How to run a local GEO audit with Agent Chat

A local GEO audit asks one question: when someone near this location asks an AI assistant for what we sell, do we come up, and if not, why not? Answering it means checking AI answers, citations, crawler access, traffic and page content for one place. Done by hand that's an afternoon of tabs and exports.

Agent Chat in Promptwatch is a conversational analyst over the project's live data. It runs multi-step workflows, can delegate parts of a question to read-only subagents, renders charts, builds and edits dashboards, and generates PDF reports. That makes it a good fit for an audit you'll repeat per location. This guide gives the questions to ask, in order, for one location. Run the same sequence for the next location and the results compare.

Before you start

The audit can only read what the project collects, so check three things.

First, a monitor for the location. Country-level monitors are on every paid brand plan. City and state monitors start on Business ($579/mo) and are on every agency plan from Kick-off ($199/mo). For a single branch, Business, Agency Scale and Enterprise can also pin a monitor to exact coordinates, with a radius from 20 m to 5 km, which affects ChatGPT, Google AI Overviews and Google AI Mode. Other models fall back to country, state or city.

Second, crawler logs. Agent Analytics needs an integration through your CDN, host or the WordPress plugin, and it's included from Professional ($245/mo) and on every agency plan. Essential has none, so skip step four on that plan.

Third, two weeks of data. A first-day audit has nothing to compare against.

Agent Chat itself is on every paid brand plan and every agency plan. Besides its own view, it opens as a side panel with ⌘J on any page.

The audit, one question at a time

Ask these one after another in the same chat, so each answer builds on the last. The examples use a hypothetical "Bristol" monitor.

1. Where do we stand?

"For the Bristol monitor, show Visibility Score, Share of Voice and Position in Answer for the last 30 days by model, with a line chart of visibility over time."

This tells you whether the problem is everywhere or concentrated in one engine. A location that's visible on Gemini and invisible on ChatGPT needs a different fix from one that's invisible everywhere.

2. Who wins instead?

"Which competitors have the highest Share of Voice in Bristol's answers over the same period, and on which prompts do they appear when we don't?"

Promptwatch auto-discovers competitors, so this can name a local rival nobody listed. Our guide to monitoring local competitors covers cleaning that list up with brand merging.

3. What do the answers cite?

"List the domains and pages most cited in Bristol's answers. Separate our own pages, competitor pages, directories, and Reddit or YouTube sources. Which of our pages are cited, and for which prompts?"

This is usually the most useful answer in the audit. It shows which sources the engines leaned on for this place. If a local directory dominates, the next step is off your site. If a competitor's location page dominates, it's on your site.

4. Can AI crawlers reach the location page?

"For our Bristol location page, show AI crawler visits over the last 30 days by bot, any errors with their explanations, and the page's citation rate."

Agent Analytics logs IP-verified crawler hits, filtering out spoofed user agents, from bots such as GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Agent. If OAI-SearchBot hits an error on the page, the content doesn't matter yet. Fix access first. Our post on agents that read crawler logs and propose robots.txt fixes goes into that.

5. Does AI send anyone to it?

"How many AI-referred visits did the Bristol page get, from which assistants, and how many converted?"

Visitor analytics attributes AI-referred traffic and conversions through a script, the GTM template or the WordPress plugin. A page that's cited but never visited may be cited for the wrong prompts.

6. What's missing from the page?

"Run content gap analysis for the prompts in the Bristol monitor against our Bristol page. What do the answers cover that the page doesn't?"

Gap analysis scores coverage against the prompts. Read the result critically. If the gap is "parking near the clinic" and you don't know the answer, the fix is to ask the branch, not to have an agent write something plausible.

7. Anything technical?

"Does the latest site audit flag anything on the Bristol page or its internal links?"

Weekly Site Health audits run 21 checks and report what changed since the last run.

Once the answers are in, ask Agent Chat to package them: "Build a dashboard called 'Bristol audit' with the charts from this conversation, and generate a PDF report summarizing the findings with the three most important fixes."

The dashboard stays live, so next month's audit starts from it. The PDF goes to the branch manager or the client. If you want the fixes tracked, ask the agent to add them to the Action Board with the evidence attached; our Action Board guide for locations explains how to assign them.

Run the same audit from your own agent

Some teams want the audit in their own tools: a Claude project, Cursor, or a script that loops over every location. Promptwatch's hosted MCP server at server.promptwatch.com/mcp exposes 110+ read and write tools over OAuth or an API key, and there's a REST API v2 for scripts. Promptwatch also lists a Claude Connector and a ChatGPT plugin. Our MCP connection guide covers setup.

With MCP, the seven questions above become a script your agent runs per location, writing one report each. For an audit you only want to read, use a read-only API key so the agent can't change monitors or prompts by accident.

Why run it here

You could assemble the same audit from four tools: a prompt tracker, server logs, an analytics suite and a content grader. You'd spend the afternoon joining exports on URL and date. Running it in Agent Chat means one conversation over data that already shares a project, a location and a page. That's why we'd run local audits on Promptwatch: Business if you need city-level monitors for your own locations, or an agency plan if you're auditing locations for clients.