AI Search · owned-data study
I measured what AI cites for 'best lawyer' queries. Firms' own websites won.
Across 236 answer-engine citations on 26 'best lawyer' searches in 8 metros, two out of three pointed at firms' own websites, not at directories and not at press.
67% of the 236 AI citations we sampled pointed to firms' own websites, not to a directory and not to earned press. The machine reads a firm's own site first. So the firm with the most citable, authoritative site wins the answer.
A person choosing a lawyer now asks an answer engine first. Ask one for the best lawyer in a city and it returns a short list of sources. We wanted to know which sources, so we counted them.
How we measured: one answer engine that runs live web searches, on Jun 16, 2026. 26 "best lawyer" queries across 8 metros plus two mass-tort searches, 236 citations, hand-classified. One engine, one dated snapshot.
Q. Why do firms' own sites win the citation?
The engine reads the page the answer stands on. A firm's own site, with original data and clear authorship, reads as the primary source. When that page carries a real number only the firm can publish, the engine cites it and names the firm.2 Directories take about 21% of the citations and earned press the smallest slice, roughly 11%.
The pattern holds because the machine wants the source the claim traces back to, not an opinion about who is good.
Q. What does that mean for a firm or its agency?
Build the site the model wants to cite. You do it with data, not opinion. Cross fatal-crash volume with carrier-safety and civil-filing records, publish where the case volume actually is by metro, and your own page becomes the citable source a competitor cannot replicate.3
Q. What moves an answer engine to cite you at all?
Not bought links. Unlinked factual mentions correlate with AI-overview visibility at 0.664, stronger than backlinks, referring domains, or domain rating, and 70 to 72% of AI brand mentions carry no link at all.4 Publishing something worth quoting, dated and sourced, is the mechanism. This study is that mechanism, run on ourselves in public.
Q. How did we count them?
We assembled the "best lawyer" queries a person actually asks, 8 metros across three injury practice areas plus two mass-tort searches. We sampled the citations the answer engine returned and classified every source: a firm's own site, a directory, earned press, or other. The full query set, the sample, and the classification rule are in the published method, so the 67% reproduces.5 We report one dated snapshot from one answer engine, not a live score.
The honest read: AI already cites firms' own sites most in legal. The firm that publishes a real number, dated and sourced, is the one it names.
Reporters and analysts: take the figure, keep the dateline and the source line. That's the whole point.
See how this becomes a Law Firm SEO engine that ranks and gets cited, not a link vendor. Law Firm SEO → · or check your own AI visibility free →
Sources & method
- Sample: 236 answer-engine citations across 26 “best lawyer” queries in 8 metros, measured Jun 16, 2026. One answer engine that runs live web searches, one dated snapshot, not a live figure. Taqtics.
- Source split of the 236 citations: firms' own sites 67.4% (159), directories 20.8% (49), earned press 11.0% (26). Taqtics legal AI-citation study, n=236.
- Earned pickup: Search Engine Roundtable cited this finding. Recorded Jul 6, 2026. Distribution proof, not a client testimonial.
- The moat asset: PI case-volume-by-metro from fatal-crash (NHTSA/FARS), carrier-safety (FMCSA), and civil-filing data. Public-records sector analysis, data moat §2.
- Unlinked mentions correlate 0.664 with AI-overview visibility; 70–72% of AI mentions carry no link. Aggregated public research.
- Published method: query set, sample, and source-classification rule, so the 67% reproduces.