The Question Is the Intervention

Every business runs a quiet audit of its customers. The post-purchase survey. The onboarding question that asks what you're here to accomplish. The email that wants to know, on a scale of zero to ten, how likely you are to recommend us. The RSVP that asks whether you'll attend. These feel like instruments — thermometers held up to demand, reading a temperature that exists whether or not you take it.

They are not thermometers. The act of asking changes the reading. Ask a customer to predict what they'll do, and you make them measurably more likely to do it. This is the question-behavior effect, and it means the measurement many brands treat as overhead is already doing the work they pay their campaigns to do — usually by accident, and usually worse than it could.

Asking is not observing

The finding is old, replicated, and consistently underused. In 1993, Vicki Morwitz, Eric Johnson, and David Schmittlein showed that simply asking consumers whether they intended to buy a car or a personal computer raised the rate at which they later bought one. No new information changed hands. No offer was made. The question itself moved the number it claimed to observe. They called it the mere-measurement effect.

It was not a fluke of car buyers. Steven Sherman had already found that people asked to predict whether they'd perform a socially desirable act over-predicted — and then, having predicted, went on to do it more often than a control group who were never asked. Anthony Greenwald's team found that asking people whether they intended to vote raised turnout. Gaston Godin's work found that a survey about blood donation increased donations, and a questionnaire about exercise raised physical activity. The pattern holds across products, civic acts, and health behaviors: put the behavior in front of someone as a question about themselves, and a fraction of them answer by doing it.

Be precise about the size, because the honest number is the whole argument. The most complete meta-analysis of intention and self-prediction questions — 116 tests — puts the effect at roughly d = 0.24. That is small. It is a small influence, not a substitute for doing important work well. It is a thumb on the scale that shifts a few percent of people who were already close. What makes it worth engineering is not its size but its cost: it is attached to activity you are already running and paying for. You are already sending the survey. The persuasion is already happening in the question, and many brands throw it away.

Why the question moves the person

The mechanisms are still argued over — the same meta-analysis found only thin support for any single explanation, which is a useful reminder that a robustly real effect can outrun a tidy theory. But the plausible drivers are worth understanding, because they tell you which questions work.

The first is accessibility. A question drags an attitude out of storage and sets it on the desk. Once "I care about my health" or "I've been meaning to switch" is active in working memory, it exerts more pull on the next decision than it did while dormant. The second is self-prediction as a small commitment. To answer "yes, I'll probably attend" is to make a statement about the kind of person you are, and people work to stay consistent with statements they've made about themselves — the same machinery that makes the small yes so durable once it's given. The third is simulation: to answer the question, you have to imagine yourself doing the thing, and imagined behavior rehearses real behavior.

All three point the same way. The question works when the customer already holds a favorable, accessible attitude and the behavior is easy. The meta-analysis confirms it: the effect is largest for socially desirable, low-risk, easy-to-perform actions. It is weakest — or absent — where the attitude is negative or the behavior is hard. You cannot ask someone into a purchase they don't want. You can ask someone into a purchase they were quietly leaning toward and hadn't yet named.

The measurement you already run is persuading — badly

Now look at what many companies do with this power, which is nothing. They design their questions to extract, not to move.

The typical NPS survey asks a detractor and a promoter the same eleven-point question at the same moment, then files the average. Treated as a reading, it's noise with a decimal point. Treated as an intervention, it's a missed one: you've just asked a satisfied customer to state, in their own hand, how likely they are to recommend you — the exact self-prediction that raises the odds they will — and then you did nothing to make that recommendation easy in the next breath. The moment you manufactured, you wasted.

The product quizzes now spreading across e-commerce are the same story with better production values. The "find your perfect match" flow is sold as a zero-party data tool, and the data is real. But the reason those quizzes convert has less to do with the recommendation at the end than with the eight questions before it, each of which asked the shopper to state a preference and picture themselves using the product. Interact's 2026 benchmark, drawn from more than a decade of data, puts the average quiz's start-to-lead conversion at about 40% — and notes it has held near 40% since 2014, through every shift in traffic and channel. A conversion rate that stable across twelve years is not a function of the traffic source or the offer. It is a function of the format: asking people to answer questions about themselves and, in answering, commit. Brands run these tools to learn about the customer. The tool's quieter job is to move the customer, and it does that whether or not anyone designed it to.

This is the mirror image of a problem we've written about before. Your dashboard runs a campaign on you — the numbers you gather to understand your market end up shaping what you believe and do. The question-behavior effect is the version pointed at the customer: the numbers you gather to understand them end up shaping what they believe and do. In both directions, the lesson is the same. Measurement is not a window. It is a hand on the thing being measured.

The honest version

There is a manipulative way to use this and a durable way, and the difference is not subtle. The manipulative move is to ask people to predict behaviors they don't want, in the hope of shaming a few into consistency, and to ask relentlessly. It fails on its own terms. The same research that found the effect found its reversal: over-asking flips the sign. Morwitz's own work showed repeated questioning can decrease the behavior, as the polarization curdles into resistance. Ask once and you nudge. Ask every week and you train the customer to see the question as pressure and to answer by leaving. The reactance does the rest.

The durable version treats the question as a design surface, not a data grab, and follows three rules.

Ask about behaviors the customer already wants. The effect amplifies an existing favorable attitude; it does not install one. So point the question at the thing the good-fit customer already half-intends — renewing, referring, completing setup, coming back — not at the thing you wish they'd do. If you have to ask someone to predict a behavior they have no appetite for, the honest read is that you have a positioning problem the question can't paper over.

Ask at the moment of decision, then clear the path. An accessible intention decays. The self-prediction that raises the odds of a referral is worth little if the referral link is three clicks and a day away. Put the question where the action can immediately follow it. The survey that asks "how likely are you to recommend us" should, for the promoter, open directly onto the means of recommending. Otherwise you've lit the intention and let it burn down untouched.

Ask once, and mean it. A single, well-placed question about a wanted behavior is an intervention. The same question on a loop is an irritant that reverses the effect you were after. Restraint is not just brand hygiene here; it is what keeps the mechanism working.

None of this replaces a real offer, and the small effect size is the reason to say so plainly. The question-behavior effect is a margin play. It converts a share of the people who were already near the line; it does nothing for the people who aren't, and it will not rescue a product no one wants or a price no one will pay. Sold as a growth hack, it disappoints. Understood as a way to stop wasting the persuasion you're already generating, it compounds.

What the question-behavior effect changes

The reframe is small and it changes everything downstream of it. Stop treating the questions you ask customers as neutral readings of a demand that exists independently of the asking. There is no such reading. Every survey, quiz, poll, and confirmation prompt is a small act of persuasion, running whether you acknowledge it or not — and the only choice you have is whether it runs by design or by accident.

So audit your questions the way you'd audit your ad spend. Which ones ask a good-fit customer to picture a behavior they already want, at a moment when they can act on it? Those are working for you. Which ones ask everyone the same thing on a schedule, harvest the answer into a chart, and leave the customer exactly where they were found? Those are the ones costing you the thing you'd have paid a campaign to produce.

You have been running an intervention this whole time and calling it a report. The brands that pull ahead are not the ones that ask more. They are the ones that understood what asking was.