Arrow An answer engine names the firm it can describe. If your practice area, jurisdiction and client type are not stated plainly and repeated consistently across the web, the model has nothing safe to say about you — so it says something about a firm it can describe instead. That is a legibility problem, not a reputation problem.
Every managing partner has had the same unpleasant afternoon. You type your own practice area into ChatGPT — “best employment lawyer in Austin”, “who handles construction defect claims in Miami” — and three firms come back. None of them is yours. Two of them you have beaten in court.
The instinct is to assume the machine is wrong, or that someone paid for the placement. Neither is true. There is no ad slot inside a generated answer, and the model is not ranking firms by quality. It is doing something much more mundane, and much more fixable.
So what is actually happening?
A model asked to recommend a lawyer has to produce a sentence it can stand behind. To name your firm, it needs to be able to complete something like: “X handles employment disputes for mid-size employers in Travis County.” If it cannot assemble that sentence from what it retrieved, naming you is a risk. Naming a firm it can describe is not.
So the question is never “who is the better lawyer”. It is “whose practice can I state without hedging”. Thirty years of results, a wall of verdicts and a reputation that fills the courthouse do not travel to a language model. Legible, corroborated text does.
This is why the firm down the street wins. It is usually not because they are better at law. It is because their site says, in plain words on a page a crawler can read, exactly who they help and where.
What do engines actually read on a law firm site?
Most firm websites are built to reassure a human who already decided to call. They open with a hero image of the skyline, a line about “decades of combined experience”, and a contact form. That is a perfectly good brochure and a terrible source.
That last one is where most firms lose. A model cross-checks what your site claims against the rest of the web. A firm whose own pages say one thing while its directory listings say another gets treated as unreliable — and quietly skipped.
Is any of this ethically acceptable?
It is the first question a good partner asks, and it deserves a straight answer. Publishing accurate, verifiable information about who you represent and where you are admitted is not a comparative claim about quality. It sits well inside ABA Model Rule 7.1, which prohibits false or misleading communication — not clear communication.
The real exposure runs the other way. If an engine is telling prospective clients that your firm handles a practice area you dropped years ago, or places you in a jurisdiction where you are not admitted, that is an inaccuracy about your firm circulating at scale. You did not publish it, but you are the only one who can correct it — by fixing the sources it was drawn from.
The uncomfortable part
You cannot see the answers you are losing.
A missed AI recommendation leaves no trace. No impression, no bounce, no line in the analytics. The only way to know is to ask the engines the questions your clients ask and read what comes back. Most firms have never done it once.
What should a firm fix first?
The order matters more than the effort. Firms routinely spend a year producing content while remaining impossible to identify, which is the equivalent of shouting without saying your name.
- Make the entity unmistakable. One phrasing of the firm name, practice areas, jurisdictions and admissions — on the site, the directories, the attorney bios, the bar listings. Contradictions cost you more than gaps.
- Rewrite the page closest to a retainer. Pick the single practice area with the highest client value and make that page answer its question in the opening two sentences, before any firm history.
- Ask the engines. Five questions a real client would type, run across ChatGPT, Perplexity, Gemini and Google's AI answers. Write down who gets named. That is your baseline.
- Fix what the answers reveal. Absent is a different problem from described-incorrectly, and the second is more urgent than the first.
- Re-measure monthly. Same questions, same engines, same week. Visibility here moves in months, not days — and without a baseline you cannot tell progress from noise.
None of this requires new software, and a firm with a capable marketing lead can run it in-house. What it requires is someone treating it as a standing task rather than a project — see the questions buyers ask AI for the version of this that applies to any business, or how Arrow works with firms if you would rather not run it yourself.