The Reason Comes Last

In 2005, a team of Swedish cognitive scientists ran an experiment that should have ended a significant portion of the market research industry. It did not. Most marketers have still never heard of it.

The setup was simple. Participants were shown two photographs of faces and asked to choose the more attractive one. The researcher then slid the chosen photo across the table and asked the participant to explain the choice. Except sometimes — using sleight of hand — the researcher handed over the photo the participant had rejected.

Only about 13 percent of participants noticed the switch. The rest looked at a face they had just declined, and explained, in detail and without hesitation, why they had picked it. She has a warmer smile. I liked the earrings. She looks like someone I'd get along with.

They were not lying. They were doing what human beings do when asked to account for a preference: constructing a plausible story after the fact and experiencing it as memory.

The phenomenon has a name — choice blindness — and Petter Johansson and Lars Hall have spent two decades demonstrating how far it reaches. It is the single most inconvenient finding in commercial research, and the industry has largely responded by ignoring it.

Introspection Is Not a Readout. It Is a Reconstruction.

The theoretical groundwork was laid nearly fifty years ago. In 1977, Richard Nisbett and Timothy Wilson published "Telling More Than We Can Know: Verbal Reports on Mental Processes", which argued that people have little or no direct introspective access to the higher-order cognition driving their judgments. When asked why, they do not consult an internal record. They generate a theory — drawing on what seems like a reasonable cause for the effect in question — and report the theory as though it were observation.

The paper has been cited thousands of times. It remains one of the most influential findings in the study of consciousness. And it carries a specific, expensive implication for anyone who buys research: the "why" in your customer data is not evidence. It is authorship.

Choice blindness took Nisbett and Wilson's argument out of the seminar room and made it physical. It is one thing to claim introspection is unreliable. It is another to hand someone the opposite of what they chose and watch them defend it fluently.

It Does Not Stop at Faces

The obvious objection is that face preference is trivial. Nobody has a deeply held position on which stranger is more attractive. Fair enough — so the researchers escalated.

In a 2012 study published in PLOS ONE, Hall, Johansson and Thomas Strandberg built a "self-transforming survey" on moral and political questions — the kind of positions people describe as core to who they are. Using a concealed adhesive, the questionnaire quietly reversed participants' own answers before they were asked to justify them.

Sixty-nine percent failed to detect at least one reversal. And they did not simply fail to notice. Many proceeded to build coherent, forceful arguments for the position they had just rejected minutes earlier.

Follow-up work found something worse for research validity: the confabulated justification often changed the actual preference. Having argued for a position, participants moved toward it. The act of explaining a choice you did not make can produce the choice.

Read that again with a focus group in mind. You put eight people in a room, show them three concepts, and ask them to explain their reactions out loud to strangers. You are not measuring preference. You are manufacturing it, then recording the output as insight.

Where the Fiction Enters Your Marketing

This does not stay in the lab. It enters commercial decision-making through four doors, and most organizations have all four propped open.

Concept testing. You ask which message resonates. Respondents tell you — confidently, with reasons. The reasons are generated on demand. What you have measured is which option is easiest to justify aloud, which is a different variable than which one moves behavior.

The say-do gap. Stated intention and observed behavior diverge sharply, and the divergence is largest exactly where identity is involved. People overstate the healthy option, the sustainable option, the disciplined option — not to deceive the researcher, but because a survey is a moment of self-presentation. The answer is shaped by who the respondent wants to be, which is rarely the person who makes the purchase.

Self-reported attribution. "How did you hear about us?" is the most trusted bad question in marketing. It inherits every defect above and adds recency bias on top. It systematically over-credits the last memorable touch and under-credits the long, unmemorable accumulation that made the last touch land. Dreamdata's own test of self-reported attribution found meaningful distortion, including a heavy pull toward whatever option sits first in the dropdown. The customer is not auditing their own journey. They are naming the thing they can recall.

Post-purchase interviews. The most seductive of the four, because the customer has real experience and real conviction. But conviction is precisely the problem. A buyer who has already committed has strong motivation to produce a rational narrative for a decision that was substantially driven by timing, social context, default settings, and mood.

