PricingContactGet Started Free

Blog / AEO & GEO / AI Search

How to Check if ChatGPT Recommends Your Law Firm

A single screenshot of ChatGPT answering “who’s the best lawyer in my city” proves almost nothing about your firm’s AI visibility. The same question can return a different answer five minutes later, from a different device, or with slightly different wording. If you want a real read on whether AI assistants recommend your firm, you…

GrowthAgents AINovember 17, 20264 min read

A single screenshot of ChatGPT answering “who’s the best lawyer in my city” proves almost nothing about your firm’s AI visibility. The same question can return a different answer five minutes later, from a different device, or with slightly different wording. If you want a real read on whether AI assistants recommend your firm, you need a repeatable panel of prompts run over time, not one lucky (or unlucky) screenshot.

Why doesn’t one ChatGPT answer settle the question?

Large language models generate a new response each time rather than replaying a stored answer, which is “not a glitch” but “how every large language model works by design,” according to a 2026 analysis from Similarweb’s AI search team. Consumer ChatGPT runs with randomness on by default, so “no two responses will be exactly the same,” and even with randomness turned all the way down, one test still produced 80 different responses out of 1,000 identical runs.

The same source notes that saved chat memory, custom instructions, and even which model version an account happens to use can all shift the answer — “The prompt is identical. The context is everything.” A managing partner’s ChatGPT, a prospective client’s, and a marketing coordinator’s can legitimately give three different answers to the same question, for reasons that have nothing to do with your firm’s actual visibility.

What should a real test panel look like?

Instead of one question, build a small, fixed set of prompts that mirror how real prospects actually search. A useful starting panel covers three types:

  • Practice area + city prompts — “best [practice area] lawyer in [city],” “top-rated [practice area] attorney near [neighborhood].”
  • Comparison prompts — “[Firm A] vs [Firm B] for a [practice area] case,” “which [city] law firm handles [practice area] cases.”
  • Plain client questions — “I was in a car accident in [city], who should I call,” “do I need a lawyer for [situation] in [state].”

Keep the wording fixed once the panel is set — rewording on the fly is itself a variable, not a shortcut to a better answer.

How do you run it and what do you log?

Run the full panel on a schedule — weekly is a reasonable starting cadence — and log the same fields every time:

  • Date, prompt, and which assistant/engine you used (ChatGPT, Perplexity, Gemini, etc.)
  • Whether your firm was mentioned at all
  • Whether your firm was mentioned with a cited URL, and which page it pointed to
  • Which competitor firms were named, and in what order
  • Whether the description of your firm (practice areas, location, specialty) was accurate

A spreadsheet with one row per run is enough to start. The point is a pattern across dozens of runs, over weeks, not one data point — that pattern tells you whether mentions are trending up, down, or holding steady for a given prompt type.

What’s the difference between a mention, a citation, and an accurate description?

These three outcomes get collapsed into “we’re in ChatGPT” far too often, but they’re not the same thing:

  • A mention is just your firm’s name appearing in the answer text, with nothing backing it up.
  • A citation is a mention paired with a specific URL the assistant points to as its source — which page it pulled from matters, because that’s the page shaping how your firm gets described.
  • An accurate description is a mention or citation that correctly states your practice areas, location, and specialty. A firm can be mentioned and cited and still be described wrong — pointed to the wrong office, assigned a practice area you don’t handle, or confused with a similarly named firm.

Logging all three separately, rather than one “mentioned yes/no” column, is what turns a screenshot habit into an actual signal.

When does checking this by hand stop working?

A panel of 15–20 prompts, run weekly across two or three AI assistants, is roughly 30–60 manual checks a week — doable for a month, painful to sustain. It tends to break down once you’re tracking several practice areas, more than one office location, or several competitors per prompt: the spreadsheet grows faster than anyone has time to fill it in accurately, and rows get skipped or logged sloppily, which quietly wrecks the pattern you were trying to build.

What can’t this method tell you?

Be clear-eyed about the limits. This panel is a sample of your own prompts, run from your own accounts and locations — it is not what any individual prospect sees at the moment they ask. It can tell you that an answer changed, not why. It doesn’t measure how many real prospects ask AI assistants this question, or what they do after reading the answer. Treat it as a directional read on your AI visibility, not a guarantee of what the next person, searching from a different city, device, or account, will see.

Does asking ChatGPT the same question twice in a row count as a test?

No. Two runs in a row tell you almost nothing — you need enough repeated runs across your prompt panel, over enough time, to see a pattern rather than noise.

Do I need to check every AI assistant, or is ChatGPT enough?

Checking only ChatGPT misses how prospects actually search, since Perplexity, Gemini, and Google’s AI Overviews all pull from different sources and can describe your firm differently. Running the same panel across each one is more work but gives a more honest picture.

If a competitor is named more often, does that mean they’re getting more clients from AI search?

Not necessarily. A mention count is a visibility signal, not a conversion number — it tells you how often a name comes up, not what happens after someone reads it.

If keeping this panel updated by hand isn’t realistic for your team, get a free AI Search Analysis from GrowthAgents to see how your firm currently shows up across AI answers.

A city seen from high above

Can AI search read your site?

Free readiness check. About a minute.