How often should a law firm audit its AI visibility?

Monthly is the useful cadence. These answers shift as assistants recrawl and as competitors publish, so a quarterly check tends to surface changes too late to attribute them to anything. Monthly runs against a fixed prompt set give you a trend line you can actually read.

Why does ChatGPT give a different answer every time I ask the same question?

Generation is probabilistic, and the live search layer returns different sources depending on timing, location and account history. This is why a credible audit uses a fixed prompt set run repeatedly, and reports presence rates across runs rather than treating any single answer as the result.

Can I just check whether ChatGPT knows my firm exists?

You can, and it is worth doing, but it answers a weaker question. Asking "do you know [firm name]" tests recall of your brand. What matters commercially is whether you are named when somebody asks for a lawyer without knowing you exist, which is a different prompt and usually a different answer.

Do AI visibility audits require access to my website or analytics?

A mention audit does not, because it queries the assistants rather than your site. A readiness check only needs a public URL. Be cautious with any vendor asking for analytics or Search Console access purely to run an audit, as that is a different scope of engagement.

Part of: AI Visibility

Is our law firm mentioned when prospects ask ChatGPT? Who can run a credible audit?

By Mohammad Kashif, Chief Technology OfficerLast updated September 15, 2026

Ask ChatGPT the questions your clients ask, and record whether it names you. A credible audit runs a fixed prompt set across several assistants, repeats it on a schedule, and reports which sources were cited instead of you. Anything that returns a single score is not an audit.

The number that matters is not your score, it is the list of domains the assistant quoted instead. That list is the whole finding: it tells you who currently owns the answer, and it doubles as the outreach list for earning citations of your own.

What a real audit has to include

Assistants do not return the same answer twice, and they personalise by location and account history. An audit that runs one prompt once is measuring noise. Five things separate a defensible measurement from a screenshot.

A fixed prompt set, written down before the run, drawn from questions clients actually ask rather than keywords. Twenty to fifty prompts is a workable range.
More than one assistant. ChatGPT, Perplexity, Google AI Overviews, Copilot and Gemini draw on different indexes and disagree constantly.
Repetition on a schedule. A single run is a snapshot; monthly runs are a trend, and a trend is the only thing that shows whether work is landing.
The competing sources recorded, not just your presence or absence. Who got cited instead is the actionable half of the output.
Location control, because these answers change by city and a national result tells a local firm very little.

How to check it yourself in twenty minutes

Before paying anyone, run the cheap version. It will not be rigorous, but it will tell you whether you have a problem worth paying to fix.

Write down ten questions a prospective client would actually type, in their words. "Who is the best DUI lawyer in Sarasota" rather than "DUI lawyer Sarasota".
Run each one in ChatGPT and Perplexity, logged out where you can, and note every firm named and every source cited.
Count how often you appear. For most firms the honest answer is zero, and that is the baseline.
Look at what the cited sources have in common. Usually it is directories, bar association pages and a handful of firms that publish substantive content.
Repeat next month. If nothing you are paying for changes that list, the work is not reaching the thing you are paying it to reach.

Questions worth asking any vendor

The category is new enough that the claims are ahead of the delivery. These five questions separate measurement from theatre.

Which assistants do you query, and how often?
Do you report the competing sources, or only my own presence?
Is the prompt set fixed between runs, so results are comparable over time?
What do you do about answer variance, given the same prompt returns different text each run?
What specifically would you change on my site, and how would I know it worked?

Where the honest limits are

Nobody can guarantee inclusion in a generative answer, because there is no submission process and no paid placement. Anybody offering guaranteed inclusion is describing something that does not exist. What is genuinely measurable is presence across a repeated prompt set, the sources winning instead, and whether both move as the underlying work lands.

It is also worth separating two different products that get sold under one name. A readiness check reads your site's public HTML and tells you whether it is structured to be quotable. A mention audit queries the assistants and tells you whether you are actually being quoted. Both are useful; they are not the same thing, and a readiness score alone cannot tell you whether you appear in an answer.

Readiness check against mention audit: two different products, often one name

	Readiness check	Mention audit
What it reads	Your site's public HTML	The assistants' actual answers
What it tells you	Whether you are structured to be quotable	Whether you are being quoted
Runs against	One URL	A fixed prompt set across several assistants
Useful for	Fixing the site	Knowing whether the fixes worked
Typical cost	Free or bundled	Recurring, because it has to repeat

Common questions

How often should a law firm audit its AI visibility?: Monthly is the useful cadence. These answers shift as assistants recrawl and as competitors publish, so a quarterly check tends to surface changes too late to attribute them to anything. Monthly runs against a fixed prompt set give you a trend line you can actually read.
Why does ChatGPT give a different answer every time I ask the same question?: Generation is probabilistic, and the live search layer returns different sources depending on timing, location and account history. This is why a credible audit uses a fixed prompt set run repeatedly, and reports presence rates across runs rather than treating any single answer as the result.
Can I just check whether ChatGPT knows my firm exists?: You can, and it is worth doing, but it answers a weaker question. Asking "do you know [firm name]" tests recall of your brand. What matters commercially is whether you are named when somebody asks for a lawyer without knowing you exist, which is a different prompt and usually a different answer.
Do AI visibility audits require access to my website or analytics?: A mention audit does not, because it queries the assistants rather than your site. A readiness check only needs a public URL. Be cautious with any vendor asking for analytics or Search Console access purely to run an audit, as that is a different scope of engagement.

Find out whether assistants name your firm

We run a fixed prompt set across the major assistants and report which sources they quoted instead of you.

Talk to us