Research Notebook
Methodological notes from building FirmRanker's research system. These notes explain measurement problems, research decisions and unresolved questions. They are not necessarily formal study findings.
- Why we separate “recommended” from “suggested” — A law firm can appear in an AI answer without being recommended. FirmRanker records how each firm is treated, not just whether its name appears.
- Why one AI response cannot establish visibility — AI systems generate answers probabilistically. The same question can produce different firms, in a different order, minutes apart. A single response is an observation — not a ranking.
- Why entity resolution matters when measuring law firms — AI systems rarely name a law firm the same way twice. Measurement depends on deciding which mentions refer to the same firm — and which do not.
- Why AI citations need careful interpretation — Seeing a source near an AI answer does not tell you that the source caused the answer. FirmRanker keeps these relationships separate.
- Prompt intent is not response treatment — What the user asked for and what the AI did are two different facts. FirmRanker stores them separately.
- What should count as a law-firm appearance? — Appearance sounds simple. In practice, a firm can show up in an AI answer in many ways, and not all of them mean the same thing.
- Why model versions matter in AI visibility research — “ChatGPT” is not one fixed system. Models, retrieval modes and interfaces change continuously. An observation means little without knowing what produced it.
- Why FirmRanker preserves the original AI answer — The original AI response is the evidence. FirmRanker stores it unchanged and never silently rewrites it after collection.