How to run unmoderated user research

The standard objection to unmoderated user research is that it loses the live follow-up. The moderator cannot lean in when the participant says something half-finished, cannot ask "what did you mean by that?", cannot read the pause between two sentences and decide to wait. Without the moderator, the argument goes, the data is thinner.

The argument has a quiet assumption inside it: that what makes interviews valuable is the moderator. On the studies we run, that assumption is usually wrong. What makes interviews valuable is the participant talking honestly about something real, on a day they are willing to do it, in a medium that does not flatten the answer. When you change "live follow-up" to "right prompt, right medium, right time", a lot of moderation turns out to be ceremony.

This is a working guide on how to run unmoderated user research: what it is, when it beats a moderated session, the steps from prompt to synthesis, and the operational details that decide whether the data comes back rich or thin.

What unmoderated user research is

Unmoderated user research is a qualitative method in which participants complete the study on their own time, on their own device, without a researcher in the session. The instructions, prompts, and recording happen inside a tool. The output is a per-participant artifact (a transcript, a voice note, a screen recording, a survey response) that the researcher reads or listens to later, asynchronously.

Nielsen Norman Group's framing holds: in an unmoderated session, the participant works through the tasks alone and the researcher reviews the output afterwards. The trade is real-time probing for two things: scale (more participants, faster) and honesty (no observer in the room).

The category covers more than usability tests. Async voice interviews, diary studies, prototype tests on platforms like Maze, open-ended survey questions with voice or text answers, take-home concept tests, and even code-along sessions where developers narrate their reasoning all qualify as unmoderated research. The shared shape is: no live moderator, async completion, qualitative output.

When unmoderated user research beats moderated

Three cases where unmoderated is the right call:

• The signal you need is honesty, not depth. A participant alone in their kitchen at 9pm, recording a voice note about why they cancelled a subscription, will say things they will not say to a stranger on a Zoom call. Unobserved is closer to truthful for a wide class of questions.
• You need reach across time zones. Ten participants in five countries cannot all be on a call this Thursday. Unmoderated research lets the same study run in parallel across the recruitment list with no scheduling layer.
• The behavior is rhythmic or rare. A morning routine, a weekly review, a once-a-month decision. The moderator cannot be there at the moment the behavior happens. Unmoderated research can be, because the participant records inside their own day.

Three cases where moderated is the right call:

• You are chasing one specific decision deep. A 45-minute moderated interview can probe the same thread five turns down. Unmoderated research can ask a follow-up, but the follow-up does not adapt in real time the way a moderator does.
• The participant is unfamiliar with the medium. B2B research with executives on tools they have never used, accessibility research with participants who need help with the device, anything that requires the researcher to clarify the task as it unfolds.
• You only have two participants. Sample of two means the cost of running a moderated session is small relative to the cost of a misread prompt. Unmoderated research pays back at six participants and up.

This is not an either/or. The pattern that works on most product teams is unmoderated to figure out what to ask, then a single moderated conversation to chase the one clip that did not make sense. The companion post on async user research methodology covers the broader case for shape; this guide is the practical version for unmoderated specifically.

How to run unmoderated user research, step by step

Six steps, in order. They are tuned for unmoderated work but borrow from the methodology playbook.

01 · Frame a question that survives without a moderator

Write one sentence: what is the question that, asked clearly once, returns a useful answer with no follow-up? If the answer is "I would need to clarify what I meant", the prompt is not ready for unmoderated research yet. Rewrite until the participant can read it once on a phone screen and start talking.

The single most common failure of unmoderated work is putting a prompt designed for a moderator into a tool that has no moderator. Prompts written for a 1:1 call assume the participant will hear "tell me more" within twenty seconds. Without that cue, ambiguity in the prompt becomes ambiguity in the data, and the researcher only finds out a week later.

The craft of writing prompts that open people up is covered separately in how to write user research questions; the unmoderated-specific rule is shorter: the prompt is the moderator. Spend more time on it than feels reasonable.

02 · Pick the medium that matches the answer

Unmoderated research is not one method. It is a category of methods, and the wrong medium guarantees thin data.

