Willingness to pay research with emotion AI compares what people say they would pay with how they react when the price is revealed. Stated willingness to pay overstates real willingness to pay by about 21% on average. A research agent captures facial expressions and attention at the price reveal, probes answers that conflict with those signals, and adjusts the stated estimate before a price decision.
Summary:
|
Ask a shopper what they would pay for a new product and you will usually get a confident number. Show them the actual price a moment later, and their face often tells a different story.
That gap is the heart of willingness to pay research with emotion AI. Teams working in Consumer Insights have long relied on what respondents say, but a stated number alone is a fragile foundation for a price decision. This guide explains why stated answers run high, what emotion and attention add at the price reveal, and how a research agent turns both into a more defensible price range.
Stated vs Observed Willingness to Pay
Willingness to pay research with emotion AI compares what people say they would pay with how they react when the price is revealed. Stated willingness to pay overstates real willingness to pay by about 21% on average. A research agent captures facial expressions and attention at the price reveal, probes answers that conflict with those signals, and adjusts the stated estimate before a price decision.
Stated willingness to pay is what people tell you in a survey or interview. Observed willingness to pay is what their choices and reactions show, whether in a purchase, an incentive-aligned task, or the way they respond when a price appears on screen.
Price decisions need both. Stated answers are fast and easy to compare across groups, while observed signals show whether those answers deserve trust. Together they give a sturdier read on price sensitivity across your buyer segments.
How Stated Willingness to Pay Is Measured
Most teams collect stated willingness to pay through a willingness to pay survey. The methods fall into two families.
Direct methods ask about price head on:
Open-ended questions ask "what would you pay?" and record the number.
Van Westendorp asks four price questions to map the range of acceptable prices.
Gabor-Granger asks about purchase likelihood at a series of set prices.
Indirect methods infer value from trade-offs. Conjoint analysis and other choice tasks ask people to pick between product profiles, then estimate the value of each attribute, price included.
Every one of these methods records what people say, not what they do.
What Counts as Observed Willingness to Pay
Observed willingness to pay sits close to revealed preference, which draws on behavior instead of opinion. Three sources qualify:
Real purchases, where people spend their own money.
Incentive-aligned tasks such as the Becker-DeGroot-Marschak (BDM) auction, where a random draw can bind people to the price they state, so overstating carries a cost.
In-market price testing, where different shoppers see different prices and sales show the response.
A fourth source is weaker but useful: implicit signals at the price reveal, such as emotion and attention. These do not measure willingness to pay on their own. They indicate whether stated answers can be trusted, and that is exactly where they earn their place in a price test.
Why Stated Willingness to Pay Runs High
Hypothetical bias is the tendency to overstate what you would pay when no money is at stake. It is one of the best documented problems in stated willingness to pay.
A meta-analysis of 77 studies by Schmidt and Bijmolt in the Journal of the Academy of Marketing Science pooled 115 effect sizes and found that hypothetical willingness to pay overstates real willingness to pay by about 21% on average. In plain terms, a price that looks like a comfortable fit in a survey can sit a fifth above what shoppers will actually spend.
The same analysis shows where the bias grows. It is larger for specialty goods than for other product types, larger for higher-value products, and larger in within-subject designs, where one person answers several price questions. It is also larger for indirect methods than for direct ones, which cuts against the common assumption that indirect methods are more accurate.
The pattern is not new. An earlier meta-analysis of 28 stated preference studies in environmental valuation reported a median ratio of 1.35 between hypothetical and actual values. The field differs, but the direction is the same: stated values run above real ones.
Three forces explain the drift:
No real money at stake. Saying yes costs nothing, so a generous answer feels harmless.
Wanting to please. Respondents often give the answer they think the researcher hopes to hear.
No competing alternative in view. Shoppers rarely judge a product alone, yet surveys often show a single option with a single price.
These forces also sit behind the say-do gap in CPG research, where shoppers report one thing and then buy another. Willingness to pay is simply the same gap with a price tag attached.
What Emotion Reveals at the Price Reveal
Emotion gives you a second channel at the moment a price appears. That matters because liking a product and accepting its price are two different judgments.
Brain imaging research makes the split visible. In a Stanford-linked fMRI study published in Neuron in 2007, Knutson and colleagues found that wanting a product engaged the nucleus accumbens, a reward area. A price that felt too high engaged the insula, which is linked to unpleasant feelings and anticipated loss.
The researchers described this tension as the pain of paying. For pricing work, the lesson is practical: product appeal and price pain are processed separately, so a person can love a product and still balk at its price. A survey that asks one question cannot separate the two.
