Methodology

The say-do gap starts in the interview. So does closing it.

Stated intent is not a forecast, it is a biased measure. We design each interview to reduce the biases that inflate it, so what you read sits closer to revealed behaviour than to the aspirational self.

5 documented

biases the interview is designed against

4 method

moves that reduce them: recall, probing, no audience, projective

1 click

from any claim to the source verbatim

Why self-report misleads

Stated intent measures the aspirational self

Ask a respondent what they will do and several well-documented biases move the answer before you ever code it. The result is a stated-preference measure that systematically overshoots revealed preference.

Bias	What it does to the answer
Social desirability bias	Respondent shades toward the socially approved choice (healthier, greener, thriftier).
Demand characteristics	Respondent infers the "expected" answer from the question and supplies it.
Intention-behavior gap	Even honest intentions predict only a modest share of later behaviour.
Acquiescence bias	Yea-saying to agree-worded items, independent of true attitude.
Rationalization	Ask "why" too early and you get post-hoc justification, not the actual driver.

A stated intention is the answer a respondent gives their audience. The method decides whether you also reach the answer they would give themselves.

Move 1 · Ask about the last time, not the next time

Anchor in episodic memory, not hypothetical projection

Hypothetical questions ("would you pay more?") recruit the aspirational self and invite anchoring on the frame you supplied. Verso anchors in the critical incident technique instead: concrete, recent, episodic events the respondent actually lived.

Hypothetical projection → episodic recall ("tell me about the last time you...")
Attitude → reported behaviour, the stronger predictor of future behaviour
Abstract preference → a described occasion, with its context, friction and trade-offs intact

Don't

"Would you switch to a greener option?"

"Walk me through the last product you switched. What tipped it?"

Critical incident techniqueBehavioral recallMeans-end laddering

Move 2 · Press the contradiction

The first answer is where the aspirational self hides

A fixed script records the surface answer and moves on. Verso's moderator is adaptive and non-directive: it holds a stateful model of the conversation and generates follow-ups toward each objective, within a probe budget, without leading or suggesting.

Detects the attitude-behavior contradiction ("price is my barrier" said moments after describing an unhesitating premium purchase) and probes the gap rather than logging both.
Ladders from surface attribute to functional and emotional consequence, toward the underlying value.
Stays neutral — no suggested response options, so no anchoring and no acquiescence handed to the respondent.

Freedom inside a frame

Dynamic follow-ups keep roughly nine interviews in ten on objective. The trade-off, versus a human chasing a truly off-frame thread, is consistency at scale; we widen the frame to catch what matters.

Non-directive probingDynamic follow-upsContradiction pressingStateful memory

Move 3 · Remove the audience

Strip out interviewer effects and social desirability

Self-disclosure rises when the audience disappears. Each interview is a real-time voice and video conversation, but with no human in the room, which removes the interviewer effects and evaluative pressure that drive social desirability bias, on exactly the topics where it bites hardest: health, money, personal habits.

The confessional effect

Respondents disclose to the AI what they hold back from a person. No one to impress, no reaction to read, so they say the awkward, honest thing.

Reducing social desirability and demand effects raises the ecological validity of the response, so the transcript reflects behaviour the respondent would actually own rather than the version staged for a moderator.

Social desirability reductionInterviewer-effect controlSelf-disclosure

Move 4 · Ask sideways

Reach what people can't, or won't, say directly

Some drivers never survive a direct question: the respondent either can't articulate them or won't admit them. Projective techniques get at them indirectly, and Verso runs them as a crescendo, from lightly to fully projective, once rapport is set.

Third-person framing — "what would people like you say about this?" lets the respondent voice a socially risky view as someone else's.
Personification — "if this brand were a person, who would it be?" surfaces associations a rating scale flattens.
Completion and association — unfinished sentences and first-word prompts bypass the rehearsed answer.

Projective techniquesProjective crescendoThird-person framingPersonification

Before the answer counts

An unbiased answer is worthless if the respondent isn't real

Reducing bias assumes a real person behind the response. Panel fraud has gotten good, and coherent answers now come from bots and click farms; video and voice expose what a text survey never sees.

Signals we read — speech patterns, facial presence, response latency, behavioral consistency.
Before launch — dozens of dry-run interviews against AI participants until the system holds up.
During fieldwork — every interview scored in real time, with a daily manual sample review by our team.
Before analysis — weak interviews flagged and excluded; we work with vetted panels and score respondents across studies, so quality compounds.

Fraud detectionVetted panelsReal-time scoringManual review

Trust, but verify

A human signs off before you read a word, and every number is audited

Reducing bias only counts if the synthesis stays honest too. Each finding is tagged as supported or contradicted against the coded answer set, then two steps stand between the data and the report.

Human checkpoint — a researcher approves the outline before anything is written, so the story follows the data, not the model's momentum.
Claim audit — every quantitative claim is checked against the tag data, so no number drifts from what respondents actually said.
Traceability — each claim links to the timestamped excerpt and supercut behind it.

Our first differentiator

This human-in-the-loop pass is the one thing no competitor runs systematically. It is why a finding survives a CFO pushing back on it.

A stakeholder can push back on a finding, and you can open the exact second a real respondent said it.

Reading stated intent

Treat intent as one signal, weighted by friction

None of this discards stated intent. It relocates it: an optimistic signal, read next to reported behaviour and the frictions the respondent names, never as a standalone forecast.

Anchor in recent behaviour before any future-facing question (order matters: behaviour first, projection last).
Reserve room to probe contradictions instead of accepting the first response.
Lower evaluative pressure so respondents stop performing.

Stated intent tells you who your consumer wants to be. Method decides whether you also learn who they are.

Ready when you are

See the method on your own question

Tell us what you want to learn about your customers. We'll show you how a Verso study would run.

Book a call →