AI Search Visibility by Country, State, and City with Personas and Languages

Assistants do not have a map pack. They still answer as if the user were in a place, in a language, with a job. We track that in Promptwatch with country, state, city, personas, and language. We do not sell it as local pack rankings. The distinction is the first thing to settle with a client, because "local GEO" gets used to mean two different things. One is the assistant answering as if it were local. The other is the Google map pack. Promptwatch does the first. The map pack is a different tool and a different playbook.

For Google Business Profile, reviews, and location pages, we still run the local SEO playbook. That work feeds the public facts models repeat. This post is the assistant log. The two playbooks are not in competition. The local playbook builds the facts the model reads. The assistant log measures whether the model read them and named you. A client who skips the playbook and asks us to measure visibility is a client measuring an empty shelf.

Where geo tracking sits

Country-level tracking is on all paid plans. State and city are on Professional and above (and on the agency plans). Essential at $95/mo is enough to start a national program. Professional is $245/mo when a second locale or a city split matters. Business is $579/mo. Kick-off is $199/mo with unlimited projects and prompts, 10 seats, and state/city included. Explore is free: 10 ChatGPT prompts, one country conversation at most. G2: 4.7/5. The tier jump that matters is Essential to Professional. Essential answers "are we visible nationally." Professional answers "are we visible in Hamburg and in Bavaria, in German, for a finance buyer." Those are different questions, and the second one is the one that needs state and city.

AI Overviews can be tracked for a region. AI Mode is available across supported countries. We still split Overviews from ChatGPT on the slide. Search Console generative reports stay in the Google appendix. The split is the habit to keep. A client who sees one blended "AI visibility" number will ask why it moved, and the answer is different for Overviews and for ChatGPT, so we keep them on separate lines from the start.

Languages we see on accounts include English, Dutch, German, French, Portuguese, Japanese, and others the product lists (Thai, Vietnamese, Chinese simplified and traditional, and more). Per-generation language on Content Agents is a separate switch when we draft. We do not mix NL and EN prompts in one average. The reason is that a Dutch prompt and an English prompt for the same product return different competitors and different citations, and an average of the two is a number that describes neither market. We report them as two columns, not one.

Personas are how we prompt like the buyer: practice manager versus patient, IT versus finance. Same prompt stem, different persona, different answer. That is not a city. Do not stack city and persona on day one or the response budget disappears. A persona changes the answer because it changes who the model thinks is asking. A city changes the answer because it changes where the model thinks the answer should come from. They are different axes, and turning both on at once on fifty prompts burns through the response budget before you learn anything.

Paid engines: ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, AI Overviews, AI Mode. Daily refresh. No instant alerts. No $29 Promptwatch plan. Daily refresh is the cadence we pay for. Instant alerts are not a feature we promise, because an alert that fires on every change is an alert that fires on noise.

What we will not call local GEO

A map-pack rank. A BrightLocal or GeoRanker AI SKU we have not verified. A doorway page for every city the model might mention. If we cannot write something true about that city, we do not target it. The playbook already says that. A doorway page for a city we have no real presence in is a page the model can read and a page a human will dismiss in two seconds, and the model learns from the human dismissal too.

Otterly at $29 with Gemini as an add-on and lag is a poor multi-city log. Peec at $95 is a mention tracker. Profound Starter at $99 annual is ChatGPT only. Fine if the client already bought it. Insufficient if they asked for Gemini in German in two states. The three competitors each miss a piece of the geo split, and the piece they miss is usually the piece the client asked for, which is why we end up on Promptwatch for this work.

Visitor analytics can show location of the people who clicked. That is not the same as the simulated city on the prompt. We keep those columns apart. One is where the model thought the user was. The other is where the visitor who clickedwas. They are different signals, and conflating them is how a program reports a city it never targeted.

How this changes the prompt list

We freeze a national list first. Then we duplicate the five revenue prompts into the one city or language that sales already loses. We do not duplicate all 50. Query fan-outs in the second locale often name a local publisher we are absent from. That is a citation ticket, not a new location template. The five-prompt duplication is the discipline that keeps the program from sprawling. Fifty prompts across fifty cities is a budget that produces no signal. Five prompts in the one city that matters is a budget that produces a decision.

Webflow and Framer can take a reviewed location page. WordPress publish is still coming soon. Autopublish stays off. Invented neighborhoods stay out. The reviewed location page is the unit. An autopublished location page is a page nobody checked, and an invented neighborhood is a page with no real facts, and both are pages we will not ship on a client domain.

FAQ

Is city tracking a replacement for GBP?

No. GBP is still the local homepage for Google. City-level Promptwatch checks are how we see whether ChatGPT or Gemini names you when the user is treated as local. Do the playbook. Then measure the assistant. The two work in sequence, not in substitution.

Can we do city tracking on Essential?

Country only on Essential. Move to Professional, Business, or Kick-off for state and city. We will not pretend otherwise to win a SOW. A SOW won on a feature the tier does not have is a SOW that fails at the first report.

How we run it

1. Ship the local playbook items that are broken (GBP, NAP, real location copy).
2. Load a national prompt list in Promptwatch. Baseline two weeks.
3. Add one city or one language, plus one persona if the buyer types differently.
4. Read mention vs citation per locale. Do not average them.
5. hello@1001seomedia.com if you want the geo split on the retainer. Tell us the countries you sell in. Do not ask for map-pack positions from this tool.