How does Claude decide which law firm to name?

Does Claude recommend lawyers at all?

Yes, at roughly the same rate as the other major engines when the question is a direct request for a lawyer. Citorian's June 13 to 15, 2026 study logged 359 personal injury answers across ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews in five US metros. Claude named at least one specific firm in 60% of its answers, level with Perplexity and within a few points of ChatGPT at 61%.

The same study found that the ten most-named firms were named by all five engines, which is the strongest public evidence that Claude does not run on a private logic of its own. A firm that is the obvious answer in Gemini and ChatGPT tends to be the obvious answer in Claude. The gap between engines shows up lower down the list, among firms that are visible in one index and thin in another.

Reach is the honest caveat. In iLawyerMarketing's August 17, 2026 survey of 1,110 US adults, 6.2% said they would use Claude to research which lawyer to hire, against 41.9% for ChatGPT and 15.1% for Gemini. Claude is a real surface for legal research, but it is a smaller one, and that should set how much of a firm's effort it gets.

What has Anthropic actually documented about how Claude finds law firms?

Four things, all about plumbing and none about ranking. Anthropic announced web search for Claude on March 20, 2025, for paid US users, and expanded it to all Claude plans globally on May 27, 2025. Anthropic states that answers drawing on web content include direct citations. Anthropic documents three crawlers. And Anthropic's subprocessor list added Brave Search on March 19, 2025.

What Anthropic has not published is the part a marketer most wants: how Claude chooses among the pages it retrieves, how many it reads, whether it weights directories over firm sites, and how it treats reviews. The table below separates the claims a law firm will meet in vendor pitches by how much evidence stands behind them. Firm AEO uses the left two categories to plan work and treats the right one as a hypothesis to test.

Is Claude answering from training data or from a live search?

It depends on whether Claude runs a search for that question, and the answer itself usually shows which happened. When Claude searches, the response carries citations to the pages it used. When it does not, the answer comes from what the model learned during training, which has a cutoff date and no citations. A firm should read those two kinds of answer differently.

A training-data answer reflects how often and how consistently the firm appeared across the public web before the cutoff. It cannot be changed quickly, and a new page will not show up in it until a future model is trained. A search-backed answer reflects what Claude's retrieval finds today, and it responds to indexation, directory profiles and on-page answers within weeks rather than model generations.

For a law firm that distinction decides where the effort lands. A specific, local question, such as a named city plus a practice area, is more likely to trigger a search, and that is the question type where present-day work pays. A broad question such as which firms are best known for mass torts is more likely to draw on training data, where reputation built over years carries the answer.

Does Claude use Brave Search, and what does that mean for a law firm?

The evidence says yes, though Anthropic has not described the arrangement in detail. Anthropic's subprocessor list added Brave Search effective March 19, 2025. TechCrunch reported on March 21, 2025 that a developer spotted the entry and that Simon Willison found a Claude search returning the same citations as Brave. Willison also found a parameter named BraveSearchParams in Claude's web search tool.

For a law firm, the practical meaning is that Claude's live retrieval probably does not read Google's index or Bing's directly. Brave runs its own independent index. That makes the usual checks (is the site indexed in Google Search Console, is it verified in Bing Webmaster Tools) necessary but not sufficient for Claude. Brave has not published a webmaster console comparable to either, so the direct check is to search Brave for the firm's name and core practice terms and see what comes back.

This also explains the Martindale-Avvo overlap figure without contradicting it. Independent indexes built from the same web tend to agree on the strong pages. A firm whose practice page ranks well in Google will usually rank in Brave too, which is consistent with Claude matching Google's top ten about 75% of the time. The 25% where they differ is where a Brave check earns its place.

Which crawlers does a law firm need to let through for Claude?

Claude-SearchBot and Claude-User, at minimum. Anthropic documents three agents. Claude-SearchBot crawls to improve search results, Claude-User fetches a page when a person's conversation asks for it, and ClaudeBot collects training data. Blocking ClaudeBot is a training-data policy choice; blocking Claude-SearchBot or Claude-User removes the firm's pages from Claude's live answers.

