1. LandOffer
  2. Blog
  3. All articles
  4. Write a Usability Test Script That Does Not Reveal the Answer

Product & Project Delivery

Write a Usability Test Script That Does Not Reveal the Answer

Five fictional workshop-booking task rewrites and a completed stuck-participant record show how to preserve the work goal without supplying an interface route.

10 min readLandOffer team

Write usability-test tasks around a person's goal, necessary context and observable outcome. Keep interface labels, preferred routes and success checks in the moderator's private notes. When someone hesitates, ask about their current thinking without naming the control you want them to find. If you supply a route so the session can continue, record the assistance and separate later observations from independent completion.

This LandOffer.ai guide contains five completed task rewrites and a fictional moderation record. All people, prototype details, utterances and outcomes below are authored teaching examples. No participant was recruited, session recorded or product tested. Primary guidance was checked October 27, 2026; the examples demonstrate script decisions rather than a measured improvement in usability.

A researcher observes a participant exploring a handmade workshop model without pointing toward a route.

Decide what the task should let you observe

A product demonstration shows someone how a feature works. A usability session gives a relevant person something to accomplish and lets you observe their approach. If you teach the route while introducing the task, successful navigation tells you little about whether they could have found it themselves. A polished script can still reveal the route you wanted the participant to discover.

Start with the research decision. “Can a workshop coordinator find which shared equipment is available for tomorrow?” is different from “Can someone use the date picker after we show it to them?” Both may be worth investigating, but they need different starting conditions. Write the research question and the observable result that would answer it before composing participant-facing instructions.

GitLab's public script-writing guidance connects research objectives, realistic goals and agreed completion measures. It warns that interface wording can turn the activity into matching words. Its page was last modified July 16, 2026. The linked internal template was not accessed; this article creates its own examples from the public guidance.

Define a result without requiring your favorite route. If a person can identify tomorrow's available tool through a calendar, a list or another supported path, note the path and whether the result meets the goal. Do not mark an alternative route wrong solely because it bypasses the feature under investigation. If the research specifically concerns a control after discovery, state that narrower scope in the analysis.

Separate the participant card from moderator notes

The participant needs enough information to act: a believable situation, relevant constraints and the outcome they want. The moderator also needs setup details, observation questions, possible valid endings, known prototype limits and a recovery plan. Those are different documents even when stored in one discussion guide. Keep private route notes off the shared screen; an exposed next column can give away the answer before the activity begins.

In the fictional Lantern workshop-booking prototype, researcher Ada studies coordinators who already arrange shared tools. The prototype contains invented equipment and members. No real booking, payment, email or permission change can occur. Observer Leon records behavior while Ada moderates. The participant card describes the work; private notes identify screens and stored state that establish the result.

Ada's opening says: “We are studying this design, not assessing you. Please work as you normally would and say what you are thinking when comfortable. You can pause, skip an activity or stop. This is a prototype with invented information; it cannot make a real booking or contact anyone.” Recording, observers and data use would need the actual study's consent explanation; a friendly introduction does not supply that consent.

The GOV.UK Service Manual, updated October 3, 2017, recommends clear, neutral tasks, a discussion guide and a practice run. Its guidance also distinguishes useful dummy-data sessions from the contextual detail that real data can reveal. Lantern's synthetic setting protects this teaching case's boundaries; it does not prove the scenario is realistic for actual coordinators.

Five completed rewrites for the Lantern study

Ada's first draft names controls before participants have seen the prototype. She revises all five activities below. These are completed editorial rewrites. No comparison study established that the revisions improved task success or removed every source of bias. The right-hand column belongs only to the moderator and note-taker.

Leading draft Revised participant task Private observation and completion note
“Open Availability and use the date filter to find a drill.” “Tomorrow, November 4, you need a cordless drill for a repair session from 10 to 11 in the morning. Find one you could use for that hour.” Start on the workshop home page with no filters applied. Result must identify an available cordless drill for the specified date and full hour; no particular navigation route is required.
“Click Reserve, select Drill B and confirm.” “You have chosen Drill B for tomorrow from 10 to 11. Arrange for it to be held for your session.” Use a separately prepared available-tool state. Confirm that the prototype records Drill B, November 4 and the correct hour; stopping on a review page is not a completed booking.
“Use Manage Booking to edit the time.” “Your repair session now starts an hour later. Change the existing Drill B arrangement to 11 to noon tomorrow.” Seed a confirmed 10–11 booking regardless of the previous activity. Check that the same booking now shows 11–12 and that no second booking was created.
“Choose Add Member and set Viewer access.” “Imani is helping coordinate the session. Let Imani see its equipment arrangements without allowing changes.” Seed an existing member named Imani. Observe access reasoning; only view permission meets the stated goal. No real invitation is sent.
“Click Cancel Booking and check the cancellation toast.” “The repair session will not happen. Make sure the drill is no longer being held for it.” Start from a seeded active booking. Check the stored cancellation state and participant's interpretation; a fleeting message alone is insufficient.

The second task intentionally gives the chosen tool. It investigates making an arrangement after selection, not discovering equipment. Reusing the same start page for both questions would obscure that distinction. Likewise, resetting the third task avoids making one participant's earlier difficulty prevent observation of the time-change flow. The reset must be recorded because the activities no longer represent one uninterrupted journey.