• Open-ended text fields. Cheap, fast, low-friction for the researcher. High friction for the participant, who has to translate a messy thought into typed prose. Average answer length on a research study we run sits around 31 words for typed responses. Use for short factual answers, screening questions, or anywhere a sentence is genuinely enough.
• Voice notes. Roughly 4× the answer length on the same prompts (around 140 words on average), with a response rate about 2.7× higher than typed equivalents. Use for "tell me what happened", "walk me through", "what did you think when". The medium does most of the work.
• Screen recording with think-aloud. A participant talks through their reasoning while completing a task in a prototype. Use when the question is "where does this design break down?" Tools like Maze, UserTesting, and Lookback handle the recording.
• Diary entries. A series of short prompts across two to four weeks. Use when the answer is an arc, not a moment. The full method sits in how to run a diary study with voice notes.

The decision is mechanical: match the medium to the shape of the answer you need. Most teams default to text because it is what their tools default to. The output is a thinner version of the truth.

03 · Design prompts that are short, specific, and self-contained

Three rules that decide whether unmoderated prompts work:

• One question per screen. A list of five prompts on one page is a survey, and participants answer it like a survey: shortest possible version, fastest possible exit. One open prompt, alone, signals a different mode of attention.
• Anchor the question to a specific moment, not a general feeling. "Walk me through the last time you tried to import a CSV" works. "How do you feel about importing CSVs in general?" does not, because the participant has no specific memory to reach for and the answer turns into hedged abstraction.
• Keep the prompt under twenty words on the screen. Long prompts on a phone screen get scrolled past. The participant taps record before they have read the second clause, and the answer aims at the first clause only.

A useful test: read the prompt out loud once, at the speed someone would read it on a phone in line at a coffee shop. If you can answer it in that moment, it is ready. If you find yourself wanting to add "by which I mean...", rewrite.

04 · Recruit, route the link, lower the start friction

Unmoderated studies live or die on whether the participant gets from "I clicked the link" to "I am answering the first question" in under thirty seconds. Most do not.

The friction stack to remove:

• No login, ever. Every required field on the entry screen drops completion. The respondent should be able to land on the link, see the first prompt, and start answering. Account creation, demographic forms, and consent walls between the link and the question are the largest avoidable cause of thin participation.
• Mobile-first means mobile-tested. Three-quarters of unmoderated participation on consumer studies happens on a phone. Test the link on an actual phone, not the responsive view in a desktop browser. The full case for designing the respondent experience for the device sits in mobile user research methods.
• Send the link in the channel the participant already uses. Email opens at single-digit rates among consumer recruits. SMS, WhatsApp, or the channel that brought them in performs three to five times better.

Recruit roughly 1.3 to 1.5× your target finishing count. Unmoderated dropouts are higher in the first thirty seconds and lower across the rest of the study than moderated equivalents. If you want twenty finishing participants, send the link to thirty.

05 · Capture context, not just answers

The reason unmoderated research has a reputation for thin data is that most tools record the answer and discard the context. The transcript arrives as text. The hesitation is gone. The half-laugh at the end of a sentence is gone. The pause where the participant was thinking is gone. What is left is a faintly irritable cousin of what the participant actually said.

The fix is to keep the audio. A voice answer with a transcript and an audio clip beats a transcript-only answer by a wide margin, because the moments that carry signal (hesitations, corrections, "actually, hold on") survive the recording and not the transcription. The longer essay on what voice catches that text loses sits in the voice versus text piece; for unmoderated work specifically, the rule is: if your tool gives you transcripts and not clips, you have rebuilt a survey in a more expensive way.

06 · Synthesize without losing the "why"

The synthesis pass on unmoderated research has the same shape as on moderated work, with one addition: because the data is async, it is tempting to read it as it comes in. Don't. The first three responses anchor the entire synthesis. Whoever finished first becomes the implicit baseline for "interesting" and every subsequent response is read against them.

Wait until the window closes (or until you have at least six to ten finishing participants), then read each participant's thread end to end before clustering across them. Code for themes second, not first. The mechanics of pulling themes from transcripts are covered in how to analyze user interview transcripts; the unmoderated-specific addition is that thematic saturation lands faster on voice data than on text data, because each answer is denser. Six to ten finishing participants per homogeneous group is usually enough.

"I think I just stopped opening it. There was no specific moment. It was, like... I'd see the icon and I'd think, oh right, that thing, and then I'd open something else. So when the renewal email came I didn't even click it. I just let it lapse."

That answer is forty-one seconds of voice. Typed, the participant would have written "stopped using it" and the renewal-email beat would not have been written down. The unmoderated work captured what the moderated call would have asked four follow-ups to reach.

Common failure modes

Three patterns that show up in unmoderated studies that come back thin, every time.

