A user's explanatory narrative exposes their model

What it is

Asking a user to say "what did you expect to happen" or "why did you tap that" surfaces an external trace of their mental model. But there's a trap: what a user says in hindsight and what actually drove the action in the moment are often two different things. In-the-moment judgment is usually fast, vague, and hard to verbalize; asked afterward "what were you thinking", users tend to construct a plausible-sounding account on the spot — this is post-hoc rationalization. It reflects what sounds reasonable to the user right now, not the rule that actually governed the action.

Why it happens

Mental models often exist as tacit knowledge — capable of generating predictions without being stored as statable propositions. Asking a user to state "what is your model" forces a translation from tacit knowledge into language, and that translation unavoidably pulls in retrospective plausibility judgments, social desirability (not wanting to look foolish), and generic guesses about how a system "should" work. The result can sound perfectly coherent while having nothing to do with the rule that actually drove the action at the time.

Studying it

• Use concurrent think-aloud rather than retrospective questioning: have users narrate what's going through their mind while acting, not after the fact. A concurrent narrative sits closer to the real-time judgment; a retrospective question hands the user both the time and the motive to rationalize.
• Ask for a prediction in a concrete situation rather than an abstract definition: instead of "how do you think this feature works", present a new scenario and ask "if I tap this now, what do you think will happen, and why". Knowledge that can predict is a model; an answer that only echoes interface copy, with no prediction, is likely improvised on the spot.
• Be suspicious of explanations that are too fluent and too complete — a genuine tacit model, spoken aloud, tends to come out halting and hedged. An answer that reads like a manual is more likely a rationalization assembled after the fact than the actual basis for the action.
• Cross-check: compare the narrative against what the user actually does in similar situations. Where narrative and behavior disagree, trust the behavior.

Where it stops holding

• Concurrent think-aloud consumes cognitive resources and can change the pace of the task, so it's a poor fit for tasks where reaction time itself is being measured.
• For highly automatized, unreflective actions (an expert's practiced path), a user may genuinely be unable to state any reason — that doesn't mean no model exists, only that it has compressed to the point of not needing conscious access. Prediction-based prompts still outperform requests for a direct statement here.
• This method reveals what model a user holds right now; it doesn't by itself explain how that model formed — that needs separate historical interviews or usage-log evidence.

Related

• Same group: A7.06.1 Recurring errors of the same kind point to a faulty mental model
• Nearby: A7.01 Definition and function of mental models · A7.14 Identifying and correcting a faulty mental model
• Search terms: concurrent think-aloud · post-hoc rationalization · verbal protocol · mental model elicitation