Facial signals offer a window into that moment. Facial coding classifies expressions into signals, and four matter most at the price reveal:
Surprise: the price differs from what the person expected, in either direction.
Confusion: the price or the value case is unclear.
Negative valence: displeasure, common when a price feels too high.
Positive engagement: interest that holds after the price appears.
Early evidence suggests these signals connect to stated answers. In a 2021 study by Suominen on willingness to pay for sport and cultural events, only joy and surprise were positively related to stated willingness to pay, and other emotions showed no clear link. The study is small, so treat it as a promising direction rather than a rule.
What Emotion Data Cannot Tell You
Emotion signals do not convert into a price on their own. A smile at $29 does not mean $29 is the right price. It means that reaction deserves a closer look next to the stated answer.
Facial movement is also an imperfect window into feeling. A major review in Psychological Science in the Public Interest, covering more than 1,000 published findings, reported that people scowl when angry only about 25% of the time on average. Any approach to AI willingness to pay that reads faces should therefore treat expressions as probabilistic signals, best read in aggregate and alongside what people say.
Emotion may also react to the product, the stimulus, or the setting rather than the price. A glossy pack shot or a slow-loading page can move a face as much as a number can. Researchers who measure emotional response to advertising face the same issue: reactions attach to whatever is on screen at that second. The test design must therefore isolate the price reveal.
How a Research Agent Runs an Emotion-Enabled Price Test
A research agent can run an emotion-enabled price test as a structured sequence rather than a single survey question. The flow below keeps the price reveal clean.
Show the value proposition first. Participants see the product, its benefits, and how it is used, with no price attached.
Reveal the price as a separate, timed moment. A fixed exposure window means every reaction maps to the same event.
Capture facial expressions at the reveal. Facial expressions are coded for surprise, confusion, valence, and engagement.
Ask the stated question afterward. Participants answer once the reaction has already been recorded, so the answer cannot rewrite it.
Rotate price points and randomize the starting price. The first number a person sees pulls later judgments toward it, so rotation across participants limits anchoring.
Attention data adds the missing context. Eye tracking shows what people looked at before they reacted: the price, the value claims, or a competing option. A negative reaction after a long look at the price means something different from one that arrives before the price was even read.
Probing Answers That Conflict With Reactions
A research agent watches for mismatches between what is said and what is shown. When a stated "would buy" arrives with a negative reaction, or a "too expensive" arrives with positive engagement, an AI moderator asks a follow-up in the moment. A typical probe might be: "You said you would buy at this price, but you paused when it appeared. What were you thinking?"
Open-ended reasons are then coded into value drivers, the features that justify the price, and price barriers, the concerns that block it. This turns a bare number into willingness to pay analysis you can act on, because each answer arrives with its reason.
Adding a Competitive Reference Point
Shoppers judge price against alternatives. Showing the price next to a realistic competitive alternative lets participants react the way they would at a shelf or on a product page, with a reference price in mind instead of a blank slate.
Kellogg researchers He, Anderson and Rucker built this idea into a comparative method of valuation, published in the Journal of Marketing in 2023. Participants pick a comparison product, such as a KitKat set against M&Ms, Reese's or Snickers at stated prices. They then name the price at which they would be equally happy with either, and a BDM-style lottery decides the outcome. The authors validated the method across 13 studies with almost 3,000 participants.
In an emotion-enabled test, the same frame gives reactions something to be measured against, so a flat expression at your price beside a rival's carries real information.
Reading the Gap Between What People Say and How They React
Once you hold both a stated answer and a reaction for each participant, you can sort responses into a simple matrix.
Reacts positively | Reacts negatively | |
Says yes | Reliable acceptance | Likely overstated willingness to pay |
Says no | The value lands, but the price is a barrier | Clear rejection |
Each quadrant points to a different price decision:
Says yes, reacts positively. This is reliable acceptance. Treat it as core demand at that price point.
Says yes, reacts negatively. This is likely overstated willingness to pay, the group where hypothetical bias hides. Probe these answers, then discount them.
Says no, reacts positively. The value lands but the price is the barrier. Test a lower price, a smaller pack, or a clearer value message before cutting price for everyone.
Says no, reacts negatively. This is a clear rejection. It tells you little about price but a lot about product fit or positioning.
The matrix makes the stated vs revealed preference debate practical: you do not pick a side, you use each signal to check the other.
Adjusting Stated Willingness to Pay With Observed Signals
The next question is how to measure willingness to pay when you hold both signals. Three steps keep the adjustment honest.
