HCI research methods
How I do the research — the methods I use
How do you find out what a person actually goes through with an AI?
You cannot learn what a person goes through by asking alone. The method is to put what they said (interviews), what they did (behavioural logs) and what they are doing (field observation) side by side, and look for where the three disagree.
Contextual inquiry and field observation
People do not report their own behaviour accurately — not out of bad faith, but because most of it is never conscious. So you go to where it is used and watch while it is used. The Toss FacePay field research worked this way.
Semi-structured interviews
A survey confirms only what you already knew to ask. A semi-structured interview fixes what you will ask but leaves room to follow an answer you did not expect. Most of the findings worth having come out of that room.
Behavioural logs and fake-door tests
Logs show what a person never said: where they stopped, what they opened and closed, how long they waited before leaving. A fake-door test measures demand for a feature before building it — put up the button and see who presses it. Better than building it and finding out nobody wanted it.
No single method is sufficient. I call it a finding only when three kinds of evidence point the same way.