How to write local prompts for AI visibility tracking
How to write the prompts behind a local AI visibility program: unbranded category questions, near-me variants, competitor comparisons, intent and prompt types, and why wording must stay frozen across locations.
Most local AI tracking goes wrong before the first result comes in. The prompts are branded, or they're keyword strings nobody would type into ChatGPT, or every city gets slightly different wording so the results can't be compared. Fixing that costs nothing but an afternoon. Here's how we write a local prompt set.
Start with unbranded category questions
The core of a local set is questions that don't name you. "Best family dentist in [city]" or "who should I call for a leaking roof" measure whether an engine recommends you to someone who doesn't know you yet. That's the number you're trying to grow.
Branded prompts ("is [your business] any good", "[your business] opening hours") measure something else: what engines say about you when asked directly. Keep a few, because a wrong address or a bad summary in those answers is worth catching. Keep them in their own group, though, so they don't inflate your visibility score. We explain the split in branded, organic and competitor prompts.
Write near-me variants on purpose
Local buyers ask in two ways. Some type the place ("plumber in [town]"). Others type "near me" or nothing at all and let the engine work out where they are. Your set should have both.
For the "near me" and placeless versions, the location has to come from the tracking tool, not the prompt. That's the point of a location-set monitor: the question reads exactly as a customer would type it, and the tool supplies the place. For the versions that name the town, use one template and swap only the town name.
A reasonable starting mix for one service line is roughly half placeless or "near me" prompts and half with the town named. Adjust once you see which shape triggers more local answers in your category.
Add competitor comparisons
People ask engines to compare businesses: "[competitor] or [another competitor] for a kitchen refit in [town]", "cheaper alternative to [competitor] nearby". Include a few that name your main local rivals. They show what engines say about the businesses you lose jobs to, and whether you get offered as the alternative.
Cover the intents that lead to a booking
Sort your prompts by what the person wants to do next. A workable set of intents for local services:
- Discovery: "good physio for running injuries near me"
- Urgency: "locksmith open now", "same-day boiler repair"
- Price: "affordable wedding photographer in [town]"
- Fit: "dentist that's good with nervous patients", "vet that sees rabbits"
- Comparison: the competitor prompts above
Urgency and fit prompts are often where a small local business can beat a bigger one, because the answer depends on specifics a chain's page doesn't mention. They're also the prompts most worth tracking daily.
Freeze the wording across locations
This is the rule people break most. If Leeds gets "best emergency plumber" and Bristol gets "top emergency plumbers", you can't tell whether a difference comes from the city or the wording. Write the set once. Copy it unchanged into every location. When you edit a prompt, edit it everywhere on the same day and note the date, so a jump in the trend line has an explanation.
The same applies to languages. If a city needs a second language, translate the frozen set faithfully rather than writing new questions. Tracking local AI search in multilingual cities covers that case.
Use real data to pick wording
Your own guesses about phrasing are a weak source. Better ones are your Search Console queries, the "People also ask" boxes on your local keywords, and the questions customers ask on the phone. Look at a competitor's location page too, since it shows what they think customers want to know.
Then prune by demand. Two wordings of the same question rarely deserve two slots. Keep the one people actually use and spend the saved slot on a new intent. For sizing the whole set, see how many prompts should you track.
Doing this in Promptwatch
Promptwatch has most of this built into the prompt workflow. Prompt Explorer suggests prompts from your Search Console data, your ranking keywords, People Also Ask, or any page you point it at, including a branch page or a competitor's. Prompt Volumes adds a search volume, a difficulty score and citation rates across AI search, Google and Bing to each prompt, which makes the pruning step a sorting exercise.
Each prompt gets a type (organic, brand or competitor comparison) and an intent. Tags and topics let you group prompts by service line, so "emergency repairs" and "installations" can be reported separately. Personas are there too, and you can stack them, though we'd add one only when a buyer type clearly changes the answer.
The frozen-wording rule is easy to follow because location lives on the monitor, not in the prompt. You write the panel once and attach it to a monitor per city, each running daily across the models you pick. Country-level monitors are on every paid brand plan from Essential ($95/mo). City and state monitors need Business ($579/mo) or an agency plan from Kick-off ($199/mo). If the neighbourhood matters, coordinate targeting on Business and Agency Scale makes ChatGPT, Google AI Overviews and Google AI Mode answer as if asked from a set point.
Write the set before you pick the plan. Once you know how many prompts and locations you need, the plan choice is arithmetic, and how to structure projects and monitors for a multi-location brand walks through it. Prompt Explorer and the plans are on promptwatch.com.