Stated vs. Revealed Preferences: Why Customers Say One Thing and Do Another (2026)

Stated vs. Revealed Preferences: The Short Answer

A stated preference is what a customer says they prefer, want, or will do when you ask them directly — in a survey, a focus group, or an interview. A revealed preference is what a customer's actual behavior shows they prefer — what they click, buy, keep, and return to. The two are often different, and when they diverge, revealed preference is almost always the truer signal.

The concept comes from economist Paul Samuelson, who introduced "revealed preference" in 1938 to argue that we should infer what people value from the choices they actually make, not from what they claim to value. Nearly a century later it is the single most important idea a product researcher can internalize, because the gap between the two — the say-do gap — is where wasted roadmaps, failed launches, and misleading survey results are born.

"To design an easy-to-use interface, pay attention to what users do, not what they say." — Jakob Nielsen, Nielsen Norman Group

---

How Big Is the Say-Do Gap?

This is not a rounding error. The divergence between stated and revealed preference is large and well documented:

• Traditional stated-preference research predicts actual purchase behavior with only about 34% accuracy, while behavioral (revealed-preference) methods reach as high as 89% when properly implemented.
• Studies of online shoppers find that roughly 38% of people act differently from the behavior they previously stated — nearly four in ten.
• In health and wellness categories, consumers overstate their intention to make the healthy choice by an average of 47% — they say salad, they buy fries.

The reasons are usually unconscious rather than dishonest. As market researchers note, the discrepancy comes from "limited self-awareness and a desire to meet the imagined expectations of others" — not deliberate lying. People genuinely believe they will use the feature, choose the healthy option, or recommend the brand. Their future selves simply do not comply.

---

Why the Gap Exists

Several well-studied cognitive forces drive the wedge between what customers say and what they do:

• Social desirability bias. People answer in ways that make them look good to the interviewer — more frugal, more health-conscious, more environmentally responsible, more sophisticated than their behavior bears out.
• Projection bias. We assume our future preferences will match our present state. Asked on a full stomach, people under-order; asked about a calm hypothetical, they under-weight how they will behave under real pressure.
• Limited self-awareness. Much of behavior is automatic and habitual. People genuinely do not know why they chose what they chose, so when asked, they confabulate a plausible reason.
• Hypothetical bias. Saying "yes, I would pay $20 for that" costs nothing. Actually paying $20 is a real decision with real trade-offs. Intentions are cheap; behavior is expensive — and only the expensive signal is trustworthy.
• The desire to be helpful. Interview and focus-group participants want to give you a "good" answer, so they manufacture enthusiasm for concepts they would never adopt.

The most famous cautionary tale in market research is the yellow Sony Walkman study: in a focus group, participants reportedly raved about a bright yellow player, but when offered a free unit on the way out — yellow or black — they overwhelmingly took black. (Researchers debate whether the specifics are apocryphal, but it endures because every practitioner has watched some version of it happen.) The lesson is durable even if the details are folklore: enthusiasm in the room does not predict the choice at the shelf.

---

When Stated Preferences Are Still Useful

Revealed preference is more reliable, but stated preference is not worthless — it is the only option in several situations:

• New or non-existent products. You cannot observe behavior toward something that does not exist yet, so early concept work necessarily relies on stated reactions — which is exactly why you must probe past them carefully.
• Understanding the "why." Behavioral data tells you what people did; it rarely tells you why. Stated methods, done well, recover motivation, context, and emotion.
• Attitudes, values, and satisfaction. Some things — brand perception, trust, frustration — have no clean behavioral proxy and must be asked.

The goal is not to abolish stated research. It is to design stated research that behaves as much like revealed research as possible — and to triangulate the two whenever you can.

---

How to Close the Say-Do Gap in Your Research

You cannot eliminate the gap, but disciplined technique shrinks it dramatically:

1. Ask about past behavior, not future intentions

"Would you use this?" invites a cheap, flattering yes. "Tell me about the last time you faced this problem — what did you actually do?" recovers revealed behavior through the interview. This is the foundational move of the Mom Test: anchor in specific, recent, real events.

2. Ask for evidence, not opinions

Instead of "do you value X?", ask "when did you last pay for X? How much? What did you switch from?" Commitments, purchases, and switches are stated data that behaves like revealed data.

3. Look for the currency of commitment

The strongest signals in any conversation are the ones that cost the customer something: an email sign-up, a pre-order, a referral, time spent, a completed task. Weight these far above verbal enthusiasm.

4. Triangulate stated with behavioral

Pair what customers tell you with what your product analytics show they do. When the two agree, act with confidence. When they diverge, trust the behavior and investigate why the words differ — that gap is itself an insight.

5. Reduce the incentive to perform

Neutral, non-leading questions and a judgment-free tone lower social-desirability pressure. Never signal the answer you are hoping for.

---

The Modern Approach: Getting Past Stated Answers with AI

The hard part of closing the say-do gap is discipline at scale. A skilled interviewer can steer one conversation away from hypotheticals and toward real past behavior — but doing that consistently across dozens of interviews, and then honestly separating "I would" from "I did" in the analysis, is where human research programs strain.

AI-native platforms like Koji are well suited to this problem precisely because the technique is systematic:

• Every conversation probes for behavior, not intention. A Koji AI-moderated interview is designed to follow a vague stated answer with "when did that last happen?" and "what did you actually do?" — automatically, over voice or text, with every participant. It applies the past-behavior discipline consistently instead of only when the interviewer remembers to.
• Structured questions separate cheap yeses from real commitment. Combine open-ended probing with yes_no, scale, and ranking questions to distinguish a hypothetical "sure, I would try it" from a graded, comparative signal — and see the reasoning behind the number. See the structured questions guide for the six question types.
• Analysis at behavioral scale. Koji's automatic thematic analysis reads across hundreds of conversations to surface where stated enthusiasm is not backed by any evidence of action — the exact pattern that predicts a launch will underperform. Pair those insights with your product analytics to triangulate say against do.

While a legacy survey tool like SurveyMonkey collects stated preferences at face value, an AI-native platform like Koji is designed to interrogate them — to ask the follow-up that reveals whether "I love it" is a preference or a pleasantry. And because the discipline is built into the moderator, you do not need a behavioral-science PhD to run research that respects the say-do gap.

---

Common Mistakes

• Trusting purchase intent surveys at face value. "Definitely would buy" routinely overstates real conversion by a wide margin — discount it heavily.
• Running feature-request tallies as if they were demand. Wanting a feature (stated) is not the same as adopting it (revealed).
• Designing focus groups that reward performance. Group settings amplify social-desirability bias; the loudest enthusiasm is often the least predictive.
• Ignoring the analytics you already have. Your product's behavioral data is a revealed-preference goldmine that most teams under-use.
• Treating the gap as failure. When words and behavior diverge, that is not noise — it is one of the most valuable insights research can produce.

---

Related Resources

• Structured Questions Guide — the six question types that separate cheap yeses from real commitment
• The Mom Test Methodology — how to ask about behavior instead of opinions
• Observational Research Guide — reading revealed preference directly from behavior
• Qualitative vs. Quantitative Research — when to ask versus when to measure
• Customer Signals — separating meaningful signals from noise in feedback
• Problem Validation Guide — validating with evidence of action, not enthusiasm
• User Interview Questions — question patterns that recover real behavior

Customers are not lying to you — they simply do not know themselves as well as their behavior does. The researchers who win are the ones who ask about what people did, weight actions over words, and use tools that make that discipline the default rather than the exception.