AI search self-check
How can a lawyer check what ChatGPT and Google's AI say about their firm?
A 30-minute self-check: the prompts to run, how to set up a fair test, what to write down, and how to trace a wrong answer back to the page behind it.
By Santiago Alvarez, Founder, Ad Hoc Digital
Last updated
The short answer
If you read one part of this page, read this.
Ask the tools the questions your clients ask, the way they ask them, in a clean browser window, more than once. Each time, write down three things: whether your firm is named, which sites the answer cites, and anything it gets wrong. Five prompts in three tools, run twice, takes about 30 minutes.
Every one of these tools says it can be wrong. Google says AI Overviews "can and will make mistakes", and OpenAI says ChatGPT's search results and citations can be incomplete, outdated or incorrect. The useful part of the check is tracing each wrong fact back to the page it came from, because that page is the thing you can fix.
This is the same method we run for clients every month, and on our own agency. We work with law firms across the US and Canada, and also in Australia and the UK.
Before you start
Set up a clean test, because your own account and office tilt the answer.
A lawyer signed into their own accounts, on the office network, is the least typical tester there is. Four small changes get you closer to what a new client sees.
- Use a private window, signed out
Google says AI Mode works better with history and personalized recommendations on, which also means a signed-in partner sees answers shaped by their own searches. Microsoft says Copilot can store facts a user asks it to remember. A private window, signed out, removes most of that.
- Type the city, then leave it out
OpenAI says ChatGPT estimates general location from the IP address and may share it with search providers. An office VPN or a home in the next town over changes which firms are in the running. Run some prompts with your city named and some without, because real clients do both.
- Make sure ChatGPT is actually searching
ChatGPT can answer from what it learned in training or from a live search. Check that Search is on in the source picker, and look for citations under the answer. A reply with no sources is a memory answer, and it tells you little about this month.
- Do one round on a phone
OpenAI says relevant ChatGPT results may include a map on iOS and Android, and Google's Ask Maps started rolling out in the US on Android and iOS in March 2026. Clients searching on a phone, near you, see a different layout than you do at your desk.
The prompts
Ten prompt patterns cover how clients actually ask.
Write them the way people talk, not the way a marketer would. Swap in your own practice, city and firm name; the examples use a hypothetical Fort Worth firm.
| Pattern | Example | What it tests |
|---|---|---|
| Recommendation | Who's a good divorce lawyer in Fort Worth? | Whether you're named at all |
| Near me | DUI lawyer near me (on a phone) | How the tool handles location |
| Problem first | My ex won't follow our custody order. What can I do in Texas? | Whether your pages are cited for the problem itself |
| Narrow case type | Who handles contested probate in Tarrant County? | The case types you want more of |
| Cost | How much does a divorce lawyer cost in Fort Worth? | Which pages are quoted on fees, and whether any number is wrong |
| Comparison | Is [your firm] or [another firm] better for custody cases? | How you're described next to a competitor |
| Brand | What is [firm name]? | The firm's own description, and wrong facts |
| Reviews | [Firm name] reviews | Which review sources the tool reads |
| Lawyer | Is [lawyer's name] a good family lawyer? | Bios, bar listings, and lawyers who left |
| Other language | Abogado de divorcio en Fort Worth | Only if you serve clients in that language |
The problem-first prompts matter more than they look. Many people describe the situation before they ever say the word lawyer, and Google says AI Mode splits a question into subtopics and searches each one. Those are the searches your practice pages need to answer.
The 30-minute version
Five prompts, three tools, two runs each.
Thirty runs at about a minute each is a fair first look. Spread it wider later; the point today is a baseline you can repeat.
Pick five of the ten patterns
Recommendation, problem first, narrow case type, brand and reviews make a good first set. Paste them into a note so every run uses the exact same words.
Run every prompt twice
Answers change between runs, and Google itself suggests asking several versions of a question. A firm named once in two runs is in a different spot from one named in both.
Open the sources
Click the citations or the Sources button. Copilot lists its hyperlinked citations below grounded answers, and Ask Maps points to links for its AI sources. Write down every domain, not only the ones that mention you.
Copy anything wrong, word for word
A screenshot plus the exact sentence. "Has an office in Arlington" is fixable once you know which page says it; "got some details wrong" isn't.
