Research

Research into AI and legal discovery

AI systems are beginning to participate in the process by which people discover, compare and choose law firms. FirmRanker studies that process systematically.

Our research focuses on what can actually be observed: which firms enter AI-generated answers, how those firms are treated, whether results repeat, how systems differ and what sources appear around the response.

We separate those observations from assumptions about why the systems behave as they do.

What FirmRanker studies

  • AI visibility — Which firms enter relevant AI responses?
  • Recommendation treatment — Are firms merely mentioned, suggested or explicitly recommended?
  • Repeatability — Does the same firm appear when the same question is repeated?
  • Cross-model consensus — Do ChatGPT, Gemini, Claude and Perplexity surface the same firms?
  • Prompt sensitivity — Does changing the user's wording change the consideration set?
  • Practice-area differences — Does visibility behave differently across legal needs?
  • Geographic differences — Does the consideration set change by market?
  • Source landscape — What pages, domains and source types are retrieved or cited?
  • Longitudinal change — How does visibility change as models, markets and websites change?

Active research

Studies currently running or being validated. No studies have yet been published.

  • System Validation 001 · Status: Validation — Sydney · Personal Injury Lawyers. 10 unbranded prompt archetypes × 4 AI platforms = 40 observations. Validates FirmRanker's end-to-end research pipeline before larger benchmark research. Not a Sydney benchmark or ranking.

From question to publication

Research question → Protocol → Data collection → Raw observation preservation → Extraction → Entity resolution → Human QA → Analysis → Limitations → Publication

FirmRanker does not treat a raw AI export as a finished research finding.

Current research questions

Research questions — not established findings.

  • Does AI recommend the same law firm twice? — When the same recommendation question is repeated, how stable is the consideration set?
  • Do AI systems agree? — Do ChatGPT, Gemini, Claude and Perplexity recommend the same firms?
  • How much does the prompt matter? — Does client context change which law firms appear?
  • What counts as a recommendation? — When does a list become an actual endorsement?
  • What sources surround the answer? — Which publishers, directories, firm websites and other sources appear around recommendations?
  • How does visibility change over time? — Do firms remain visible as models and indexes change?

Published research

No FirmRanker studies have completed the publication gates yet. Published studies will appear here with their methodology version, collection period and limitations.