Methodology
How we measure whether AI engines cite a law firm.
One method behind every audit, monthly report and research page: a fixed query set, four engines, fresh sessions, repeated runs, and a published definition of a citation.
What is a citation rate?
The share of AI answers, across a fixed set of client questions and four engines, in which the firm is named or its page is linked as a source. A firm asked about in 50 questions on four engines has 200 answers; named in 30 of them, its citation rate is 15%. We report it overall, per engine and per question group.
| Counts as a citation | Does not count |
|---|---|
| The firm's name appears in the answer text as a recommendation or example | A directory page that lists the firm among fifty others |
| A page on the firm's domain appears in the answer's sources | The firm's name appears only inside a competitor's quoted review |
| The firm's Avvo, Justia or Google profile is linked as a source for a recommendation | An answer that names the firm to say it does not handle the matter |
How is the query set built?
Forty to sixty questions per firm, written the way a client types them, across five groups: do I need a lawyer, process and timeline, cost, courthouse and local, and who should I hire. Each carries the firm's city or courthouse where a client would, and a language variant where the firm's clients search in another language. The set is fixed at the audit and reused monthly so results are comparable.
- 1Sources. The audit intake, the firm's own intake calls, Reddit and forum threads in the practice area, AI Mode and ChatGPT follow-up suggestions, and the questions tracked on our practice-area guides.
- 2Balance. No group is more than 30% of the set; 'who should I hire' questions are capped at ten so branded and directory-dominated answers do not swamp the rate.
- 3Change control. Questions are added, never silently replaced. A changed set is versioned and both versions are reported for one month.
Which engines, and how are they queried?
ChatGPT with search, Perplexity, Gemini, and Google's AI Overview (with AI Mode noted where it appears). Every query runs in a fresh session so earlier answers cannot shape later ones, from a Los Angeles location, with no account history. Each query runs at least twice, a week apart, and the reported figure is the average.
| Engine | What is recorded |
|---|---|
| ChatGPT (search on) | Firms named, sources linked, whether any citation was given at all |
| Perplexity | Numbered sources and firms named in the text |
| Gemini | Sources panel and firms named |
| Google AI Overview / AI Mode | Whether an overview appeared, its links, and the firms named; the map pack and organic top ten are logged separately |
Answers vary run to run. Repetition and averaging are what make a month-to-month comparison meaningful; a single pass overstates or understates any firm.
How are the cited firms' signals audited?
For every firm cited more than once in the set, and for the audited firm, we record the same checklist: entity consistency across the twenty listings, schema types present, answer position on the cited page, review count and recency on Google and Avvo, languages published, crawler access, and rendering. That is the comparison the fix list is built from.
- Entity consistency is scored as the share of the twenty listings that match the State Bar record character for character.
- Answer position is the percentage of the page's HTML at which the first direct answer appears.
- Review recency counts reviews in the last ninety days, not the lifetime total.
- Crawler access is tested by fetching the page with each retrieval bot's user agent, not by reading robots.txt alone.
What is reported, and how often?
Monthly, in one report: citation rate overall, by engine and by question group, with the month-over-month change; the firms cited instead and their movement; entity score; review velocity and response time; content shipped; map-pack position and organic sessions to money pages. Community-tier firms get a weekly alert when a tracked query changes.
What we do not claim
We do not attribute every call to an AI answer, because a client who read the firm's page inside ChatGPT and called from Google shows up as a Google lead. Citation rate is reported as a leading indicator alongside attributed leads, not as a substitute for them. No results are guaranteed; the method and the reporting are.
How does the LA AI Citation Index use this?
The same query construction and engine procedure, run across practice areas and sub-markets rather than for one firm, with every firm named counted. Results are published quarterly with the query set, the run dates and the raw counts, so anyone can reproduce them. Nothing is published from a single run.
The research page carries the index as it is published. The free AI visibility check is this method applied to one firm.
FAQ
Questions about the method.
Rankings describe a results page; AI answers do not have one. A firm can rank third and be absent from every AI answer, or rank on page two and be cited for procedural questions. Citation rate measures the outcome that now decides who gets called.
For a firm starting from zero, 15% on its tracked set by month three is normal progress. Thirty to fifty percent by month twelve is what a consistent program produces in mid-tier practice areas.
Yes. Every client report links the spreadsheet with each query, engine, run date, firms named and sources. The audit report includes the same for the audit runs.
We use scripted runs where an engine's terms allow it and manual runs where they do not. Either way the session is fresh, the location is Los Angeles, and the log records the same fields.
Still have questions? Talk to us →
Free citation audit
Find out who AI recommends instead of you.
Written report plus a thirty-minute walkthrough. No pitch until you ask for one.