What to record
One row per run, in a table you can print and repeat.
Five columns are enough. The example rows are hypothetical, to show the level of detail that makes the next step possible.
| Prompt, tool and run | Named? Position | Sites cited | Wrong or missing, word for word |
|---|---|---|---|
| Divorce lawyer Fort Worth, ChatGPT, run 1 | Yes, 2nd of 3 | Two legal directories, a review site, the firm's divorce page | None |
| Divorce lawyer Fort Worth, ChatGPT, run 2 | No | Two legal directories, a news article | Not named |
| What is [firm name], AI Mode, run 1 | Yes | Business Profile, firm home page, one directory | "Founded by [partner who left in 2023]" |
| Contested probate Tarrant County, Copilot, run 1 | No | County court page, two firm sites | No probate page on the firm's site to cite |
The sites-cited column is the one most firms skip, and it earns its place. It tells you which directories, review sites and pages the tools trust for your city and practice, which is a ready-made list of what to check first. Our guide on where AI tools find lawyers explains why those sources show up.
Wrong facts
Each wrong answer traces back to a page, and that page is what you fix.
Arguing with the AI in the chat changes nothing it reads next time. Find the source, correct it, and the next crawl carries the fix.
| What the AI got wrong | Where it usually comes from | How to fix it |
|---|---|---|
| Old phone, address or hours | An outdated directory listing, or a Business Profile Google updated from other sources | Correct the listing; keep the website and profile identical |
| A lawyer who left | An old bio page still live, directory profiles, a bar listing | Remove or redirect the bio; update every profile |
| A practice you don't take | An old page or blog post, or a directory category | Update or remove the page; change the category |
| Claims the firm never made | A third-party profile someone else wrote | Ask the site to correct or remove it, in writing |
| Mixed up with a similarly named business | A listing for the other business in a directory | Build your own profiles with your exact name; leave theirs alone |
| Wrong fees | An article quoting general ranges | Explain how your fees work on your own site, within your bar's rules |
Google says Business Profile information is compiled from the business's own website, licensed third-party data and user contributions, as well as what the owner enters. That's why a carefully edited profile can still show an old number: some other source disagrees. Anyone can suggest an edit to a profile; owners can correct theirs directly. Our Business Profile guide covers the setup.
Send feedback too, but don't rely on it. AI Mode and Ask Maps have thumbs up and down, AI Overviews has a feedback link, and OpenAI suggests asking ChatGPT to search again using a specific source. None of that changes the page that misled the tool.
Our monthly method
We run the same prompt list every month and score the share of runs that name the firm.
The self-check above is the first month of it. Repeated monthly, with the same words, it becomes a trend line instead of an anecdote.
- A fixed list per client
Each client's list covers its practice areas and the cities it serves, built from the patterns above, with brand checks at the end. Prompts change only at a quarterly review, so month three is comparable with month one.
- Run in a private window, signed out
ChatGPT with search on and Google's AI Mode for every prompt, with Perplexity and Copilot added for the core set. Each run is logged with the same five columns as the table above.
- Scored over three months
The score is the share of runs that name the firm. We judge the trend over three months, never one screenshot, and report new cited sources to target and any wrong claim to correct.
- Checked against the platform reports
Google's Generative AI performance report in Search Console shows impressions of the firm's links in AI Overviews and AI Mode, and Bing's AI Performance report shows citations in Copilot and Bing's AI summaries. Our guide to measuring AI search traffic covers both.
A worked example
Hypothetical: what one month's check turns up for an immigration firm.
This is an illustration, not a result. The firm and every number are invented to show how a log turns into a to-do list.
More on how immigration clients look for a lawyer in immigration marketing.
What we see
In our client work, the first check almost always finds something to fix outside the website.
Firms usually expect the problem to be their pages. More often it's a listing or profile they forgot existed.
We see the same short list again and again: a directory profile with an old suite number, a bio for a lawyer who left, a review site category that's slightly off. Each is a small fix, and each is a page the AI tools were reading.
The baseline also keeps everyone honest. Our AI search work starts with this check in week one, so the firm can compare later months against the same prompts. One family law firm now shows up first in ChatGPT, adding 50 to 100 calls a month over the last three months; that's the kind of change the monthly log is there to show, or to show isn't happening yet.
