How to track Perplexity, Gemini, and Copilot for local queries

Promptwatch's Local GEO/SEO targeting pins a monitor to an exact point, and three engines answer as if asked from there: ChatGPT, Google AI Overviews and Google AI Mode. Every other model in the monitor uses its country, state and city instead. That includes three engines most local programs also care about: Perplexity, Gemini and Microsoft Copilot.

That's no reason to drop them. It's a reason to design the setup so you always know which answers came from which location grain. Otherwise you'll read a difference in precision as a difference between engines.

What "falls back" means in practice

Say you pin a monitor to a bakery's front door with a 100 m radius and set the city to Manchester. ChatGPT, AI Overviews and AI Mode get the pin. Perplexity, Gemini and Copilot get "Manchester". Same prompt, same day, same monitor, but two different location questions: one asked from the street, the other from the city as a whole.

If the bakery shows up in ChatGPT and not in Perplexity, there are at least two explanations. Perplexity may recommend other places. Or a city-wide answer has less room for a small neighborhood shop than an answer for that street. One monitor can't separate the two. Two monitors can.

The two-monitor design

For each location, build:

A coordinate monitor with ChatGPT, AI Overviews and AI Mode. Every answer in it reflects the pin.
A city monitor with Perplexity, Gemini and Copilot, plus ChatGPT.

The ChatGPT in the second monitor is the bridge. It's the one engine you now see at both grains. If ChatGPT gives similar answers at the pin and at city level for this location, the grain doesn't change much here, and putting Perplexity's city answer next to ChatGPT's pin answer is reasonably fair. If ChatGPT's two answers differ a lot, you know the city-level engines are working with a coarser question, and you read their results with that in mind.

The bridge costs one extra model in one monitor. Responses are prompts times models times runs, and runs are daily, so 20 prompts on one extra model is about 600 responses a month. Business includes 42,000.

If your customers use Amazon Alexa or Meta AI, add them to the city monitor. Promptwatch monitors both through their real interfaces, and both fall back to the city like the others.

Keep everything else identical

A comparison across monitors only holds if one thing changes. Check these before the first run:

Prompt wording. Write the set once, without place names, and copy it into both monitors. Promptwatch lets you copy prompts between monitors.
Language. Same tracking language in both.
Persona. If you use one, attach it to both. Promptwatch's "Copy to new monitor" pre-fills location, models and frequency but not the persona, so set it again by hand.
City. Give the coordinate monitor the same city as the city monitor. State is a US-only field, so outside the US the city is the finest setting the fall-back engines get.
Start date. Create both monitors on the same day so the baselines line up.

During model selection, some models read "Not available in this region" for certain countries. Check this when you pick models, because it decides which engines you can compare outside the US.

Two branches, one city

Know this limit before you promise anyone a per-store report. If a brand has two branches in the same city, coordinate monitors can separate them on ChatGPT and Google's two AI surfaces. On Perplexity, Gemini and Copilot both branches get the same city context. You can still see which branch gets named, but on those engines you can't attribute a difference to each store's surroundings.

So report the city-level engines once per city and the coordinate engines once per store. Don't copy a city number into every store's row and call it store data.

Reading engines side by side

Filter per monitor and per model. Promptwatch's custom dashboards save filters for models and monitors, and the heatmap compares any two dimensions, such as models against topics, which is a quick way to put engines next to each other. Read Visibility Score and Position in Answer for each engine separately.

Then look at citations. When engines disagree about a location, check whether they also cite different sources. Citation analytics shows the pages and domains behind each answer, and the Citations view has a monitor filter, so you can compare what Perplexity cites for the city with what ChatGPT cites at the pin.

Two weeks of daily runs is the minimum before calling any engine "weak" for a location. For engine-by-engine tool comparisons, see our roundups of Perplexity tracking tools, Gemini tracking tools and Copilot tracking platforms.

What to do when one engine is missing you

Say Perplexity never names the location while ChatGPT does at the pin and at city level. That's a Perplexity finding worth chasing. Check whether PerplexityBot or Perplexity-User has reached the location page at all. Promptwatch's Agent Analytics logs IP-verified crawler visits per page, with errors and failure explanations. If the bots arrive and the page still isn't cited, compare the sources Perplexity does cite with the ones ChatGPT uses. For Copilot, Search Insights in 360 Insights (paid brand plans) overlays Bing Webmaster data, including Microsoft Copilot traffic, on AI crawls and citations.

Which plan this needs

City monitors are on Business ($579/mo) and every agency plan from Kick-off ($199/mo). Coordinates are on Business, Agency Scale ($799/mo) and Enterprise. So the full mixed design runs on Promptwatch Business for a brand, or Agency Scale for an agency. An agency on Kick-off or Growth can run the city half for every engine now and add coordinate monitors after upgrading.

The reason to keep both halves in one workspace is that they feed the same citation analytics, crawler logs and visitor analytics. When Perplexity does start naming a location, visitor analytics shows whether those Perplexity-referred visitors convert, which is the number that decides whether the engine deserves more of your time.