The prompt assumes a moderator. Phrases like "in your own words" or "feel free to elaborate" are vestigial; they exist because the participant in a 1:1 call needs reassurance to keep going. In unmoderated work, those phrases just take up screen space and signal that the question is not specific enough to stand on its own. Strip them.

The friction is in the wrong place. Most unmoderated dropouts happen between "clicked the link" and "started answering". The temptation is to add more questions to compensate for fewer participants, which makes it worse. The fix is the inverse: fewer questions, less entry friction, shorter prompts.

The transcript is the only artifact. A study that produces transcripts and no audio is half a study. The participant said the words; the transcript is a flattened summary of the words; the signal is in the part the transcript discards. Most general-purpose research tools default to transcript-only. Voice-first tools, including Talkful, keep both and let you read fast and listen to the entries that matter.

Where unmoderated research is heading

Two shifts that change the math on unmoderated work in 2026:

The first is voice. Phone keyboards, AirPods, and a generation of users who already send voice notes more than typed messages on WhatsApp have moved the cultural baseline. Unmoderated voice research now has the same friction profile that text-based research had a decade ago, while text-based unmoderated research has steadily lost ground because typing on a phone, alone, in the evening, is a worse user experience than it has ever been.

The second is AI synthesis. The bottleneck on unmoderated research has historically been not collection but reading. Twenty voice answers at two minutes each is forty minutes of audio plus thirty minutes of skimming to find the three useful clips. Modern transcription and clustering close that loop to under five minutes per study, with the audio kept alongside for the moments that matter. The full guide to voice user research and where the field is going sits in the voice user research pillar.

FAQ

What is unmoderated user research?

Unmoderated user research is a qualitative method in which participants complete the study on their own time, on their own device, without a researcher in the session. The instructions, prompts, and recording happen inside a tool, and the researcher reviews the output asynchronously. It includes async voice interviews, prototype tests, diary studies, open-ended surveys with voice or text answers, and screen recordings with think-aloud.

Is unmoderated user research the same as a survey?

No. A survey aggregates closed-ended answers across many participants and is read as a number. Unmoderated user research collects qualitative artifacts (voice notes, transcripts, recordings) per participant and is read as text. The output of a survey is a chart; the output of an unmoderated study is a stack of transcripts you read end to end. Most unmoderated work uses open-ended prompts, not multiple-choice.

How many participants do you need for unmoderated user research?

Six to ten finishing participants per homogeneous group is usually enough for thematic saturation, following the same logic Guest, Bunce and Johnson found in their interview-saturation study. Recruit roughly 1.3 to 1.5× your target to absorb the entry-friction tail. Sending the link to thirty people to close twenty finishers is a reasonable planning ratio for a five-question voice study.

What is the difference between moderated and unmoderated user research?

Moderated user research has a researcher in the session live, asking questions and probing follow-ups in real time. Unmoderated user research has the participant complete the study alone, asynchronously, with no live researcher. Moderated work is best for one-on-one depth on a specific decision; unmoderated is best for honesty, scale, and reach across time zones. Most product teams use both, in that order: unmoderated to figure out what to ask, then moderated to chase a specific clip.

Can unmoderated research replace user interviews?

Not entirely, and not for every question. Unmoderated voice interviews replace the bulk of standard discovery work (sentiment, satisfaction, friction, "what happened when") because the medium itself does most of what the moderator was doing. Moderated interviews remain the right method when you are chasing a single decision several turns deep, doing accessibility work that needs help in the room, or running a study with a sample of two where the per-session cost is small relative to the risk of a misread prompt.

What tools work best for unmoderated user research?

It depends on the medium. For unmoderated voice and async interviews, Talkful is what we make and use. For unmoderated prototype tests with screen recording, Maze, UserTesting, and Lookback are common picks. For unmoderated diary studies, voice-first tools that keep the audio alongside transcripts produce richer data than text-only platforms. The shared rule across tools: the entry path from "clicked the link" to "answering the first question" should be under thirty seconds, and the audio (if there is any) should be kept, not just the transcript.

---

Unmoderated user research is not a lesser interview, and it is not a survey with extra steps. It is a different shape: short, specific prompts, completed alone, in a medium that does not flatten the answer. The reason most unmoderated studies come back thin is that the prompt was written for a moderator who is not there, the friction was on the entry screen, and the transcript discarded the part of the answer that carried the signal. Fix those three and the data behaves. Talkful has a free plan that's enough to run a first unmoderated voice study against the way you do research today, and the end-to-end voice user research guide covers the wider methodology once the answers start landing.