On this page
Hendrick’s Violin Shop repairs and sells stringed instruments in a college town of sixty thousand people. In a good week, four or five new customers walk in for a repair, a bow rehair, or a rental referral from the university’s music department. If Hendrick’s tried to build an AI prompt set the way most guides teach it, the first step would already fail: there is no keyword tool that returns a monthly search volume for “violin repair shop,” because in a town this size, close to nobody is typing that phrase into anything with a volume estimate attached.
That is not a Hendrick’s problem. It is the normal condition for any business in a narrow local category: a probate attorney handling one county, a single-location orthodontist, a shop that reupholsters antique furniture. The prompt-set advice built for a national brand with thousands of monthly searches to sort through does not fail gracefully when the category shrinks. It just runs out of data to work with.
Why “search volume” doesn’t exist for a category like yours
Google-era SEO ran on a real number: type a keyword into a tool, get back an estimate of how many people search it a month, and prioritize accordingly. Nothing equivalent exists for AI assistants. None of the major engines publish query-level volume the way Google Search Console does for keywords, and Conductor’s own breakdown of “AI prompt volume” tools concludes that any product reporting a specific number is modeling it from a small panel of sampled conversations, not reading real usage logs - useful for spotting broad trends, not for deciding whether “best violin shop near [college town]” gets asked twice a month or twice a year.
For a category with real national volume, that modeling noise mostly washes out; there is enough underlying signal that an estimate lands in the right neighborhood even if the exact number is wrong. For a category with ten buyers a week in one town, there was never enough volume to model in the first place. Waiting for a tool to tell you which questions matter is waiting for a number that does not exist.
The buyers are asking anyway
The absence of a volume figure does not mean the category is quiet. BrightLocal’s 2026 Local Consumer Review Survey found that 45% of US consumers say they’ve used a tool like ChatGPT, Gemini or Perplexity for a local business recommendation in the past year, up from just 6% the year before, with ChatGPT the single most common choice at 31% and Google’s AI Mode close behind at 23%. That jump happened across every category, including the small ones with no keyword data to show for it. A rise that fast in one year means the buyer asking about your category is very likely doing it now, whether or not any tool can quantify how often.
The same BrightLocal survey found that 88% of people who get an AI recommendation for a local business say they verify it before acting, mostly by checking real reviews on the platforms they already trust. That matters for what to track, not just whether to: the AI answer is functioning as a shortlist a buyer then checks, which is exactly the job a best-of or recommendation prompt does. It is a weaker signal for a definitional prompt nobody double-checks anyway.
Build the list from where buyers actually talk, not from a keyword tool
For a category this size, the closest thing to a real dataset is the language your own customers already use, not a guess at what a keyword tool would have said if one existed. Pull the actual questions from your call log, your contact form submissions, your booking emails: “do you fix bows,” “how much to restring a viola,” “can you do a same-day repair before a recital.” Those are the phrasings a buyer would put to an assistant, because they are the phrasings that were already in their head before they found you.
Our guide to choosing the prompts worth tracking breaks prompts into four intents: best-of, comparison, problem-led and category. All four still apply here, but the weighting shifts hard toward best-of and problem-led. Comparison prompts (“Hendrick’s vs. [competitor]”) barely matter when there is only one other real option in town; category prompts (“what does a bow rehair cost”) matter more than they would for a national brand, because in a thin market they are often the first place a buyer’s trust gets built.
Size the list to the category, not to a formula
General guidance for AI visibility tracking suggests 15 to 30 prompts as a starting set. For a category this narrow, that number is usually too high in one direction and too low in another. Too high on distinct question shapes: a violin shop does not need thirty different ways of asking about a repair, because buyers in a town this size do not phrase it thirty different ways. Too low on geography: a real service area, even for a small shop, often spans the town plus two or three neighboring ones, plus a campus or a specific neighborhood people identify with. That means fewer intents, multiplied across more places, rather than the other way around.
The same logic that governs how a small business earns a place in a local AI recommendation applies to which questions are worth watching: test the phrasing a stranger would actually use, in the specific place they would use it, not your own business name.
Check more often, not less
A thin category cuts the other way on frequency. For a brand with thousands of monthly mentions, one odd answer on one engine on one day is noise that a larger sample absorbs. For a business with ten buyers a week, a single week where an assistant names a competitor instead of you is a meaningful share of that week’s total opportunity. The fact that different engines pull from different sources and disagree with each other matters more here, not less, because there is no volume cushion to fall back on if one engine happens to be wrong about you this week. Track the trend, not a single check: our guide to tracking topic volume and trends covers how to read movement over time instead of reacting to one answer.
Track it like the business it actually is
A prompt set built for a category this size looks different from the standard template: shorter on distinct questions, wider on geography, weighted toward the best-of and problem-led questions a buyer actually checks against reviews afterward, and reviewed more often because each answer carries more weight. Monitor drafts a starting set from your brand, category and competitors and lets you edit it down to what your buyers actually ask, and the free Visibility Check will show you, in about a minute, whether an assistant already names you for the handful of questions that matter most in a category this size.