A law firm finds out by auditing it. Write the firm's facts down once, ask ChatGPT, Gemini, Perplexity and Google AI Overviews six fixed questions about the firm by name, and score each stated fact as correct, wrong or missing. Then trace every error to the page the engine read.
Firm AEO searched for a published method on October 31, 2026 and found vendor pages that suggest asking an engine about the firm, but no scoring method or source trace. This post sets out one, and it is a different job from tracking how often a firm is named.
- A law firm finds out what AI says about it by writing one fact sheet, asking each engine the same six prompts about the firm by name, and scoring every stated fact as correct, wrong or missing.
- Citorian's June 2026 five-engine test found engines named at least one firm 62% of the time, so a firm that is named is also being described, and the description can be wrong.
- Citorian's source data put Super Lawyers at 33% and Justia at 20% of citations, which makes a directory profile the most likely origin of a wrong office address, attorney list or practice area.
- A fix goes to the source the engine read, not to the engine, and the audit uses only public facts, so no client information enters a prompt.
Why does it matter if AI gets a fact about a firm wrong?
Because a named firm is a described firm. Citorian's June 2026 test of five engines across five metros found the engines named at least one firm in 62% of answers, and a name arrives with an office, a practice area and a phone number. A wrong address or a practice area the firm does not handle sends a prospect to the wrong place or to a conversation the firm cannot take.
The iLawyerMarketing survey of 1,110 US adults, published August 17, 2026, found 41.9% would use ChatGPT to research which lawyer to hire. Clio's 2025 Legal Trends Report found that more than half of consumers had used or would consider using AI to answer a legal question. Accuracy is therefore a front-door problem, not an edge case.
What goes on the firm-facts sheet?
One page, written before any prompt is run, listing only facts the firm already publishes. The sheet is the answer key. Without it, a reviewer judges each AI answer by memory and misses the small errors, such as a suite number, a retired partner or a service the firm dropped two years ago.
- Legal name, trading name and the exact spelling of each attorney's name.
- Every office address and phone number, with hours and the date each was last confirmed.
- Practice areas the firm accepts, and the ones it refers out.
- Jurisdictions and bar admissions, each as the state bar or law society lists them.
- Languages offered, fee structure as published (for example, contingency or flat fee), and consultation terms.
- Awards or directory listings the firm holds, with the year.
Which six prompts should a firm run, and on which engines?
Six prompts that ask about the firm by name, run on ChatGPT, Gemini, Perplexity and Google AI Overviews or AI Mode, three runs each. Naming the firm removes the ranking question and isolates the facts. Three runs per prompt matter because engines give different answers on repeat, which is why one screenshot proves nothing.
The set below is a typical starting point. Freeze the wording once chosen, so a change in the score reflects a change in the answer and not in the question.
| Prompt | Fact it tests |
|---|---|
| Who is [firm name] and where are its offices? | Name, office addresses, city |
| What practice areas does [firm name] handle? | Practice areas accepted |
| Who are the attorneys at [firm name]? | Attorney roster, retired or moved names |
| What is the phone number for [firm name] in [city]? | Phone and hours per office |
| Is [firm name] licensed in [state]? | Bar admissions and jurisdictions |
| What do clients say about [firm name]? | Review summary, star rating, awards claimed |
How do you score the answers and trace each error?
Score every stated fact against the sheet as correct, wrong or missing, and record the cited source beside each wrong one. A wrong fact with a cited URL can be fixed at that URL. A wrong fact with no citation is logged and retested, because the engine may have drawn on training data, and no page edit reaches that.
Origin matters more than the error count. Citorian's source analysis put Super Lawyers at 33% and Justia at 20% of citations in its personal injury test, and Martindale-Avvo reported on April 6, 2026 that ChatGPT matches Google's top 10 less than 25% of the time for legal queries. Engines lean on directories and third-party pages, so the stale listing is the usual first suspect. Firm AEO treats that as a working hypothesis from those studies, not a measured rate of error causes.
How do you fix a wrong fact without sharing client data?
Fix the source the engine read, then retest after the engines have re-read it. Update the directory profile, the Google Business Profile and the firm's own page, and say the corrected fact in plain text near the top of that page. Because the audit draws only on the firm sheet, no client name, matter or file ever enters a prompt, which keeps the exercise clear of ABA Model Rule 1.6 on confidentiality.
A correction does not license new claims. ABA Model Rule 7.1 bars false or misleading communications about a lawyer's services, so the fix restores the accurate fact and adds no superlative. Firm AEO is a marketing company, not a law firm, and the firm's ethics counsel decides what the corrected pages say. Re-run the six prompts after a few weeks, since Google documents that recrawling a changed page can take a few days to a few weeks.
How often should a firm run a firm-facts audit?
Quarterly is a typical rhythm, plus after any change of address, phone number or attorney roster. The facts change slowly, so a monthly run adds cost without new information.
Can a firm ask ChatGPT or Google to remove a wrong fact?
Engines offer feedback controls, but the dependable route is correcting the page or listing the engine read. A fix at the source is what a retest can confirm.
What if an engine names the firm but cites no source?
Log the wrong fact, correct every public listing, and retest. An uncited error may come from training data, which a firm cannot edit directly.
- 1.Citorian, which personal injury lawyers AI recommends (June 2026)
- 2.Citorian, sources AI cites for lawyers
- 3.iLawyerMarketing, What online sources do people use to research and find attorneys in 2026 (August 17, 2026)
- 4.Clio, 2025 Legal Trends Report press release
- 5.Martindale-Avvo, AI visibility for law firms (April 6, 2026)
- 6.Google Search Central, ask Google to recrawl your URLs
- 7.Attorney at Work, 2 ChatGPT prompts to keep your law firm visible
- 8.ABA Model Rule 7.1, Communications Concerning a Lawyer's Services