Is Your Law Firm Showing Up in ChatGPT and AI Search?

What we measured across 26 legal practices

The figures below come from AIOInsights research: 466 public websites we chose to evaluate, scored on six pillars from what a visitor, a crawler or an answer engine can read from the public web. Nobody asked us to evaluate them and none are our clients. Every individual evaluation is published, with its own date and engine version, at estate planning and divorce mediation in public signal research, and the corpus-wide findings are written up in AI Trust Signals: What 466 Websites Show.

Two legal specialties in that corpus have enough sites to report, and the gap between them is the whole argument of this page:

Both groups are legal practices serving individuals. Both are staffed by people who write for a living. The mediation firms are nearly two points ahead, and almost none of that gap is technical: entity consistency differs by 1.5 while reputation signals differ by nearly double and local presence by more than two points.

Estate planning scores ranged from 1.1 to 9.1. That spread inside one specialty matters more than the average: the ceiling is reachable, and most firms are nowhere near it.

Why ChatGPT does not mention your firm

The usual assumption is that the website is not optimized. Across the corpus that is rarely the binding constraint. Legal sites tend to score well on the parts a developer ships: structure, schema, a consistent name and address, readable service pages.

What an answer engine is short of is a reason to name you specifically. When someone asks for a probate attorney in a named city, the model has to choose. It resolves that with corroboration: independent sources that mention the firm, agree on who it is, and say something specific about what it does. A firm that only appears on its own website has given it nothing to work with.

This is why mediation outscores estate planning. Mediators are routinely listed in directories, referral panels and court-adjacent resources, and those listings are exactly the corroboration an engine can read.

What is different about legal, and it is not marketing

Every state bar restricts what a lawyer may say about outcomes, and rules on client testimonials vary by jurisdiction. That is a real constraint and it is not an excuse, because the signals that move this score are mostly not testimonials:

• Attorney identity. Named lawyers with bar admissions, jurisdictions, and a biography that reads as a person rather than a placeholder. This is verifiable and uncontroversial.
• Practice specificity. "Estate planning" is a category. "Revocable living trusts for blended families in Ventura County" is an answer to a question somebody typed.
• Third-party presence. Bar association listings, legal directories, local organizations, speaking, writing published somewhere other than your own blog.
• Location that resolves. Consistent office details wherever they appear. This is the second weakest pillar for estate planning at 4.70 and it is almost entirely clerical work.

None of that requires a claim about results, and none of it requires a client to say anything.

What to do, in order

1. Find out where you actually stand. The distribution above is a corpus, not a diagnosis of your firm. The spread inside estate planning ran from 1.1 to 9.1.
2. Fix location consistency first. It is the cheapest of the weak pillars and it is clerical rather than creative.
3. Make the attorneys legible as people. Named, credentialed, with jurisdictions.
4. Then work on corroboration, which is the slow one, never finishes, and is the reason the firms ahead of you are ahead.

Run my free check How it is scored

Limits of this data

• 26 firms across two specialties is a small cohort, and it is mediation and estate planning specifically. We have not evaluated a cohort of personal injury, family, criminal, immigration or corporate firms, and this page does not claim to describe them.
• These figures pool more than one engine version. The evaluations behind them were stored between late June and late August 2026, and the rubric changed during that window: versions 2.3.1, 2.4.0, 2.7.0 and 2.9.3 all appear among the published legal reports, each of which states its own version and date on the report itself. Two of those releases corrected faults that moved scores upward on re-evaluation, so the pooled averages here are indicative of the gap between the two cohorts rather than a precise reading of either. A same-version cohort large enough to publish on its own does not exist yet, and we are not going to manufacture one by pretending the versions are interchangeable.
• These are sites we chose, not a survey. The cohorts are the firms we evaluated during competitor research. No figure here should be read as "the average firm" in any field.
• A score measures published signals on a date. It is not a judgment of the practice. A firm with an excellent reputation and no public trace of it scores low here, correctly, because the question being asked is what a machine can verify.
• We have not shown that these signals cause an AI recommendation. This reports what sites carry. Our share of voice research measures what engines actually answer. Connecting the two properly needs both, over time.
• Nothing here is legal or ethics advice. Advertising rules vary by state bar and compliance is yours.

A check is a baseline, not monitoring. The free evaluation reads what your firm publishes on the day it runs and returns a score, the evidence behind it and the highest-priority issue. It does not watch what changes afterwards, and re-running it on an unchanged site returns the same reading. If what you need is the record over time, and somebody completing the authorised work, that is ongoing AI visibility tracking from Digilu.