category is a route to being considered, not a description you are stuck with once somebody is reading your page.
The trade is worth stating plainly. Being filed under a category that undersells you gets you onto lists where a buyer can then discover the difference. Being filed under nothing gets you onto no lists at all, and there is no second step from there.
The exception is a category that would actively disqualify you, where appearing on the list costs more than missing it. That is rarer than most positioning conversations assume, and it is worth testing rather than asserting: ask an engine for options in the category you are avoiding and read whether the answers describe a buyer you would want.
How do you test whether you have the problem?
Two questions, asked of several engines, and the gap between the answers is the finding.
First, ask for options in the common category without naming yourself. Count how often you appear. This is your discovery position and it is the number with revenue attached.
Second, ask what your company does, by name. If the engine describes you fluently here and never names you in the first question, you have a vocabulary problem rather than an awareness problem. The model knows exactly who you are and does not file you under the thing buyers ask for.
That specific pattern, known but never shortlisted, is diagnosable and fixable, and it is separate from simply being unknown. Which of the two you have is the fork in third party sources ai citations audit, since third-party pages are usually where the common vocabulary lives or fails to.
Run both questions more than once before concluding anything. A single run tells you what one sample looked like, and category answers vary more than most, because the set of brands an engine considers plausible for a category sits close to the boundary where variance is highest.