How to Map a Category in AI Answers

Mapping a category means measuring a market instead of a company. You name the market and start a report. It writes the questions buyers actually ask, puts them to ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews, and reads back the list of brands the answers name. Then you keep asking them, because one day of answers is a snapshot and the ranking is the product.

It takes five steps, and the second one does most of the work. The brands come out of the first report rather than going into it, and so do the first prompts.

Most AI visibility work starts from a brand you already own. You know the subject, you write prompts around it, you name the competitors you care about, and the numbers tell you how you're doing. That breaks down in two situations that come up constantly. You're an agency pitching a vertical and you don't have the client yet, so there's no brand to center the measurement on. Or you're in-house and the question isn't "how are we doing," it's "who is AI sending these buyers to before they've heard of anyone."

A category answers both. There's no website, no client, and no focal brand. There's a market, and a ranking of everyone AI names inside it.

This guide works with any AI visibility setup; the callouts show how each step runs in Spyglasses.

Step	What you're doing	What tells you it's working
1. Name the category	Describing the market and the buyer	A subject with no brand at the center of it
2. Start the report	Getting a first set of prompts asked across the platforms	One completed report and a first list of brands
3. Review the brands	Cutting the noise, adding the misses	Five or more real brands in the ranking
4. Round out the prompts	Building the starter set into a real panel	15 or more prompts, two buying stages covered
5. Track it every day	Putting the prompts on a schedule	A 7-day change on every brand

Name the category

Start with the market, phrased the way a buyer would say it out loud. "Wireless headphones." "Project management software for construction firms." "Home saunas." Not a segment name from a deck, and not a phrase only an analyst uses.

Then describe it in one sentence covering what buyers are choosing between and who those buyers are. This sentence does more work than anything else in the setup, because it's what the report reads when it writes your first prompts. A vague sentence produces vague prompts, and vague prompts produce answers about the wrong market.

Add the other phrases buyers use for the same thing. Plenty of markets have two or three names depending on who's talking; the industry term, the plain-English term, and whatever the biggest vendor calls it. If buyers say all three, all three belong here.

One rule that saves rework later. Keep brand names out of the category description entirely, including your own. The whole point is to see who AI volunteers on its own, and a description that mentions a brand contaminates every prompt generated from it.

Start the report

Start it as soon as the market exists. There's nothing to prepare first. If the category has no prompts yet, the report writes its first five from your description, spread across the buying stages, and saves them to the Prompts page. Then it asks every one of them across the major AI platforms in a single pass and comes back with the brands each answer named. That list is the map, and it's the first time anyone sees it.

Give it four to fifteen minutes. Then read it in a specific order, because the ranking on its own is the least surprising part.

Read the top first, and expect to already know those names. The leader in most markets is the brand everybody would have guessed, and confirming it took ten minutes instead of a research project.

Then read the middle, which is where the useful part usually starts. Brands sitting between five and fifteen percent are the ones with real presence and no lock on the category.

Then read for the names you didn't expect. Every first category report turns up two or three of them. Sometimes it's a regional player punching above its size. Sometimes it's a brand from an adjacent market that AI has decided belongs in this one. Occasionally it's a company that shut down two years ago and lives on in the model's memory, which tells you something about how stale the sources are.

Review the brands AI named

The report suggests the brands it found. Your job is to decide which of them are real, and that judgment is what makes the ranking mean anything.

Cut three kinds of noise. Product names that got read as companies. Brands from a neighboring market that don't compete for this buyer. And the generic terms that slipped through as names, which happens more in markets where the category label and a vendor name are close.

Deactivate the noise instead of deleting it. A deactivated name stays on the list, switched off, so the next report can't add it again. A deleted name is forgotten, and a later report can find it all over again.

Then add what AI missed. If buyers in this market ask about a company by name and the ranking doesn't have it, it belongs on the list. A brand at zero is a finding, not an omission; it means buyers know a name that AI never volunteers.

Last, add the aliases. Short names, nicknames, the old name after a rebrand, the parent company where people use it interchangeably. A brand tracked under one spelling while AI uses another reads as absent, and there's nothing on the dashboard to tell you that's what happened.

Five real brands is the working minimum for a ranking anybody will trust. Most markets land between ten and twenty-five.

Round out the prompts buyers ask

You have a starter set now, written from your description. It's enough to see the shape of the market and not enough to trust week to week.

A category has no site to measure, so the prompts are the entire instrument. Get them right and everything downstream works. Get them wrong and you'll measure a market you don't sell into. Read what the report wrote, cut what missed, and build from there.

The test for a good category prompt is simple. It describes a problem and a buyer, and it names nobody. "What headphones should I take on a long flight?" is a category prompt. "Are Acme headphones good for flying?" is a brand prompt wearing a costume, and it will hand you an answer about one company.

Cover at least two buying stages, and three if you can. The stages pull different names out of the model, which is the reason to separate them at all.

Stage	What the buyer is asking	What the answer usually contains
Awareness	How the problem gets solved at all	An explanation, sometimes with no brands in it
Consideration	Who the options are	A list of names, which is where the ranking comes from
Decision	Which option fits a constraint	A shorter list, often with different names on it

Consideration prompts carry most of the ranking, because the answer is literally a list of who gets named. Decision prompts are worth the effort anyway, since they routinely surface brands that never appear in the broader questions. That gap is often the most interesting thing the second report turns up.

Fifteen prompts is a working floor, three times what the report starts you with. Tag the whole set with the market's name while you're building it, so every dashboard downstream can filter to this market in one click.

Track it every day

One report is one day of a market, run on one set of answers. The ranking is what you were after, and a ranking needs history before it can tell you anything.

Put the same prompts on daily tracking under a project named for the market. From then on the number worth reading isn't the ranking; it's the 7-day change next to each brand. Leaders barely move week to week. The brand that climbed four points did something, or somebody wrote something about it, and that's the thread worth pulling.

A few weeks in, the map starts answering questions you didn't set it up for. Which brands are gaining without any coverage you can find. Which new entrant showed up the first day AI named it. Which claim started appearing in answers about the market right after a trade outlet ran a feature.

If the market matters in more than one place, add cities or regions as locations. Each one runs the same prompts against local answers and gets its own ranking and its own leader, and the regional leader is different from the national one often enough to be worth checking before anybody builds a territory plan. Budget for it, though; a tracked prompt on a category with three locations uses four prompt runs a day rather than one.

What the map is good for once it exists

A finished category map does three jobs, and only the first one is obvious.

It tells you who the field is, which is worth the setup on its own if you're walking into a market you don't know yet. It tells you who's moving, which is the part that stays useful in month six. And it tells you where the answers come from, because every tracked answer carries its sources, so the publishers AI reads for this market are sitting right there before you've pitched anyone.

That third one is why the same map gets used two more ways. An agency turns the ranking into a pitch list, since a brand buyers know well that AI rarely names has a gap it can be shown; turning a category map into prospects covers that end to end. A comms team goes at the claims instead of the names, tracking which parts of the market's story AI repeats and which outlets it takes them from, which is watching the narrative in a category.

When a brand in the market becomes a client, track them as their own property and keep the category running beside it. The two answer different questions, and baselining the brand properly is a different first week's work. If you want to see what a single market's answers look like before committing to any of this, a free AI Visibility Report on a brand in it is the fastest way to get a feel for the data.

Part of the Spyglasses workflow series. This guide is one of twelve workflows for PR and communications teams, and it also appears in the guides for marketing agencies and in-house marketing teams. The rest are collected in the AI visibility guides for PR and communications teams.