AI cannot generate empathy map data. It can simulate a frustrated user, but simulation is not evidence. Nielsen Norman Group draws a hard line: the Says, Thinks, Does, and Feels quadrants of an empathy map require input sourced from actual user research, not language model inference.
The article is worth reading in full because it does not stop at the obvious warning. It works through the specific mechanics of where AI fails in this process, and where it can legitimately help, organizing research you have already collected rather than inventing research you have not done. That distinction has real consequences for how teams run workshops.
The core problem is calendar pressure. A workshop is scheduled, the data is not ready, and an AI tool produces a full page of sticky notes in seconds. The speed makes it feel like a solution. NNG's argument is that what those notes are made from matters more than how fast they appeared. Read the full piece to understand exactly where that substitution breaks the map.
[READ ORIGINAL →]