How to track local AI visibility for clients: our agency process
Our agency process to track local AI visibility for clients: location setup, local prompts, engines, citation checks and reporting, run on Promptwatch.
Local clients ask us a simple question that has a complicated answer: "If someone near me asks ChatGPT for a recommendation, do they hear about us?" We have a process for answering it. It is not glamorous, and it depends on one rule: never judge from a single answer.
Why a single check misleads
AI answers are generated, not listed. Run the same prompt twice and you may get different businesses. Ask from a different location and the answer can change again, on engines that use local signals. A screenshot someone took on their phone tells you almost nothing about the whole picture.
So we measure the way we would measure anything noisy: a panel of prompts, run repeatedly, anchored to specific locations, aggregated over time.
Our process
Scoping the locations
We ask which locations make the money. A client with twelve outlets rarely needs twelve identical programs. We start with the top three or four by value and add more once the setup is stable.
Writing local prompts
We write them as locals phrase them: "emergency plumber in [town]", "best family dentist near me", "where to buy [product] nearby", "cheap lunch near [landmark]". We add comparison prompts, and a few brand prompts, which show what AI says when asked about the client directly. Each prompt is tagged by location and intent so reports can slice by either.
Anchoring to the location
For each prompt we need the engine to behave as if asked from the right place. In Promptwatch, country-level tracking is on every paid plan, state and city tracking starts on Business and is on every agency plan, and Local GEO/SEO targeting pins a monitor to exact coordinates with a radius from 20 meters to 5 kilometers. That coordinate targeting works for ChatGPT, Google AI Overviews and Google AI Mode; other models fall back to country, state or city, and we say so in the report. Coordinate targeting is on Business, Agency Scale and Enterprise.
Running the panel
We track daily. Local answers shift, and a weekly check can miss a drop for days. Promptwatch runs prompt panels daily per model, country and language, so we get trend lines instead of anecdotes.
Reading the citations
When a client is missing in a location, we list the cited sources. Often it is a directory listing with inconsistent details, a competitor with more review content, or a local news piece. The citation view in Promptwatch gives page-level and domain-level breakdowns, and the offsite mentions report shows where the brand is named without a link. These point us at fixes outside the client's own website.
Checking the client's own pages
We use IP-verified crawler logs to confirm AI bots reach each location page, and to read error explanations where they do not. Crawler logs start on the Professional plan and are included on every agency plan. Thin or blocked location pages are one of the most common problems we find.
Fixing and re-measuring
We correct listings, strengthen location pages with real details (services, hours, staff, directions), encourage genuine reviews, and add content that answers local questions in plain language. Then we compare visibility, position in answer and citations against the baseline.
Why we run it on Promptwatch
We run client programs on Promptwatch, and the agency plans suit multi-client local work: Kick-off at $199/mo, Growth at $399/mo and Scale at $799/mo. Each includes unlimited projects, unlimited pitch projects, unlimited prompts, 10 seats, state and city tracking, crawler logs, visitor analytics, custom reports and a Looker Studio connector. Scale adds coordinate targeting and an audit log. Agency plans do not include the Content Agent or AEO articles, so we handle content through our own workflow.
Pitch projects let us run an audit for a prospect before they sign, which helps in local sales conversations. A white-label workspace and client portal are listed for agencies, and we confirm the details with Promptwatch before promising them to a client.
For reporting, we use PDF reports of any dashboard and scheduled reports, with visitor analytics showing AI-referred visits and conversions where the client's site has the script installed.
Other tools worth knowing
Local Falcon runs grid scans, sending the same prompt from many points around a location. It is a good visual for a pitch or a review. BrightLocal's Local AI Visibility module is convenient for clients already inside BrightLocal, though it tracks three engines and caps prompts per location. We treat both as complements. They stop short of crawler logs and conversion tracking, which we need to close the loop.
What we tell clients
We say that results show direction and comparison between locations, not a guaranteed ranking. We say that coordinate-level results reflect the engine's answer from that point, not what every person sees. And we say that no one, including us, can guarantee a recommendation. Clients take that well, especially when the report shows concrete fixes and a trend.
Starting small
If you are building your own local tracking, pick two locations and fifteen prompts, run them for two weeks, and read the citations. You will learn more from that than from a month of planning. Promptwatch's free Explore plan (10 prompts, ChatGPT only) is enough to try the first run.