Your users aren’t lying to you. Their memory is reconstructing.

An interview participant can be candid, thoughtful, and wrong.

That is not necessarily deception. Memory is not a recording that plays back on demand. When people cannot retrieve the exact episodes a question assumes, they estimate, combine similar events, use the wording as a cue, and build an answer that feels true.

The practical response is simple:

• Ask about one recent episode instead of “usually”.
• Use the participant’s words rather than supplying adjectives.
• Check frequency and recency in analytics when you can.
• Shorten the recall window and anchor it to a known event.
• Use diaries or observation when routine behaviour is the research question.
• Work from transcripts, not your own memory of the interview.

These moves reduce the burden on memory. They do not make recollection infallible, so keep the evidence limits visible.

“Usually” is an estimate

Suppose you ask, “How often do you export a report?”

The participant may not keep a mental count. They might recall last week, think about what their role normally requires, choose a round number, and answer “most days”. The answer is not a memory of every export. It is an estimate assembled during the interview.

Survey researchers Norman Bradburn, Lance Rips, and Steven Shevell described this problem in Science. Questions often assume people can remember and enumerate autobiographical episodes. When they cannot, respondents use partial information from memory to construct a numeric answer.

Teresa Torres offers a useful, explicitly illustrative example in her story-based interview guidance. A participant estimated that she watched Netflix for about two hours a week. Asked about the last occasion, she described 2 hours and 40 minutes on one day. That single story does not reveal her typical weekly use either. It reveals why the generic estimate should not be treated as behavioural data.

If your product records exports, logins, or feature use, let the product answer the frequency question. Use the interview to understand the surrounding episode: what the person was trying to do, what else was happening, what went wrong, and what they did next.

Generic judgements and recent episodes answer different questions

The Day Reconstruction Method asks people to rebuild the previous day as a sequence of episodes. In the original study, 909 employed women described activities and feelings episode by episode rather than giving only broad judgements about their lives.

The method matters because a category judgement and an episode report are not interchangeable. “Onboarding is easy” is a belief about a category. “I signed up last Tuesday, opened the import page, and stopped when I saw the CSV format” is an account of an event.

The event report is usually more useful for product decisions because it contains a trigger, sequence, context, and consequence. But it is still recalled evidence. A recent story can omit details or borrow from similar occasions. Asking about the last time improves specificity; it does not certify accuracy.

Nielsen Norman Group recommends asking about specific events to jog memory and obtain meaningful detail. The same guide is clear about the boundary: interviews gather reported behaviour. If you need to know what people actually do, use observation, analytics, or other behavioural measures alongside the interview.

Four ways an honest answer can go wrong

1. The participant infers instead of recalling

Frequency, duration, and totals are especially demanding. “How many times did you contact support last quarter?” assumes the participant can find and count several similar episodes. If they cannot, they may estimate from a typical month or from how active they believe they are.

Treat rounded numbers as leads, not measurements. Ask for the most recent occasion, then check the count elsewhere.

2. Events move across time boundaries

People can remember that something happened but misplace when it happened. Neter and Waksberg’s Census Bureau study of household repairs found both omissions and forward telescoping: some older expenditures were reported as more recent than they were. Longer recall periods also produced fewer reports of small jobs.

The product-research version is familiar. A customer says a failure happened “last month”, then the support ticket shows it was three months ago. The underlying incident may be real even when its date is wrong.

Use shorter windows and personal landmarks:

• Since the plan changed
• After your colleague joined
• During the last reporting cycle
• The week you cancelled

Then walk backwards and forwards through the sequence. A landmark gives the memory a boundary; it does not guarantee correct dating.

3. The question supplies part of the answer

In 1974 Elizabeth Loftus and John Palmer showed participants films of car accidents and changed one verb in a speed question. The mean estimate was 40.5 mph when the cars “smashed” and 34.0 mph when they “hit”. In a second experiment, participants were asked a week later whether they had seen broken glass. There was none in the film, yet 32% of the “smashed” group said yes, compared with 14% of the “hit” group.

The study does not mean every adjective creates a false memory. It does show that wording can affect an estimate and a later report.

In an interview, “Was the setup frustrating?” supplies both an evaluation and a label. Ask: “What happened during setup?” If the participant says “frustrating”, use their word in the follow-up. Do not upgrade “slow” into “painful” or “confusing” into “broken”.

This is one reason leading questions are a data-quality problem, not merely a breach of interview etiquette.

4. A remembered evaluation can overweight particular moments

Retrospective evaluations do not always reflect the duration or average intensity of an experience.

