What people ask AI assistants to find a car accident lawyer in Phoenix

Run these yourself

The pack is only worth anything once somebody asks the questions. This is the protocol we use, and it is short enough to do by hand for one practice area in an afternoon.

1. Log out first. A signed-in assistant has your history, your memory, and your saved location. You would be measuring your own account, not what a stranger sees.
2. Ask each question three times. Not twice, not once. Identical reruns do not reliably return the same firms, so a single answer tells you about one draw rather than about the assistant.
3. Record which of five states you saw for each firm you care about, per answer: absent (not named), mentioned (named in passing), cited (treated as a source), recommended (offered as someone to hire), or primary recommendation (elevated above the others). Only the last two count as recommendations when computing Case Recommendation Share.
4. Log every answer, not a score. Keep the full text with its date and assistant, so a number you compute later can be checked against what was actually said. The answer log is set up for this pack and this market if you want a structure for it.
5. Throw away failed runs, do not score them as zero. A refusal, a timeout, or an empty answer says nothing about a firm. Counting it as “not recommended” manufactures a trend out of an outage. Our full rules for that are on the methodology page.

What this page does not tell you

Worth reading before this pack gets forwarded to anyone as evidence of something.

• It is a question generator, not a measurement. Nothing here reports what any assistant answered, in Phoenix or anywhere else. The counts on this page count questions we wrote, not results.
• Asking these questions does not make an assistant recommend you. Neither does writing pages that answer them. We have not measured that a firm can move its position this way, and we do not claim it. Anyone who does claim it owes you the data.
• The same question asked twice does not reliably return the same firms. That instability is exactly why the protocol above says three runs, and why a single screenshot of a good answer is not a result.
• Being named by an assistant is not a statement about quality. It says something about coverage and about which sources the assistant leaned on. It says nothing about whether an attorney is any good, and this page is not legal advice or a recommendation of anyone.
• There is no freshness promise. None. The pack is deterministic — the same catalog and the same templates produce the same questions on every build. Determinism is the feature; there is no schedule and we are not implying one.

If it helps to see the reasoning in longer form, the methodology page covers what we ask and what we throw away, and the writing works through the instability problem and the sources behind these answers in more detail than a tool page should.

Packs in other markets

Phoenix is the only Arizona metro in the catalog, so there is no same-state pack to compare against yet.

All 25 market packs · The other free tools