For a newer firm, the check is often mostly blank. A business and labor law firm that started that way spent six months on a credible presence everywhere a prospect might look, and prospects and referral partners now approach it first. More on how family clients choose a lawyer in family law marketing.
Common mistakes
Where firms go wrong.
We see these when firms show us their own screenshots of what an AI tool said.
Testing from your own account
Your history, saved details and office network all lean the answer toward what you already know. A signed-out private window is closer to a stranger's view.
Asking once and concluding
One run that names you, or doesn't, says little. Run each prompt at least twice and judge the share across runs.
Only checking ChatGPT
Google's AI Mode, Copilot and Perplexity draw on different material. A firm named in one can be missing from another for reasons that have nothing to do with its website.
Skipping the citations
Whether you were named is the headline. Which sites were cited is the to-do list.
Trying to correct the AI in the chat
Telling the tool it's wrong doesn't change the page it read. Fix the source, send feedback, and check again next month.
Real results
What this looked like for real firms.
One firm whose monthly checks now show it named first in ChatGPT, and one new firm that went from a blank check to people approaching it first.
Identifying details are anonymized to protect our clients. Individual result, not a promise or prediction of any specific outcome for your firm.
FAQ
Questions lawyers ask us.
Straight answers to the questions that come up most.
How often should a law firm check what AI says about it?
Monthly, with the same prompts, is enough for most firms. Answers change between runs, so weekly checks mostly measure noise, and a quarterly check misses problems for too long. Run the brand prompt again any time something changes: a new office, a lawyer joining or leaving, a new practice area.
Why does ChatGPT give a different answer every time I ask?
The answer is written fresh from whatever its search returns at that moment, shaped by the wording and your general location. OpenAI says it may rewrite one question into several searches. That's why we count the share of runs that name a firm rather than trusting any single answer.
Can I tell ChatGPT or Google to correct what they say about my firm?
Not directly. You can send feedback through the thumbs or feedback links, and OpenAI suggests asking ChatGPT to search again using a specific source. The lasting fix is correcting the page the tool read: a directory listing, an old bio, your Business Profile or your own website.
What if an AI tool says something false and damaging about my firm?
Copy the exact wording and the sources it cites, then trace the claim to the page it came from and ask that site to correct it in writing. Report it through the tool's feedback option too. If the claim could amount to defamation, that's a legal question for you to assess as a lawyer, separate from the marketing fix.
Does Search Console show what Google's AI says about us?
It shows part of it. Google's Generative AI performance report counts impressions of your links in AI Overviews and AI Mode, by page, country, date and device. It doesn't show the wording of the answers or which competitors were named, so the manual prompt check is still the only way to read what clients see.
Should we check Perplexity and Copilot, or just ChatGPT and Google?
Include at least one of them. Copilot grounds its answers in web search results and cites them, and Perplexity searches on every question. Both often cite different directories from ChatGPT, which can surface a wrong listing you'd otherwise never see. Our Perplexity guide explains how it chooses sources.
Can you run this check for our firm?
Yes. We run it in week one of our AI search work, then monthly against the same prompts, and report what moved and what to fix. If you'd like a first read for your practice and city, schedule a consultation.
Where we do this
The services this guide touches.
What this looks like when we run it for a firm, with a demo for your practice on each page.
Sources
Where these facts come from.
Official pages we read when writing this page. Platforms and rules change, so check the current version before you act on any of it. This is marketing guidance, not legal advice.
- OpenAI Help Center, Searching the web with ChatGPT
- Google Search Help, AI Overviews in Google Search
- Google Search Help, AI Mode in Google Search
- Google Maps Help, Ask questions in Google Maps
- Google Blog, Ask Maps and Immersive Navigation (Mar 2026)
- Microsoft Support, Microsoft Copilot Transparency Note
- Google Business Profile Help, How Google sources and uses info in Business Profiles
- Search Console Help, Generative AI performance report (Search)
- Bing Webmaster Blog, Introducing AI Performance in Bing Webmaster Tools (Feb 2026)
Written by
Founder of Ad Hoc Digital. Leads strategy and works directly with every client firm on AI search, Local Services Ads, Google Ads and Meta ads.
More about SantiagoWant a second pair of eyes on this?
Book a free 30-minute call. Tell us how cases come in today, and we'll tell you straight what we'd change, and whether we can help.