How AI chooses which lawyers to name

What this page is and is not

This is our working model. The answer engines do not publish their selection logic, they change it without notice, and no outside party can see it directly. Everything here is inferred from documented behavior, platform guidance, and repeated testing. Where we are uncertain we say so.

We publish it because an attorney should be able to examine a method before paying for it. If a claim here cannot be checked, treat it as our opinion.

The model, in four steps

When a person asks an answer engine for a lawyer, the engine appears to work through something like the following.

• Resolve the question. Practice area, location, and urgency are extracted. A vague question is narrowed using the person's location and prior context.
• Retrieve candidates. The engine draws on its training data and, for most current systems, a live web search. Candidates are firms that appear in sources the engine already treats as reliable: directories, bar listings, review platforms, news, and firm sites it has crawled.
• Check for agreement. A firm whose name, practice area, and location are consistent across its own site, its structured data, and third-party sources is kept. A firm with conflicting or missing information is dropped, because naming it is a risk.
• Prefer the clearest answer. Among the firms that survive, the engine favors those whose content directly addresses the question asked, in plain language, with a specific jurisdiction. It quotes that content when it can.

What the evidence supports

Consistency matters more than volume. In repeated testing, firms with modest sites but clean, agreeing data across their own pages and major directories are named more reliably than firms with large sites and inconsistent listings.

Specificity beats authority claims. A page that answers one procedural question for one state is quoted more often than a page that describes a firm as experienced, aggressive, or award-winning. Engines do not quote adjectives.

Third-party mentions carry weight the firm's own site cannot. A firm that appears in a state bar directory, a reputable review platform, and a local news story is treated as verified. A firm that appears only on its own site is treated as unverified.

What we do not know

We do not know the relative weight of any one signal, and we suspect it differs by engine and changes over time. We do not know how much of a given answer comes from training data versus live retrieval. We cannot explain why the same question sometimes produces different firms on different days, beyond noting that the systems are not deterministic.

Anyone who claims more certainty than this is selling it.

How we test it, every month

For each client, we fix a set of real client questions for their practice area and market before the engagement begins. The set does not change month to month, because a moving target cannot show a trend.

On a fixed schedule we ask each question of ChatGPT, Gemini, Google's AI Overviews and AI Mode, and Perplexity. For each answer we record whether the firm is named, whether the practice area and location are correct, and whether the firm's own site is cited as a source. We also record which other firms were named.

The result is a one-page report a partner can read in two minutes. It shows the trend, it shows the competitors, and it shows the questions where the firm is still absent. It does not show a score we invented, because a score invites the question of what it means.