An AI assistant asked a broad question hedges; asked a narrow one, it commits to a name. That asymmetry turns niching from good advice into a visibility mechanic: the expert who narrows their claim until a real buyer query has only one defensible answer becomes that answer. This piece gives you the niching ladder, a five-check test for whether your niche can carry a category of one, and the honest math on why narrow converts better than broad.
Positioning advice has said "niche down" for forty years. The machines finally made it enforceable: the generalist gets summarized, the specialist gets named.
Why AI answers force the niche question
Ask an assistant "who is the best business consultant" and you get a polite essay: it depends on your industry, your stage, your budget. Ask "who should a UK dental group hire to fix associate retention" and the engine behaves differently. The query is specific enough to have a small set of defensible answers, and if one person has visibly done the work on exactly that problem, the engine can commit to a name with justification. Confidence is the currency of these systems. Broad questions dilute it across thousands of candidates; narrow questions concentrate it on whoever built the deepest record.
This matters commercially because the narrow questions are where buying happens, and increasingly they are asked in a chat window rather than a search box. Google's own results now often resolve without a click, with zero-click searches growing from 56% to 69% in the twelve months after AI Overviews launched, according to SEO Sherpa's AI search statistics, and AI Overviews themselves appearing on roughly 25% of searches. The surface that rewards being one option among ten is shrinking. The surface that rewards being the single committed answer is growing. That is the whole strategic argument for a category of one, and it lands hardest for whoever moves while their lane is still empty, the dynamic we described in the gold rush and the empty lane.
How narrow should my niche be? The niching ladder
Most experts stop narrowing two rungs too early because each rung feels like it discards revenue. Walk the ladder deliberately instead. Each rung stacks a qualifier, and each qualifier removes competitors faster than it removes buyers.
- Role. "Marketing consultant." Millions of competitors, zero chance of being named. This is a job title, not a position.
- Role plus industry. "Marketing consultant for law firms." Thousands of competitors. Better, still a crowd.
- Role plus industry plus problem. "Client intake conversion for law firms." Dozens of credible competitors. Now the engine's shortlist is small enough for record depth to decide it.
- Role plus industry plus problem plus context. "Client intake conversion for US personal injury firms spending six figures on ads." Here the query often has one defensible answer, and whoever holds the record is it.
The test at each rung is not "does this sound impressive." It is "would a buyer with money type this, and is the answer set small enough for me to dominate." Rungs three and four are where categories of one live.
The category-of-one test: five checks before you commit
| Check | The question | Pass looks like |
|---|---|---|
| 1. Typed | Would a real buyer phrase this as a question to an assistant? | You can write five natural buyer prompts for it without straining |
| 2. Paid | Does winning this query lead to engagements at your target rates? | The buyer asking it controls budget and is close to a decision |
| 3. Defensible | Can you back the claim with real work you have actually done? | You could talk for an hour, with cases, on this exact problem |
| 4. Winnable | Is the lane empty or held by someone with a shallow record? | Assistants currently hedge, or name someone without much justification |
| 5. Extendable | Does owning this lane open adjacent lanes you want next? | You can name the two or three neighbouring queries you would claim after |
Run every candidate niche through all five. Four passes and one fail means keep looking: a niche nobody types is a hobby, a niche that cannot pay is charity, a niche you cannot defend is a liability waiting for its first hard question, and a niche with a deeply entrenched owner is a war of attrition you do not need when the next lane over is open.
Before committing, spend an hour asking ChatGPT, Gemini and Perplexity the buyer questions in your candidate niche. Where they hedge, there is an opening. Where they name a competitor with weak justification, there is a contest you can win. Where they name someone with a deep record, move one qualifier over.
Will niching down shrink my business? The math says otherwise
The objection is always the same: narrowing feels like turning away money. Three things are true that the fear ignores. First, broad positioning was never winning you the broad market; it was making you invisible in all of it. You cannot lose revenue you were never going to be named for. Second, the narrow query converts at a completely different rate, because the buyer asking "who fixes intake conversion for personal injury firms" is not researching, they are choosing. The conversion gap between AI-referred, high-intent visitors and generic organic traffic, which we broke down in the AI referral economy, is the same physics: specificity of intent is what converts, and narrow positions harvest specific intent.
Third, the premium follows the narrowing. A category of one is, by construction, comparison-proof: there is no like-for-like alternative to benchmark your fee against. That is the pricing ladder we mapped in pricing power: the premium of being the named recommendation, and niching is the single fastest way to climb it. Fewer clients, better clients, at rates set by you rather than by a shortlist.
What if the lane is taken? Contest or sidestep
Sometimes the audit reveals an incumbent. Decide with evidence, not emotion. If the assistants name them consistently, across models, with rich justification drawn from years of work, sidestep: add a qualifier, shift the context, claim the adjacent lane. Fighting a deep record head-on is the slowest possible route to being named. But if the naming is inconsistent, thinly justified or varies by model, the lane is held, not owned. AI answer sources churn heavily, with 40 to 60% of cited sources rotating month over month according to Semrush's AI Visibility Index, as reported by Similarweb. Churn is opportunity: the engine is still shopping for its durable answer, and the deeper record wins the audition.
Naming the lane so machines can repeat it
A category of one needs a handle: the short phrase that buyers, hosts and machines can all repeat without mangling it. Get the language wrong and you can own the expertise while losing the query. Two rules govern the choice. First, buyer-phrased beats clever. "Intake conversion for personal injury firms" will be typed; a trademarked coinage like "ClientFlow Alchemy" will not, and an assistant asked the buyer's question has no reason to reach for your private vocabulary. Second, the phrase must appear, verbatim and often, in the record you build: your bio line, your canonical page, your article titles, the way podcast hosts introduce you. Machines learn associations from repetition across independent sources, and a consistent phrase is the thread that ties the whole record to your name.
Then write the piece that defines the lane. Every young category has a moment where no definitive explainer exists, and the person who writes it collects an outsized share of the citations that follow. Research on generative engines backs the instinct: the Princeton GEO study found that content with quotable structure, statistics and clear sourcing earns measurably more visibility in AI answers. Your defining piece should do three jobs at once: name the problem in the buyer's words, lay out your method for solving it, and stake the claim that this is a distinct specialism with distinct rules. That single asset becomes the thing assistants paraphrase when they explain your lane, and the thing they attribute when they name you.
A quiet test that the naming has worked: ask an assistant to explain the problem, not to recommend anyone. When the explanation starts echoing your framing, your terms and your way of cutting up the topic, the category is forming around you, and the recommendation usually follows a few months behind the vocabulary.
Holding the category, then widening it
A category of one is rented on the same terms as every other AI visibility asset: it decays without maintenance and it compounds with it. Holding the lane means refreshing the record, adding new proof, and re-running your money queries monthly so drift is caught early. Widening happens on top of holding: once the engine reliably names you for the beachhead query, your corroborated authority starts bleeding into adjacent answers, and each neighbouring lane costs less to claim than the first one did. Sequencing that expansion over a year, which signals to build in which quarter and when to claim the next lane, is exactly what the 12-month PEO plan lays out. And if you would rather have the audit, the niche selection and the record-building run for you, that is the shape of a full PEO engagement.
The strategist's summary: in a market where machines must commit to names, the surest way to be the name is to ask a question only you can answer, and then answer it in public, better and longer than anyone else.
Questions
How narrow should my niche actually be? +
Will niching down shrink my business? +
What if someone already owns the niche I want? +
See what AI says about you today.
Start with a reading. We show you the words the engines return about your name, then map the fastest signal to move.
Get named →