What AI assistants say about your business, and how to check

AI answers change from run to run. Studies that repeat the same local prompt find only about a fifth to a third of the named businesses overlap between runs, and the answer also changes with where the asker appears to be. So a single "who's the best plumber in Albuquerque" check, from your office, signed in, mostly tells you what the assistant said once, to you.

The fix is a small panel, repeated: enough prompts, enough runs, and a simple record of what came back. That is the same method we run monthly for clients, and you can run a starter version yourself for free.

Write 15–30 prompts. Mix the head terms ("best [service] in Albuquerque"), the specific jobs ("who can [job] in Rio Rancho this week"), price questions ("what does [service] cost in Albuquerque"), comparisons ("[your business] vs [competitor]"), and trust checks ("is [your business] good?"). Run them from more than one part of town if you can, the Northeast Heights and the Westside behave differently, and logged out or in a clean browser profile.

Run each prompt 3–5 times on each assistant you care about: ChatGPT, Google AI Mode or AI Overviews, Gemini, Perplexity, Claude, and Copilot. It takes an afternoon the first time. Monthly after that, it takes an hour.

For every run, record whether you were named, where in the list, which competitors were named instead, every source the assistant cites, and any fact it got wrong about you: hours, prices, services, service area. The cited sources are the gold. Each source that names a competitor and not you is a concrete task: a directory to join, a review site to build up, a list to earn a place on, a licensing record to fix.

Turn the notes into three numbers you track monthly: how often you're mentioned, your share of mentions against your top three competitors, and a count of factual errors. The error target is zero, and every error gets fixed at its source listing, not argued with on your own site.

Work the source gaps first. Get listed, completely, where assistants already look for your trade, with the same name, phone, hours, and services everywhere. Publish the prices and policies the assistants couldn't find. Fix the wrong fact at the listing that fed it. Then re-run the panel next month and watch the mention rate and the error count.

If you'd rather not do this yourself, our free Agent Readiness Check runs this panel for your business and emails you the report within two business days. Ongoing, it's the core of our AI Visibility Monitor, from $279/mo.