How many cities should you track in AI search?
A practical way to size a city-level AI visibility program: which cities earn a monitor, how many prompts each gets, and how to fit it all inside a response budget without averaging away the signal.
Once a team finds out its AI visibility tool can target cities, the first instinct is to add all of them. Forty locations, forty cities, fifty prompts each. Three weeks later nobody reads the report, the response budget is gone, and half the cities show the same answer every day.
City tracking pays off when it's selective. Here's how we'd size it.
Start by asking whether cities differ at all
Before you build forty monitors, test whether location moves the answer in your category. Pick your two most different markets, a big metro and a smaller city for example, and run the same 15 to 20 prompts in each for two weeks.
If the recommended brands and cited sources are nearly identical, your category is mostly national in AI answers. Track by country and spend the budget on more prompts. If they differ, keep going. Categories where people buy close to home, like trades, clinics and law firms, are the likeliest to split by city, but test rather than assume.
Sort cities into three tiers
Not every city deserves the same attention.
Tier A is cities that drive revenue now: your largest locations, or the markets your sales team reports on. These get a full monitor with your core prompt set and the engines you care about.
Tier B is cities you're expanding into or where you think you're losing. These get a smaller set, the five to ten prompts closest to a purchase.
Tier C is everywhere else. Don't track these continuously. Rotate one or two in each quarter as a spot check, then drop them again.
For most multi-location brands that's three to eight Tier A cities and a similar number in Tier B. That's enough to see patterns without drowning in rows.
Do the response math before you buy
Every check of one prompt on one model in one place is one response. The budget runs out faster than people expect.
Take Promptwatch, which refreshes daily and includes city and state targeting from its Business plan and on every agency plan. Location sits on a monitor, so each city is its own monitor. Twenty prompts on three models in one city uses about 1,800 responses a month (20 x 3 x 30).
- Business ($579/mo) has 42,000 responses and 350 prompts. The prompt cap binds first: about 17 cities at 20 prompts, or more if Tier B cities get fewer prompts.
- Agency Kick-off ($199/mo) has 10,000 responses and unlimited prompts: about 5 cities at that shape.
- Agency Growth ($399/mo) has 25,000 responses: about 13 cities.
- Agency Scale ($799/mo) has 65,000 responses: about 36 cities.
Essential and Professional are country-level only, so they don't enter this calculation.
A tiered plan stretches the budget much further. Five Tier A cities at 20 prompts plus eight Tier B cities at 8 prompts is 164 prompts. On three models daily that's about 14,760 responses a month, which fits on Growth with room to spare, or on Business within both caps.
Other tools meter differently. Local Falcon charges one credit per grid point, so a 9x9 grid is 81 credits per keyword per scan. BrightLocal and Synup price per business location, which means a city without an address isn't trackable at all. Run the same arithmetic for whatever you're evaluating.
Don't stack every variable on day one
Promptwatch monitors can also carry a persona and a language. It's tempting to cross cities with personas: five cities times three buyer types is fifteen monitors. Each combination costs responses, and you won't learn which variable moved the answer. Add cities first. Add a persona to one city only once you suspect buyer type matters there.
Report cities one at a time
The point of tracking cities is to see differences, so don't erase them in the report. Show one row per city, per engine. A brand that's recommended in Denver and missing in Colorado Springs doesn't have "moderate" visibility. It has a Colorado Springs problem.
For each weak Tier A city, follow the chain. Which sources did the answer cite instead of you? Did AI crawlers fetch your page for that city? Agent Analytics in Promptwatch logs those visits, with errors. And when AI did send visitors from that market, did they convert? That chain turns a city score into a to-do item.
Review every quarter
Cities move between tiers. A Tier B city where you've closed the gap can drop to a lighter prompt set. A Tier C spot check that shows a strong competitor might earn a monitor. Prune prompts with no search volume, which Promptwatch shows on each prompt, before adding new ones.
The answer
Fewer than you think, chosen on purpose, with prompts sized to the tier. For most brands that's five to fifteen cities. Fit them to a response budget, report them separately, and trace the weak ones to a cause. Promptwatch is the tool we'd use for that, since the city data, crawler logs and conversion data sit in one workspace. Size your plan at promptwatch.com.