The Five Prompts That Decide Who Owns Your Category
By
Izzy Gregorio
·
5 minute read
How-To · Prompt Testing
The Five Prompts That Decide Who Owns Your Category
Only 15.2 percent of categories have a clear brand owner in ChatGPT. More than half have no consistent leader at all. Most brands have never run the test.
By Izzy Gregorio · Updated August 2026 · 10 min read
In short
Category ownership in AI answers is measurable with five prompt types: definition, comparison, alternatives, use case, and buying. Run them, log who gets named, and you learn two things: whether you are present, and whether the category is already claimed. Both answers change what you should spend money on next, and one of them has a closing date.
The finding
More than half of categories are unclaimed
Kevin Indig's Semrush analysis ran between January and June 2026, covering 1,094 US categories, more than 50,000 brands, and 600,000 citations. Four numbers from it matter here.
| Finding | Figure |
|---|---|
| Categories with a clear brand owner in ChatGPT | 15.2% |
| Ownership rate in the highest-demand categories, which carry 98% of AI search volume | 11.3% |
| Categories with no consistent leader at all | 53.7% |
| How often an owner holds position month over month | 90.4% |
Read the first and last together. Ownership is rare and, once established, it is sticky. A brand that claims a category holds it nine times out of ten the following month.
That combination is what makes this urgent rather than interesting. Unclaimed does not mean unclaimable forever. It means the seat is empty right now.
The instrument
The five prompt types
Each shape maps to a different moment in how a buyer interrogates a category. Skipping one produces a blind spot in exactly the place your competitor may be winning.
- 1
Definition. What is [category]?
Tests whether you are associated with the category at all. Brands cited here shape how the category itself is understood, which is the most durable position available and the least commercially urgent.
- 2
Comparison. [Your brand] versus [competitor]
Tests whether the model can describe you accurately when placed beside a rival. This is where inaccurate or outdated descriptions surface fastest, and where a stale directory listing does the most damage.
- 3
Alternatives. Alternatives to [competitor]
The prompt most brands never think to run and the one with the clearest commercial payoff. A buyer running this is actively shopping and already dissatisfied. Being named here is being handed a warm lead.
- 4
Use case. Best [category] for [specific situation]
Tests whether you own a defensible niche. Broad category ownership is expensive and slow. Situation-specific ownership is cheap and available, and it converts better.
- 5
Buying. Which [category] should I buy?
The highest-intent prompt and the shortest answer. Usually one to three names. There is no second page.
Do the work
How to run it properly
Six steps. Budget an afternoon for the first pass. Every month after that takes a fraction of the time, because the instrument already exists.
- 1
Write your variants. Five types times five real customer situations. That produces 25 prompts, enough to be stable and small enough to actually run.
- 2
Use real language. Phrase prompts the way a buyer speaks, not the way your category writes about itself. If your customers say cheap and you say value-tier, write cheap.
- 3
Name real competitors. Types two and three require specific names. Use the three to five brands you actually lose deals to, not aspirational ones.
- 4
Run each prompt three times per engine, logged out. Personalization flatters and is not reproducible, so use a private window with no account signed in. Three runs is the minimum that survives answer volatility.
- 5
Run all four engines. ChatGPT, Claude, Gemini, Perplexity. Engine behavior varies widely, and one engine gives you one engine's answer.
- 6
Log four columns. Were you mentioned. Were you cited with a link. Which competitors were named. What date and which engine.
Step four is the one most methods skip, and it decides whether the number means anything. AirOps found only 30 percent of brands remain visible across consecutive AI responses. Ask the same question twice and roughly seven times out of ten a different set of brands comes back.
Record the rate at which you appear across the three runs, not whether you appeared once.
Then freeze the prompt list verbatim. It is the instrument. A prompt set you edit every month measures your editing.
Read the result
What the pattern tells you
| What you observe | What it means | What to do next |
|---|---|---|
| Same one or two brands named in nearly every response | Category is owned. Claimed and sticky | Compete on use case prompts. Broad ownership is expensive to take |
| Different brands each run, no consistency | Category is unsettled. This is the 53.7 percent | Move now. Ownership is available and it consolidates |
| You appear in definition prompts only | You are known and not shortlisted | Structural and citation supply work on commercial pages |
| You appear in buying prompts only | You convert and you have no category authority | Definition and use case content, plus third-party corroboration |
| You appear nowhere in any prompt | Access or entity problem before anything else | Check robots.txt and listing consistency before writing content |
The last row catches more brands than people expect, and it is also the cheapest to fix.
Translate it
Which prompt matters most, by business
| If you sell | The prompt type to run first | Why |
|---|---|---|
| A consumer product | Use case. Best [product] for [specific need] | Situation-specific ownership is where product attributes actually decide the answer |
| Professional services | Alternatives. Alternatives to [competitor] | Your buyers shop by dissatisfaction with an incumbent more than by category research |
| A local service business | Buying. Which [service] should I use in [city] | The answer names one to three, and the local shortlist collapsed hardest |
| Software | Comparison. [You] versus [competitor] | Comparison content dominates software buying and inaccurate descriptions cost deals directly |
| A nonprofit or ministry | Definition. What organizations near me do [thing] | Recognition and accurate description matter more than shortlist position |
The honest limitation
What this test does not tell you
It is a snapshot of a system that has already moved. Half of all AI-cited content is less than 13 weeks old per Amsive's 2026 analysis, so a single month's run is a photograph rather than a measurement. Repeat it before you trust the trend.
It also does not tell you why. A prompt set shows you are absent. It does not distinguish between blocked crawlers, unextractable pages, and weak third-party corroboration, and those three problems have three different budgets attached.
A test that tells you where you stand is not the same as a test that tells you what to do. Run it anyway. Knowing the shape of the problem before you buy anything is the entire point.
One question worth sitting with. If you ran the alternatives prompt for your closest competitor right now, and your brand was not in the answer, you would have learned that a buyer actively shopping away from them is being handed a list you are not on. So what is the actual reason that test has not been run yet?
Common questions
Category prompt testing, answered
What prompts should I test to see if AI recommends my brand?
Five types: definition, comparison, alternatives, use case, and buying. Definition tests category association, comparison tests description accuracy, alternatives captures active shoppers, use case tests niche ownership, and buying tests shortlist position. Run each across several real customer situations.
How many prompts do I need?
Around 25 for a single category, produced by five prompt types across five real customer situations, with each prompt run three times per engine. Fewer than twenty produces results that swing month to month. More than fifty becomes hard to run consistently by hand, and consistency matters more than volume.
Which AI engines should I run prompts on?
ChatGPT, Claude, Gemini, and Perplexity at minimum. Engine behavior varies widely. SOCi's 2026 index found ChatGPT surfaces a local business 1.2 percent of the time against Gemini at 11 percent, so a single-engine test produces a single engine's answer.
How do I know if my category is already owned?
Two different tests, and you need both. Within a session, run each prompt three times: if the names change between runs, that is answer volatility rather than a signal. Then repeat the whole set a few weeks later: if the same one or two brands dominate both times, the category is owned. Semrush found 53.7 percent of 1,094 US categories had no consistent leader.
Should I run prompts logged in or logged out?
Logged out, in a private window. A signed-in session applies personalization based on your history, which will tend to surface your own brand and produce a result nobody else would see. Logged-out runs are reproducible, which is what makes a trend meaningful.
Now what do the numbers mean
The Citation Gap Report 2026
You now have the method. What you do not have is the benchmark. The report contains the citation data behind every figure on this page, what an owned category looks like against an unsettled one, and what each result pattern actually costs to fix. Sources you can trace, and the numbers your own run will be compared against.