1. Discount stated estimates by the share of conflicting responses. Count the answers where the stated response and the reaction disagree, and report the adjusted range next to the raw range.
Consider an illustrative example with hypothetical numbers. Of 200 participants shown a price of $29, 120 say they would buy, a raw acceptance of 60%. If 36 of those 120 reacted negatively, adjusted acceptance falls to 84 of 200, or 42%. You report 42% to 60% as the range instead of a single point.
2. Check against an incentive-aligned task or in-market data where possible. The distance between your adjusted estimate and real behavior gives you a calibration factor you can reuse within a category.
3. Report by segment. The gap differs across buyer groups and categories, so a single blended adjustment hides where overstatement is concentrated. Segment-level willingness to pay shows which groups hold firm and which only sound enthusiastic.
Adding Emotion to Van Westendorp, Gabor-Granger and Conjoint
Emotion layers onto the common willingness to pay survey methods without replacing them.
Van Westendorp: capture reactions to the product and value framing before the four price questions, so you know whether people understood the value they are about to price.
Gabor-Granger: record a reaction at each price shown, to locate where resistance starts. It often appears before stated purchase likelihood drops.
Conjoint: track attention to the price attribute against other attributes in choice tasks. AI agents can run these choice tasks at a scale that makes attention patterns easier to compare.
In each case the stated method still produces the number. The emotion and attention layer tells you how far to trust it.
Limits and Where Researchers Decide
Emotion-enabled price tests have real limits, and honest reporting names them:
Lighting: poor light degrades facial reading.
Webcam quality: image quality affects how reliably faces and gaze are read, a trade-off covered in this comparison of webcam and hardware eye tracking.
Cultural differences: people show emotion differently across cultures, so expression norms vary by market.
Small samples per price cell: rotating prices splits your sample, and thin cells make reaction shares unstable.
Researchers set the adjustment rules and own the final price decision. The agent runs the sequence, flags conflicts, and codes reasons, while people decide how much to discount and what price to take to market. This is the same principle behind human oversight in AI moderated research generally.
Capturing the Price-Reveal Moment at Scale
Facial coding and eye tracking in the same session link how people react to a price with what they looked at before reacting. That pairing is hard to reproduce by hand across many participants and markets.
Decode supports emotion-enabled price tests with 90%+ facial coding accuracy across 62 facial expressions, 96% eye tracking accuracy, and 70+ languages supported for multi-market pricing.
Look for a consumer insights platform that runs both channels in one session and lets researchers set the adjustment rules.
Our guide to consumer research platforms compares the main options.
Frequently Asked Questions
1. What is the difference between stated and observed willingness to pay?
Stated willingness to pay is what people say they would pay. Observed willingness to pay is what their purchases, choices, and reactions show. Price decisions need both, because stated answers are easy to collect but tend to run high.
2. How much do surveys overstate willingness to pay?
A meta-analysis of 77 studies found that hypothetical willingness to pay overstates real willingness to pay by about 21% on average. The overstatement is larger for specialty goods, higher-value products, indirect methods, and within-subject designs.
3. Can emotion AI measure willingness to pay?
Not on its own. Emotion AI reads reactions such as surprise, confusion, and valence at the price reveal. Those signals show whether a stated answer can be trusted. A price still comes from stated or incentive-aligned measurement, adjusted with the emotion data.
4. What facial expressions signal price resistance?
Negative valence and confusion at the price reveal are the clearest warning signs. Surprise needs context, because it can mean the price is higher or lower than expected. Positive engagement that holds after the price appears suggests the value case is landing.
5. How do you reduce hypothetical bias in a willingness to pay survey?
Show a realistic alternative next to the price, use incentive-aligned tasks where possible, and rotate price points to limit anchoring. Add observed signals at the price reveal, then probe any answer that conflicts with the reaction.
6. How does a research agent combine emotion data with Van Westendorp or Gabor-Granger results?
The agent records a reaction alongside each price question, flags answers where the stated response and the reaction conflict, and asks follow-up questions. Researchers then report an adjusted range next to the raw range from the original method.
7. Is willingness to pay different for specialty and everyday products?
Yes. The meta-analysis found hypothetical bias is larger for specialty goods than for other product types, so stated estimates for specialty products usually need a stronger adjustment.
8. How many participants do you need for an emotion-enabled price test?
There is no single number. Plan for enough participants in each price cell to read the share of conflicting responses with confidence, and report segment results only where cells are large enough to be stable.
Price Decisions That Include How People React
A price decision built only on what people say inherits the bias in what they say. Adding how people react gives you a way to check each answer before it shapes a launch.
