Usability Test Plan
When To Use
The prototype is ready and the script asks "do you like it". Run it once the sessions are booked and before anyone sits with a participant. It answers: which tasks, which words and which notes will show where people actually get stuck?
When Not To Use
With no moderator available, use Unmoderated Test Plan; to track metrics across rounds, use Usability Benchmark Study. If what you test is the AI's answers, wrong outputs and trust, use AI Feature User Study; if the AI is incidental and only the flow is tested, stay here.
Inputs
- The research questions, the parts of the design under test, and the prototype (a link, screenshots, or frames read through the Figma connector) with invented test data
- Who takes part and the session length
If you have none of this, I start from the one flow you most need to see used and mark the output as a first draft.
Approach
Moderated usability testing as the GOV.UK Service Manual describes it, with thinking aloud and the five-users-per-round rule and its limits, both from Jakob Nielsen at NN/g. Tasks, not tours: a realistic goal with the path unstated, then silence while the person works. Five people per distinct group per round finds problems; it measures nothing. The failure it prevents: a script that says "click the expenses tab and see how easy it is", which only ever finds happiness.
Workflow
- Ask up to three questions: the research questions, what part of the design each task tests, and who takes part (one group or several).
- Tie each test goal to a research question and a part of the design. A goal with no question behind it is cut.
- Write task scenarios: "you need to [goal]; go ahead", with the starting point stated and no interface words that give the answer. Write success criteria per task before testing: what counts as completed, completed with help, not completed.
- Write the think-aloud ask and the neutral prompts ("what are you thinking?", "what did you expect?"). When they struggle, wait. Talking changes timing, so no time on task is reported from these sessions.
- Write the moderator script: intro ("we are testing the design, not you"), consent check, tasks, post-task questions that stay open, close. Never "do you like it" or "would you use it".
- Build the observer note grid per task: outcome against the criteria, path taken, hesitations, verbatim quotes with timestamp.
- Set the sample and logistics: five per round per distinct group as a rule of thumb for finding problems, not a measurement; session length you set, with gaps between sessions for notes. Pilot once and fix the tasks before the first real session.
Output Format
# Usability Test Plan
**Design under test:** [name, version] | **Groups:** [groups] | **Sessions:** [n per group, dates] | **Length:** [min]
## Test goals
| Goal | Research question | Part of the design |
|---|---|---|
| [goal] | [RQ] | [screen or flow] |
## Tasks and success criteria
| # | Scenario as read aloud | Start point | Completed | Completed with help | Not completed |
|---|---|---|---|---|---|
| 1 | "You need to [goal]; go ahead" | [screen] | [criteria] | [criteria] | [criteria] |
## Moderator script
1. Intro: "We are testing the design, not you." Consent check: [items].
2. Think-aloud ask: [words]. Neutral prompts: "What are you thinking?" "What did you expect?"
3. Tasks 1 to [n], each followed by [open post-task question]; close with [open question] and thanks.
## Observer note grid
| Task | Outcome | Path | Hesitations | Quote and timestamp |
|---|---|---|---|---|
| [#] | [completed / with help / not completed] | [path] | [where] | "[verbatim]" [mm:ss] |
## Sample and pilot
[n per group, why; what this round can and cannot show; pilot date and changes]
## Decision
[Research lead] signs off the tasks and criteria after the pilot by [date].
Done When
- Every task has a scenario with the path unstated and success criteria written before testing
- The script contains no "do you like it", "would you use it" or leading prompt
- The observer grid matches the tasks, and a pilot is booked
Quality Bar
- Prototypes use invented test data, never a participant's real account on screen
- No time on task from think-aloud sessions; metrics belong to the benchmark
- Five per round is stated as a rule of thumb for finding problems, with its limits
- Claude writes the tasks; real people attempt them, and only what they did becomes a finding
Next
Run uxr-usability-findings (Usability Test Findings) to turn the sessions into rated issues.
About the makers
This pack is made by 90n (90n.ai): AI-native ways of working and tools for Product, Marketing and HR teams, built by ex-McKinsey founders. We map where time and money leak, redesign the work around AI and stay until your people run it themselves, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).
