How to Run Unmoderated User Testing: A Practical Guide

Learning how to run unmoderated user testing is mostly about learning to write tasks that survive without you in the room. In an unmoderated study there is no researcher on the call: as the Nielsen Norman Group puts it, a software application “provides instructions to users, records their actions, and may ask them predetermined followup questions.” That is the whole method and its whole risk. Get the instructions right and you can collect feedback from dozens or hundreds of participants, sometimes within a few hours. Get them wrong and, as NN/G warns, “if users misinterpret your instructions and perform the wrong task, your test is wasted.”

This guide covers when unmoderated testing is the right tool, how to write tasks that stand on their own, how many participants to run, and the jobs it is wrong for. It is vendor-neutral. Where GhostClick fits, this guide says so plainly, and it says where it does not.

What unmoderated testing actually is

Unmoderated testing means the participant works alone, on their own schedule, following written tasks while a tool records what they do and say. Nobody is watching live. That difference is the source of both its strengths, scale and speed, and its one structural weakness: there is no one to catch a misunderstanding in the moment.

In a moderated session the facilitator is a safety net. In an unmoderated session there is none, so the instructions have to be the safety net instead.

When it is the right tool

Unmoderated shines when the question is narrow and the thing being tested already works. NN/G is specific: unmoderated studies “work best for evaluating live websites and apps or highly functional prototypes,” and, in their guidance on remote testing, are best kept to a few specific elements rather than an overall review. It is at its strongest for evaluative, measurable questions: can people complete this checkout, find this setting, understand this label?

• You are testing a live product or a realistic prototype, not a rough sketch.
• The question is focused, and success is observable (a task completes, or it does not).
• You need results fast, or at a scale a moderator could never schedule.

Write tasks that stand on their own

This is the craft of unmoderated testing, and where most studies are won or lost. Because no one can clarify, every task has to be self-contained. NN/G’s rule is that tasks should be specific, realistic, and actionable, without including hints that make the task too easy. A task that gives away the answer tests nothing; a task that is ambiguous wastes the session.

• Set a scenario, not a click path. “You want to change the card your account is billed to” tests findability; “click Settings, then Billing” tests nothing.
• Do not leak the interface’s own wording. If the button says “Manage plan”, do not use that phrase in the task, or you are testing reading, not finding.
• Say explicitly when to stop. NN/G notes that instructions must tell users when they are done, because the moderator is not there to move them on.
• Pilot every task on one or two people first. A single pilot catches the ambiguous wording that would otherwise waste your whole sample.

How many participants

Unmoderated studies are often run for measurement, which pushes sample sizes up. A rough rule: around five participants is enough for a qualitative pass that just surfaces the big problems, but a quantitative study measuring success rates or task times needs roughly 20 to 40. Because recruiting and running are cheap once the tasks are written, unmoderated is where larger numbers actually become practical. The full reasoning is in our guide to how many participants qualitative research needs.

Where unmoderated is the wrong tool

The honest part. Unmoderated testing is a poor fit whenever the value is in the conversation. NN/G is blunt that early-prototype testing is difficult without a moderator to explain and help people recover from errors, and that participants “behave less realistically in tasks that depend on imagination, decision making, or emotional responses.” There is also no way to ask a detailed follow-up about something surprising a participant just did, which is exactly where discovery research lives.

• Generative or exploratory research, where you do not yet know the right questions.
• Early, rough prototypes that need a human to explain them.
• Anything emotional or imaginative: real buying decisions, sensitive topics, reactions you need to probe.

Where an AI moderator changes the trade-off

Classic unmoderated testing forces a hard choice: you get scale and speed, but you lose the follow-up question. GhostClick sits between the two. It is asynchronous, so participants still complete the interview on their own schedule at scale, but the interview is run by an AI moderator that can ask the follow-up a static task list cannot. You bring your own participants, GhostClick gives each a private link or emails the invites, then interviews them by text or voice in seven languages and turns the transcripts into analysis.

It is not a fit for everything, and the limits above still bite. If your study is deeply generative, needs a human reading body language in real time, or hinges on emotional reactions you must probe live, a moderated session is still the right call. GhostClick is strongest for structured discovery and evaluative interviews at a scale live moderation cannot reach. The fuller comparison is in async AI-moderated versus live interviews, and once you have recruited people, the questions themselves decide the outcome, which is the subject of writing a discussion guide that gets honest answers.