Confidence Is the Tell, Not the Signal

Here is the part that makes this a strategy problem rather than a methodology footnote.

Confabulated answers do not arrive hedged. They arrive articulate. In the choice blindness studies, the manipulated justifications were often longer and more detailed than the genuine ones. The participant, unconsciously working to reconcile a mismatch, generates more explanation, not less.

Which means the research output that most impresses a room — the eloquent verbatim, the customer who explains their own psychology with unusual clarity, the quote that ends up on slide four of the strategy deck — carries no more truth value than the mumbled one. Possibly less.

Organizations do not weight evidence by validity. They weight it by vividness. A single articulate customer voice will overrule a quiet statistical pattern in almost any meeting. That is how confabulation gets promoted to strategy: not because anyone was careless, but because the fiction was persuasive and the truth was boring.

This is a distinct failure from the one we've written about in your dashboard. There, the fiction is authored by platforms with an incentive to claim credit. Here, it is authored by the customer, with no incentive at all beyond the ordinary human need to make sense. Both produce confident numbers. Neither produces causes.

What Self-Report Is Actually For

None of this means stop talking to customers. It means stop asking them to do a job their cognition cannot perform.

People are unreliable narrators of why. They are excellent sources of what and how it is described.

Ask a customer why they switched vendors and you get a rationalization. Ask them to walk through the week before the switch — who was in the room, what broke, what their boss said, what they searched — and you get retrievable episodic detail. Events are stored. Causes are inferred.

Likewise, the single highest-value output of qualitative research is not explanation. It is vocabulary. The precise words customers use for the problem, the workaround, the failure state, the stakes. That language is real data. It is what they actually said, not a theory about what they actually think. Most category-defining copy is transcribed, not invented.

So the reframe is narrow and specific: treat interviews as a source of language and sequence, and treat behavior as the only source of causation.

Measure What Costs Something

Preference expressed at zero cost is close to worthless, because the respondent is free to answer as their aspirational self. Preference expressed at a cost is a signal, because costly action is hard to fake — the same logic that governs why cheap evidence stopped persuading anyone.

Practically, that means moving your evidence base toward methods where the customer pays something to register a preference — money, time, effort, reputation, or opportunity.

• Holdouts and geo experiments. Withhold spend from matched regions and measure the difference in total outcome. No attribution model, no survey, no self-report. Geo holdout testing gives you a clean causal boundary precisely because it never asks anyone anything.
• Priced choice over rated preference. Do not ask which package is more appealing. Present the packages with real prices and real commitment and observe selection. Trade-off forces disclosure in a way that a rating scale never will.
• Friction as measurement. A form field, a scheduling step, a waitlist deposit. Each one filters aspiration out of the data. What survives friction is closer to preference.
• Behavioral pre-registration. Before running the research, write down what result would change the decision. If no possible answer would change what you do, the study is theater, and its main function is to generate justifications for a plan already chosen — which is the organizational version of the same bug.

The Discipline

Three questions will clean up most research programs.

Did the respondent pay anything to give this answer? If not, treat it as vocabulary and directional interest, never as a cause.

Am I asking for an event or an explanation? Events are recalled. Explanations are generated. Design every question to retrieve the former.

Would I have believed the opposite answer just as easily? If a finding and its inverse would both have produced a confident-sounding slide, the research had no discriminating power and the story is coming from you.

The Reason Comes Last

The uncomfortable truth in the Johansson and Hall work is not that people deceive researchers. It is that the reason genuinely does come last — the choice happens, then the mind supplies an account, and the account feels indistinguishable from the cause. Your customer is not withholding the real reason. There was no real reason available to withhold at the moment you asked.

Which puts the burden where it belongs. If buyers cannot narrate their own decisions, then understanding those decisions is not a matter of asking better questions. It is a matter of building conditions in which behavior has to reveal itself, and having the discipline to trust that behavior over the more articulate story sitting next to it in the deck.

Most brands will not do this. Asking is cheap, fast, and produces quotable material. Measuring costs money and frequently returns an answer nobody wanted.

That asymmetry is the opportunity. In a market where nearly everyone is optimizing against fiction, the firm operating on revealed behavior is not marginally better informed. It is playing a different game entirely.