1---2name: uxr-usability-test-plan3description: Writes a moderated Usability Test Plan with test goals, realistic task scenarios with success criteria set before testing, a moderator script with think-aloud and neutral prompts, and an observer note grid with pilot and sample notes. Use for "run uxr-usability-test-plan", "write a usability test script", "usability testing plan", "prep the prototype sessions", "task scenarios for a usability test", "think aloud script", "how many users do I need", "my script asks do you like it", part of the UX Research with Claude Pack by 90n.4metadata:5 source: https://skillmd.com/skills/polar-bear-org/uxr-usability-test-plan6---78# Usability Test Plan910## When To Use11The prototype is ready and the script asks "do you like it". Run it once the sessions are booked and before anyone sits with a participant. It answers: which tasks, which words and which notes will show where people actually get stuck?1213## When Not To Use14With no moderator available, use Unmoderated Test Plan; to track metrics across rounds, use Usability Benchmark Study. If what you test is the AI's answers, wrong outputs and trust, use AI Feature User Study; if the AI is incidental and only the flow is tested, stay here.1516## Inputs17- The research questions, the parts of the design under test, and the prototype (a link, screenshots, or frames read through the Figma connector) with invented test data18- Who takes part and the session length19If you have none of this, I start from the one flow you most need to see used and mark the output as a first draft.2021## Approach22Moderated usability testing as the GOV.UK Service Manual describes it, with thinking aloud and the five-users-per-round rule and its limits, both from Jakob Nielsen at NN/g. Tasks, not tours: a realistic goal with the path unstated, then silence while the person works. Five people per distinct group per round finds problems; it measures nothing. The failure it prevents: a script that says "click the expenses tab and see how easy it is", which only ever finds happiness.2324## Workflow251. Ask up to three questions: the research questions, what part of the design each task tests, and who takes part (one group or several).262. Tie each test goal to a research question and a part of the design. A goal with no question behind it is cut.273. Write task scenarios: "you need to [goal]; go ahead", with the starting point stated and no interface words that give the answer. Write success criteria per task before testing: what counts as completed, completed with help, not completed.284. Write the think-aloud ask and the neutral prompts ("what are you thinking?", "what did you expect?"). When they struggle, wait. Talking changes timing, so no time on task is reported from these sessions.295. Write the moderator script: intro ("we are testing the design, not you"), consent check, tasks, post-task questions that stay open, close. Never "do you like it" or "would you use it".306. Build the observer note grid per task: outcome against the criteria, path taken, hesitations, verbatim quotes with timestamp.317. Set the sample and logistics: five per round per distinct group as a rule of thumb for finding problems, not a measurement; session length you set, with gaps between sessions for notes. Pilot once and fix the tasks before the first real session.3233## Output Format34```markdown35# Usability Test Plan36**Design under test:** [name, version] | **Groups:** [groups] | **Sessions:** [n per group, dates] | **Length:** [min]37## Test goals38| Goal | Research question | Part of the design |39|---|---|---|40| [goal] | [RQ] | [screen or flow] |41## Tasks and success criteria42| # | Scenario as read aloud | Start point | Completed | Completed with help | Not completed |43|---|---|---|---|---|---|44| 1 | "You need to [goal]; go ahead" | [screen] | [criteria] | [criteria] | [criteria] |45## Moderator script461. Intro: "We are testing the design, not you." Consent check: [items].472. Think-aloud ask: [words]. Neutral prompts: "What are you thinking?" "What did you expect?"483. Tasks 1 to [n], each followed by [open post-task question]; close with [open question] and thanks.49## Observer note grid50| Task | Outcome | Path | Hesitations | Quote and timestamp |51|---|---|---|---|---|52| [#] | [completed / with help / not completed] | [path] | [where] | "[verbatim]" [mm:ss] |53## Sample and pilot54[n per group, why; what this round can and cannot show; pilot date and changes]55## Decision56[Research lead] signs off the tasks and criteria after the pilot by [date].57```5859## Done When60- Every task has a scenario with the path unstated and success criteria written before testing61- The script contains no "do you like it", "would you use it" or leading prompt62- The observer grid matches the tasks, and a pilot is booked6364## Quality Bar65- Prototypes use invented test data, never a participant's real account on screen66- No time on task from think-aloud sessions; metrics belong to the benchmark67- Five per round is stated as a rule of thumb for finding problems, with its limits68- Claude writes the tasks; real people attempt them, and only what they did becomes a finding6970## Next71Run uxr-usability-findings (Usability Test Findings) to turn the sessions into rated issues.7273## About the makers7475This pack is made by 90n ([90n.ai](https://90n.ai)): AI-native ways of working and tools for Product, Marketing and HR teams, built by ex-McKinsey founders. We map where time and money leak, redesign the work around AI and stay until your people run it themselves, and we run our own company on Claude. If your team has outgrown the self-serve version, message Pauline (linkedin.com/in/paulinebertry).