In a 1993 cold-water experiment, 22 of 32 participants chose to repeat a 90-second trial whose final 30 seconds were slightly less unpleasant rather than a 60-second trial at the original temperature. A later randomised study of 682 colonoscopy patients found that adding a short, milder ending improved remembered unpleasantness.

These studies concern physical pain, not SaaS onboarding. Do not turn them into a claim that every product experience follows a universal peak-end rule. They offer a bounded warning: a person’s overall verdict can be shaped by salient moments and the ending, while ordinary parts of the experience receive less weight.

When a customer says the product was “always clunky”, ask for separate episodes:

• the last ordinary use;
• the worst recent use;
• the final use before cancellation.

Comparing those stories helps you see whether the verdict describes a repeated pattern, one severe incident, or the ending.

Why a direct “why?” can produce a tidy theory

“Why did you choose us?” sounds like a request for memory. It often asks for an explanation the participant has never had to formulate.

Nisbett and Wilson’s classic 1977 paper, Telling More Than We Can Know, included a study in which people chose among identical stockings displayed in different positions. The right-most pair was selected far more often, but participants explained their choices using qualities such as knit and sheerness rather than position.

One old laboratory result should not be treated as proof that people never know their motives. The useful lesson is narrower: confidence and fluency do not show that an explanation was retrieved intact.

Replace “why?” with a sequence:

• What was going on before you started looking?
• What did you try first?
• What changed?
• Who else was involved?
• What happened immediately before the decision?

The reason often appears inside the story. Our guide to asking why without asking why gives more examples.

A practical rewrite table

Notice that “tell me about the last time” is not always enough. Nielsen Norman Group’s critical-incident guidance separates three prompts:

• “Tell me about a time” lets the participant choose any example.
• “Tell me about the last time” retrieves the most recent example.
• “Tell me about a time it helped you succeed” deliberately retrieves a significant incident.

Use the question that matches your research goal. If you want the ordinary workflow, do not ask only for the best or worst event. If you want to understand a rare breakdown, a critical incident is appropriate.

Help memory without filling the gaps yourself

Five techniques help: some belong in the conversation, others in the study design.

Shorten the recall window

Prefer last week to last quarter. For churn, contact participants soon after cancellation when practical. For onboarding, invite them after the first session rather than months later.

Timing is part of study design. A tool cannot make a distant event recent after the fact.

Give the participant time

Do not fill every silence. NN/g notes that recalling an incident can take time. A pause after “tell me about the last time” may be retrieval, not confusion.

Decompose the event

Move through one episode:

1. What started it?
2. What did you do first?
3. What happened next?
4. Where did you stop or change course?
5. What was the consequence?

Avoid asking five questions at once. Use them as successive probes.

Match the participant’s language

If you say “integration” while the participant thinks “weekly upload”, your term may not cue the relevant memory. Ask what they call the task, tool, or hand-off.

Use diaries for routine or longitudinal behaviour

Routine events are easily blended. Diary studies let participants record activities and context as they happen, with follow-ups while the experience is fresher. They are often a better fit for repeated workflows, changing habits, or behaviour that unfolds over days.

Your memory is part of the method too

Interviewers also reconstruct.

You may leave a call remembering the sharpest complaint and the enthusiastic ending. A colleague may remember a different moment. If you synthesise from memory or sparse notes, the most vivid participant can become “what users said”.

Record with consent, keep the transcript, and connect every finding to a source passage. Separate:

• what the participant said;
• what happened in the recalled episode;
• what you infer;
• what behavioural evidence confirms or contradicts it.

See how to take notes in a user interview for a compact evidence format.

How Maren can support recall-aware interviews

Maren lets a researcher define the objective and guide, then runs adaptive interviews that probe short or vague answers. Its Feature Feedback style asks for specific moments and step-by-step accounts when a participant stays general. Every conversation produces a transcript, an individual summary, and evidence that can feed cross-interview synthesis.

You still need to design the study correctly. Put recent-event questions in the guide. Invite participants close to the event when that matters. Review the transcript and corroborate frequency, recency, and sequence with product data where possible.

Maren does not observe behaviour merely because it conducts the interview, and synthesis does not remove recall bias from the source material. The advantage is consistency and traceability: you can ask the same recall-aware core questions across many conversations, follow each participant’s thread, and return to their exact words instead of relying on your memory of the call.

Your users are usually trying to help. Ask questions that give memory a fair chance, then keep the difference between a story and an event log visible.