Anthropic's documentation states that all three bots honor robots.txt, including the Crawl-delay directive. That is stricter than some other vendors, whose user-triggered fetchers generally ignore robots.txt. The consequence cuts both ways: a firm can rely on a robots.txt block being respected, and a firm that blocks by accident with a blanket wildcard rule will see Claude respect that mistake too. Firm AEO covered the training-versus-search decision for all three major vendors in a separate post on GPTBot, ClaudeBot and PerplexityBot.

• •Check robots.txt for a wildcard "User-agent: *" block with "Disallow: /" that would catch Claude-SearchBot and Claude-User.
• •Check the content delivery network or firewall bot settings, which can block AI agents even when robots.txt allows them.
• •Check raw server logs for Claude-SearchBot and Claude-User requests over the last 30 days. Presence is evidence of retrieval, not of citation.
• •Decide ClaudeBot separately, as a policy question for the firm's leadership.

What do the vendor guides say about Claude, and how much should a firm trust them?

Most of what the vendor guides say about Claude is general answer-engine advice relabelled. The research behind this post found no Anthropic guidance on law firm recommendations, and the guides that name Claude specifically repeat the same checklist used for every engine: consistent entity data, answer-shaped pages, directory profiles and third-party corroboration. That advice is sound. It is not Claude-specific.

The test for any Claude claim in a sales pitch is simple. Ask whether it cites an Anthropic statement, a published dataset with a method section, or neither. A claim that Claude rewards a particular word count, a particular schema type or a particular review volume falls in the third group unless someone shows a controlled test. Firm AEO has not found one for any of those claims as of November 2026.

The same discipline applies to the numbers that are cited. The roughly 75% overlap figure comes from Martindale-Avvo's own analysis, and Martindale-Avvo sells the directory profiles that the overlap would favour. The figure is plausible and consistent with the Brave evidence, but it comes from one interested party and its method is not public. Citorian's 60% name rate comes from one practice area, personal injury, in five large metros. Both are useful. Neither should be read as a law of nature.

What can a law firm do to be named by Claude?

The same work that earns citations in the other engines, plus two Claude-specific checks. Because the most-named firms in Citorian's study appeared in all five engines, the fundamentals carry across: accurate and identical entity data, complete legal directory profiles, practice pages that answer client questions directly, and corroboration from sources a model trusts. The Claude-specific steps are crawler access and a Brave index check.

The ordering below is typical for a firm starting from a reasonable website and thin directory coverage. The timings are illustrative, not promises. Any work that touches reviews, case results or comparative claims stays inside ABA Model Rule 7.1, which prohibits false or misleading communication about a lawyer's services, and the firm's own ethics counsel signs off on the wording.

• •Week 1. Confirm Claude-SearchBot and Claude-User can reach the site, at robots.txt and at the network edge.
• •Week 1. Search Brave for the firm name, each attorney name and the top three practice-plus-city phrases, and record what ranks.
• •Weeks 1 to 3. Reconcile name, address, phone and practice focus across the website, Google Business Profile, Bing Places and the legal directories.
• •Weeks 2 to 6. Complete the directory profiles that the published datasets show engines citing for law firm answers, starting with Super Lawyers and Justia.
• •Weeks 3 to 10. Rewrite practice pages so each opens with a direct answer to one client question, with a source for every number.
• •Month 3. Re-run the test in the next section and compare against the baseline.

How can a firm test whether Claude names it?

Run a fixed set of prompts in Claude, record whether the firm is named and what Claude cites, and repeat the identical set monthly. Twelve prompts across four question types, each run three times in a fresh conversation, gives 36 answers, which is enough to see a pattern without becoming a project. The table sets out the four types and why each one matters.

Record three things per answer: whether the firm was named, which competitors were named, and which URLs Claude cited. The cited URLs are the most useful column, because they show which directory or page Claude is reading for that question, and that is the asset to improve. Results from a single month are noise. Firm AEO reads a change only once it holds across two consecutive monthly runs.