William Jones·April 15, 2026·5 min read

The Say-Do Gap: Why Your Interviewees Are Lying to You

research methodologybehavioral scienceproduct research

In 1985, Coca-Cola tested a new formula on 200,000 people. They loved it. The company launched New Coke. The public revolted. Within 79 days, Coca-Cola brought back the original.

200,000 data points. All of them wrong.

This is the say-do gap — the persistent, well-documented divergence between what people tell you and what they actually do. It is the single biggest threat to the validity of your user research, and almost nobody designs their research process to account for it.

Why interviewees lie (without knowing it)

Erving Goffman described it in 1959 and the mechanism hasn't changed. Every social interaction is a performance. When someone sits across from a researcher — someone who clearly cares about this product, who is recording the conversation, who is being visibly attentive — the interviewee performs.

They perform competence. They perform helpfulness. They perform the version of themselves that this social situation seems to demand.

This isn't deception. It's impression management, and it happens automatically. The interviewee isn't thinking "I should lie to be polite." They're unconsciously calibrating their responses to maintain a comfortable social dynamic.

The result: your interviewees overstate enthusiasm for features they'd never use, understate frustration with problems they've learned to tolerate, and give you the answer they think a "good participant" would give.

Social desirability bias is the specific mechanism

Researchers call the formal version of this social desirability bias. When a response carries social weight — when one answer sounds smarter, kinder, or more reasonable than another — people drift toward the socially desirable option.

Ask someone if they read the terms of service. They say yes. (They didn't.)

Ask if price matters more than sustainability. They say no. (It does.)

Ask if they'd pay $20/month for your product. They say probably. (They won't.)

The bias scales with the social stakes of the question. The more your question touches on identity, values, or competence, the less you can trust the answer. And the questions that matter most for product decisions — willingness to pay, switching costs, feature prioritization — are exactly the ones most contaminated by social desirability.

Personality makes the gap predictable

Here is where it gets useful. The say-do gap isn't random. It correlates with personality traits in predictable ways.

High agreeableness makes people conflict-averse. An agreeable interviewee will tell you your prototype is "really nice" even if they found it confusing. They won't volunteer criticism unless you push hard, and even then, they'll soften it.

High neuroticism makes people risk-sensitive. A neurotic user will fixate on what could go wrong, express anxiety about changes, and resist anything unfamiliar. In a real interview, social desirability pressure suppresses this. They tone it down. They say "I'd probably get used to it" instead of "this makes me nervous."

Low conscientiousness makes people impulsive. They'll tell you they'd use your budgeting feature every week. They won't. They'll download the app and forget about it.

When personality is invisible — when your persona is just "marketing manager, 30-40, urban" — you can't predict which direction the say-do gap will pull. When personality is explicit, you can.

Synthetic personas don't perform

A synthetic persona with low agreeableness and high conscientiousness doesn't soften its feedback. It doesn't worry about hurting your feelings. It doesn't perform competence or helpfulness.

It tells you your onboarding flow has three unnecessary steps, that the pricing page is confusing, and that it wouldn't pay more than $10/month for something it can do in a spreadsheet. Not because it's been told to be harsh, but because that's how a disagreeable, conscientious person actually responds when social pressure is removed.

Synthicant makes this concrete through two mechanisms.

First, OCEAN scores. Every persona has explicit, calibrated personality dimensions. You can create a persona at Agreeableness 1 (blunt, critical, direct) and interview the same prototype that your Agreeableness 5 persona loved. The delta between those two conversations is a map of your say-do gap.

Second, the biases and flaws field. You can assign specific cognitive biases — confirmation bias, anchoring, status quo bias — and the persona will exhibit them naturally during the conversation. A persona with status quo bias won't just say "I prefer the current version." It will rationalize its preference, minimize the benefits of the change, and find reasons the old way is better. Just like a real user with status quo bias would, if they weren't sitting in an interview trying to be helpful.

How to use this in practice

Run the same interview twice. First with a high-agreeableness persona. Then with a low-agreeableness one. The high-agreeableness persona tells you what your polite customers would say in an interview. The low-agreeableness persona tells you what they'd actually do.

The gap between those two conversations is your say-do risk. If both personas like the feature, you probably have a winner. If the agreeable persona likes it but the disagreeable one tears it apart, you have a feature that will test well and perform poorly.

This doesn't replace real user research. It replaces the dangerous assumption that what users tell you in interviews reflects what they'll do in the wild.

References

Goffman, E. (1959). "The Presentation of Self in Everyday Life." Anchor Books. — The foundational work on impression management, establishing that all social interaction involves performance and audience management.

Jiang, H., Zhang, X., Cao, X., et al. (2024). "PersonaLLM: Investigating the Ability of Large Language Models to Express Personality Traits." Proceedings of NAACL 2024. — Showed that assigned Big Five personalities produce consistent, distinguishable behavior in LLMs with large effect sizes across all five traits.

John, O.P. & Srivastava, S. (1999). "The Big Five Trait Taxonomy: History, Measurement, and Theoretical Perspectives." Handbook of Personality: Theory and Research. — The most-cited overview of Big Five personality taxonomy, establishing agreeableness and neuroticism as predictors of social behavior patterns.

Cohen, R., Keidar, D., Matero, M., et al. (2025). "Personality-Driven Negotiation." arXiv preprint. — Demonstrated that Big Five traits significantly affect AI agent outcomes in negotiation, with agreeableness and extraversion changing concession patterns and agreement rates.