Key findings
More than half of America’s search-winning personal-injury firms do not exist in AI answers. Of 481 top-10 Google organic positions measured, only 47.6% belong to a firm that any of four AI assistants named when asked the same consumer question — “Who are the best personal injury lawyers in {city}?” — for the firm’s own city.
A first-page ranking is not carried over. It is re-earned, source by source, in a layer most firms have never audited.
A top-3 Google ranking buys a coin flip, not a mention
Visibility falls with organic position, but never approaches certainty:
| Organic position | Firm slots | Mentioned by ≥1 assistant |
|---|---|---|
| 1–3 | 150 | 73% |
| 4–6 | 150 | 45% |
| 7–10 | 181 | 29% |
Even among firms Google ranks in its top three, one in four was never named by any assistant.
The assistants barely agree on who exists
| Assistant | Firm slots mentioned |
|---|---|
| Perplexity (sonar) | 41% |
| Claude (Sonnet) | 20% |
| Gemini (Flash) | 14% |
| ChatGPT (GPT‑5.5) | 11% |
The most widely used assistant named the fewest actual firms, answering instead with general guidance and directory references. Every city produced at least one visible firm on at least one assistant, so the gap is not a measurement artifact of any single market.
AI answers are built on directories, not on law firms
Across all 200 answers, the assistants attached 1,404 source citations. Only 18.8% pointed to any population firm’s own website. The rest went overwhelmingly to third parties:
| Source | Citations |
|---|---|
| SuperLawyers | 104 |
| Expertise.com | 95 |
| Justia | 85 |
| BestLawFirms | 48 |
| Avvo | 44 |
| Lawyers.com | 41 |
| 27 |
When an AI system describes a firm, four out of five pieces of evidence it works from were written by someone other than the firm.
Structured data: near-universal adoption, minority correctness
Of the 358 firm homepages that could be fetched, 96% emit JSON-LD structured data — adoption is not the problem. Quality is:
| Signal | Share of firms |
|---|---|
| Any JSON-LD present | 96% |
| LegalService or Attorney typing | 68% |
| Any attorney marked up as a Person | 39% |
| Deprecated SearchAction markup (defunct since 2024) | 63% |
| Deprecated FAQPage markup | 21% |
| Self-serving AggregateRating (ineligible for review display) | 37% |
| JSON-LD that fails to parse | 10% |
Fewer than four in ten of these firms tell machines, in machine-readable form, who their attorneys are — while a majority broadcast markup that has been dead for years. The types a firm actually needs, and the mistakes validators miss, are documented in Hyrizen’s law firm schema markup reference.
Methodology
Population. The law firms holding top-10 Google organic positions for “personal injury lawyer {city}” in the 50 most-populous U.S. cities (US/English/desktop, collected 2026-08-31). Directories, media, bar associations, and aggregators were excluded from the firm population by a published filter but retained for citation analysis. Result: 481 position slots held by 394 unique firms; every city contributed at least 8 firms.
AI measurement. One consumer-phrased prompt per city — “Who are the best personal injury lawyers in {city}? Name specific firms.” — run once against ChatGPT (gpt‑5.5), Claude (Sonnet), Gemini (Flash), and Perplexity (sonar), with web retrieval enabled where the platform supports it, in a single collection window on 2026-08-31.
Matching. A firm counts as visible if its domain appears among an answer’s cited sources or its name appears in the answer text. Names were derived from search-result titles with generic phrases excluded; a 10% sample was manually verified. Metrics were defined before data collection.
Structured data. Each unique firm homepage was fetched once (initial HTML, no script execution) and its JSON-LD blocks parsed for schema.org types; 358 of 394 homepages responded.
Limitations
This study measures one practice area, one prompt phrasing, and one collection window; AI answers vary between runs, so rates are point-in-time observations, not stable properties. Name matching is string-based with manual verification of a sample; some mentions phrased without the firm’s name or domain will be missed, which biases visibility slightly downward. The structured-data crawl reads initial HTML only, so markup injected by client-side scripts is not counted. No causal claim is made connecting AI mention rates to client acquisition, and no individual firm is identified negatively; all findings are aggregates.
What firms should take from this
The AI layer is currently assembled from directories and third-party content because those sources are consistent, structured, and retrievable — qualities most firm websites measured here do not yet exhibit. The controllable work is unglamorous: a site whose facts are stated once and identically everywhere, attorneys machines can verify as people, practice areas that resolve to real services, and the removal of markup that undermines trust. How a firm’s current site performs on exactly these checks is what an AI visibility audit measures, and the broader architecture is described in Hyrizen’s law firm practice.
Dataset and findings are published under CC BY 4.0. Cite freely with attribution to Hyrizen. Journalists and researchers can request the aggregate tables at hello@hyrizen.com.