Models do not invent categories. They reinforce the strongest existing ones, which means a company claiming a new category usually gets mapped to the nearest recognised neighbour instead. That neighbour is almost always a category a competitor already leads.
Category creation was always slow. What has changed is that the intermediate stage now has a specific cost: while your category does not exist to a model, every query about it resolves to somebody else's.
The mechanism that fixes this is not publishing. It runs through taxonomies other people maintain, and it is more concrete than most positioning advice suggests.
Key Takeaways
- Declaring a category does nothing on its own: one source making an unusual claim is noise.
- Third-party taxonomies carry the weight: review platforms, directories, and reference sources.
- Placement can move quickly: a category created in a major taxonomy is recognised far faster than one asserted on a website.
- Dual framing is necessary in the meantime: stay findable in the adjacent category while the new one forms.
- Most companies should not attempt this: the cost is real and the alternative is usually better.
Why Assertion Alone Fails
A model builds category associations from repetition across independent sources. Your own site making a claim about a new category is a single source, and it is the source with the most obvious incentive to make it.
When the claim has no external support, the model does what any retrieval system does with an unfamiliar term: it maps to the nearest thing it recognises. Your positioning document says you created a category. The answer places you inside an existing one, usually alongside the competitors you were trying to differentiate from.
This produces a recognisable failure. Traffic and pipeline hold up, buyers who already know you find you, and category-level questions return a list you are absent from because the question itself was framed in vocabulary you rejected. It is the same pattern our AI Recommendation Gap report traces between accurate presence and favourable framing, applied to vocabulary rather than to preference.
The Taxonomies That Actually Register
Review platforms and directories are not marketing surfaces in this context. They are structured taxonomies that models read as authoritative classification, maintained by parties with no stake in your positioning.
The clearest documented example is recent. G2 created a GEO Tools category in March 2025, later renamed to AEO Tools. Profound was among the products placed there at creation, and by February 2026 had been named a Top 50 AI Product in G2's annual rankings, under a year from the category first existing.
The sequencing matters more than the outcome. The category became a recognised classification maintained by a third party, and placement within it became a fact about the company rather than a claim by it.
Four taxonomy types do most of the work:
- Review platforms. G2, Capterra, and equivalents add categories regularly and publish the additions, which makes them the fastest available route.
- Reference sources. Slower and harder to obtain, with correspondingly more weight once established.
- Analyst and research coverage. A named category in an analyst framework travels through every publication that covers it.
- Industry directories and event programming. A conference track or awards category is a classification decision made by someone else.
The Order That Works
Category creation in this context has a sequence, and reversing it wastes the most expensive part.
Name it, then stop talking about it for a while
Fix the term and define it precisely enough that someone else could apply it consistently. Then resist the instinct to build the entire content programme around it before anything external recognises it. Content published into a vocabulary vacuum does not create the vocabulary.
Get one external party to use it
A single credible third party using your term in their own words is worth more than fifty pages using it in yours. Review platform categorisation, analyst coverage, or a publication adopting the framing all qualify.
Then publish explanatory content
Once external usage exists, explanatory content compounds it rather than asserting alone. This is where the category definition, comparison to adjacent categories, and use case material belong, and where it starts being retrieved rather than ignored.
Reinforce across ecosystems
A category framing established in one place does not automatically propagate, since platforms draw on largely different source sets. Recognition in a review platform taxonomy may not reach an engine that leans on community sources instead.
What to Do While the Category Does Not Exist
The interim period is where most category plays lose money, because companies stop competing in the old category before the new one exists.
| Adjacent category | New category | |
|---|---|---|
| Model recognition | Established | None initially |
| Query volume | Real today | Near zero |
| Competitive position | Weak fit | Uncontested |
| What to do | Stay findable | Build recognition |
| Content role | Comparison and fit | Definition and explanation |
Dual framing means being retrievable for the adjacent category question while building the new one, and being explicit about how you differ rather than refusing the comparison. Buyers asking the old question are real; the new question has no volume yet by definition.
Refusing to appear in the adjacent category until the new one is recognised is a decision to be absent from every query that currently exists.
How to Tell Whether It Is Working
Category recognition has an observable progression, and each stage takes a different intervention.
First, the term is unrecognised and queries using it return nothing coherent. Then it returns a definition, frequently drawn from your own material, which means the term is recognised but not yet populated. Then other companies appear alongside you, which sounds like failure and is the clearest sign the category has become real.
That third stage is the one worth planning for. A category with only one member is not a category to a model; it is a product name. Competitors appearing is confirmation, and the position to defend at that point is leadership within the category rather than sole occupancy of it.
Tracking this requires running both the new term and the adjacent one on a schedule, which is a small addition to standard AI visibility tracking and easily forgotten.
When Not to Do This
Category creation is expensive, slow, and frequently the wrong call. Three conditions where the alternative is better.
Your positioning is still moving. Establishing a term across external sources and then changing it costs more than never establishing it, since models update slowly and corrections propagate at the pace of the sources carrying them.
The adjacent category describes you adequately. If buyers find you through it and the fit is reasonable, differentiation within a recognised category is cheaper and faster than building a new one.
You cannot sustain two to three years. Recognition builds over that horizon. A category play abandoned at month nine leaves you with weakened positioning in the old category and no recognition in the new one, which is worse than either alone.
The Underlying Point
Positioning has always required other people to repeat your framing before it becomes real. What is different now is that the intermediate stage is measurable, and the cost of staying in it is visible in every category answer that names your competitors and not you.
The work is external rather than editorial. Third-party sources determine what models say more consistently than owned content does, and category vocabulary is the clearest case of it: the term becomes real when someone other than you maintains it.
FAQs
1. Can you create a new category in AI search?
Not by assertion. Models reinforce existing categories rather than inventing them, so a term used only on your own site gets mapped to the nearest recognised neighbour. Recognition comes from third parties maintaining the classification, particularly review platform taxonomies, analyst frameworks, and reference sources.
2. Why does AI place my company in a competitor's category?
Because that category exists in the model's frame and yours does not yet. Retrieval systems map unfamiliar terms to the nearest recognised concept, which is usually the established category your product most resembles.
3. What is the fastest way to get a new category recognised?
Review platform categorisation is generally the fastest available route, since platforms like G2 add and publish new categories regularly. G2 created its GEO Tools category in March 2025, and products placed there at creation gained recognised classification within months rather than years.
4. Should I stop competing in my old category while building a new one?
No. The new category has almost no query volume by definition, while the adjacent one has real volume today. Remain retrievable for the established category, be explicit about how you differ, and build recognition for the new term in parallel.
5. How do I know the category is becoming real?
When competitors start appearing alongside you in answers about it. A category with a single member reads as a product name; other companies appearing confirms the classification has been recognised, and the objective shifts to leadership within it.
References include documented G2 category creation dates and published coverage of AEO tooling classification. Category recognition timelines vary considerably by sector, taxonomy activity, and how distinct the proposed category is from established ones.