The setup
Two complete personal-injury panels in two states — one Florida metro, one Texas metro — on the standard protocol: frozen customer questions (“best car accident lawyer,” “who should I call,” truck-accident and wrongful-death questions, a fee question) across six engines, three runs each on the four automated engines and two on the ChatGPT and Claude apps, every run logged. Two states means two different bar systems, which turned out to be the point. All firms anonymized.
Finding one: the engines are openly hostile to PI advertising
In no other vertical we’ve measured do the engines editorialize this hard. Live answers included: “whoever ranks #1 on Google, dominates the map pack, or gets named ‘best’ by a directory is mostly telling you about ad budget, not trial skill” — and one engine dismissed the pay-to-play directory results it had just retrieved, by name. The most-advertised firms in each market were regularly mentioned with an explicit caveat: high volume, case-manager handoffs, ask who actually works your file.
For the biggest advertisers, that’s a headwind. For everyone else, it’s the opening: the engines are actively looking for reasons to recommend someone other than the billboard.
Finding two: the credential is the filter — and the engines verify it
In both states, board certification in the trial-law specialty was the engines’ declared sorting mechanism. They didn’t just mention it; they taught it: how rare it is (a low single-digit percentage of the state’s attorneys), what it requires (documented jury verdicts, peer review, an exam), and — crucially — where to verify it, pointing users to the state board’s official lookup and the bar’s disciplinary search “rather than taking a firm’s website’s word for it.” One engine ranked firms by how many board-certified attorneys each had. Another warned that a lapsed credential gets you marketed to rather than represented.
Finding three: every market splits into a recommended tier and a footnoted one
The strangest finding in the data: when we cross-referenced which firms’ websites the engines read against which firms they recommended, the market split cleanly in two. One cohort converted nearly every citation into a recommendation. The other — including firms whose sites were among the most-retrieved in the market — appeared in map cards and “other notable firms” tails but essentially never cracked the main recommendation list. One firm’s site was read dozens of times across a single panel and never once earned a top-list slot.
What separates the tiers isn’t publishing effort. Two of the footnoted firms publish detailed verdict tables — seven- and nine-figure results — and stayed footnoted. The recommended tier pairs its results with the things the footnoted tier lacks: review mass an order of magnitude higher, presence on the editorial best-of lists the engines ground answers in, and third-party corroboration of the results and credentials they claim.
Finding four: publishing your verdicts is necessary, not sufficient
That’s the uncomfortable correction to the standard advice. Published, specific results absolutely appear in answers — engines quoted verdict amounts, recovery totals, and “longest-serving” superlatives directly from firm sites. But in this vertical, published numbers without independent corroboration read to the engines like advertising: one answer flagged a firm’s recovery claims as “self-reported and not independently verified anywhere I found.” The firms whose numbers carried weight were the ones whose numbers appear somewhere other than their own site.
The before-state, at its purest
Both panels used a deliberately chosen subject: a decades-long, board-certified trial specialist with a spotless disciplinary record — full real-world authority. Across every API-measured surface, on more than a hundred and fifty unbranded who-should-I-hire runs: zero namings. Asked about the firm by name, every engine assembled a glowing, accurate profile from the bar record and the certification. The authority is fully legible to the machines; it simply never surfaces when it matters. And on the one surface whose credential-first reasoning runs deepest, the subject did break through — named in the credential discussion in some runs, gone again in others. Eligibility without stability: the position exists to be won, and nothing holds unassisted.
What to do with this
If you’re a certified trial specialist who doesn’t advertise heavily, this market structure is your opening — the engines are looking for you and mostly can’t find you. The order of operations from these panels: make the credential loud, published, and registry-consistent; get your results corroborated somewhere that isn’t your own site; build review mass with attorney-naming review text; and get onto the editorial lists the engines actually read. Then measure it per engine — because the tier you’re in differs by surface — which is what a Nameworthy panel produces.