Specific details are not automatically hints. A date, duration, item and permission constraint can define the problem without revealing its solution. By contrast, a capitalized navigation label or an instruction to look in the upper-right corner can supply the very discovery you hoped to observe. Remove route cues without making the goal a riddle.

Nielsen Norman Group's January 12, 2014 guidance makes this distinction explicit: scenarios should support realistic action without describing the interface steps, while familiar established terms may still be necessary. Do not replace ordinary words with elaborate synonyms solely because the interface also uses them. Check whether the participant understands the work goal, rather than testing their ability to decode your prose.

Pilot the instructions and preserve the changes

Ada rehearses the guide with colleague Tomas on November 2 in this authored case. Tomas asks whether “tomorrow” refers to his actual calendar date or the prototype's example date. Ada changes the shared context to “For these activities, imagine today is November 3, 2026,” and leaves November 4 on each relevant task card. This repairs a script ambiguity; it is not evidence about the booking interface.

Tomas also asks whether “held” means a real reservation. Ada adds the explicit no-real-booking boundary to the introduction. During rehearsal, the time-change branch stops on an unfinished prototype screen. Engineer Priya confirms that limitation in the fictional setup notes. Ada prepares a seeded screen for the next activity and records that downstream completion cannot be observed on that branch yet.

She saves guide version 3, the fixed November 3 context, the no-real-booking introduction and the unfinished time-change branch before the planned sessions. A colleague rehearsal can catch missing context and broken setup; it does not replace research with relevant users. If later sessions use materially different wording or a repaired prototype, keep those versions visible rather than pooling all outcomes as if conditions were identical.

Task order also teaches. Someone who has already changed a reservation may know where cancellation lives. Ada keeps the connected order for this round because later management tasks matter, but flags prior exposure when interpreting discovery. If first-time discovery is the primary question, she would need a different design, such as separate task assignments or carefully varied order. Neither five tasks nor one ordering scheme is a universal requirement.

An observer records a pause while a participant inspects an unfinished cardboard mechanism; a separate prepared model waits aside.

A completed record of getting stuck without being led

The following fictional November 3 transcript concerns the third activity, with participant Jules and guide version 3. Its clock measures elapsed time within this invented activity. These are not recordings or observed research results. Leon keeps actions, words, moderator interventions and interpretation separate.

Elapsed time Authored action or utterance Moderator response and evidence consequence
00:00 Ada gives the revised 11-to-noon task from the seeded 10–11 booking. Jules opens the booking summary. No hint or route instruction is supplied. The start state is recorded.
00:38 Jules opens equipment details, returns, then says, “I can see when it is booked, but not how to change it.” Ada waits while Jules continues reading; she does not identify a button or approve a route.
01:02 Jules asks, “Should I cancel it first?” Ada asks, “What would you do if you were working on this on your own?” Jules says they would check the existing arrangement once more. This is a prompt, not proof of unprompted behavior.
01:27 Jules opens the existing arrangement again and says, “This only tells me the details.” Ada asks, “What are you looking for at this point?” Jules replies, “A way to keep the drill but change the hour.” The reply clarifies the goal without establishing the cause of difficulty.
01:54 Jules finds a change action and reaches the documented unfinished prototype screen. Ada identifies the prototype limitation and ends this activity. No successful time change is recorded.
02:10 Ada loads a separate seeded arrangement for the next activity and explains the reset. Later permission observations can continue, but they do not establish recovery or independent completion of the time-change task.

The useful record is narrower than “Jules failed to edit a booking.” Jules understood the desired change, initially explored equipment details, needed conversational prompts and eventually found the change action. The unfinished branch prevents observing whether the new time would be correctly saved. The record cannot assign all difficulty to navigation, terminology or participant knowledge.

Ada's exact wording matters. “Have you tried the three dots?” would reveal location; “Was that confusing?” would offer an interpretation. Even a neutral question interrupts the person's activity. Record it instead of calling everything before an explicit click instruction independent. Record what the participant says they would do separately from what they actually try. Neither hesitation nor a request for help establishes carelessness or lack of skill.

GDS practitioner advice from August 5, 2019 discusses observing quietly, avoiding cues and prompting cautiously. It is practical moderation guidance, not evidence that any prompt is unbiased in every setting. Give people room to think, respond to their actual behavior and avoid turning a pause into an automatic countdown to intervention.

Report the boundary, then decide what needs another session

Before sessions, agree how to stop for withdrawal, distress, unsafe disclosure, a prototype failure or exhausted session time. Do not withhold necessary support merely to preserve a completion measure. If you provide a location hint to reach a later screen, mark the earlier discovery as assisted and explain what later question remains answerable.

For Lantern, Ada's completed disposition is to retain the neutral task wording, repair the unfinished time-change branch and revisit that flow with appropriate users. The script's ambiguity and the prototype's missing behavior are separate fixes. No participant success percentage, time benchmark or usability score is calculated from the teaching transcript.

Review one task by covering the private notes and reading only what a participant will receive. Confirm that it supplies a believable goal, the necessary constraints and a safe stopping boundary without naming the answer. Then rehearse the whole guide, including failure recovery, and preserve the wording actually used. A cleaner script prepares a study; it does not certify the design or replace sessions with relevant users. That gives your next study a clearer question to investigate while keeping assistance and uncertainty visible.

LandOffer.ai publishes this practical workplace guide; it does not conduct or certify your user research. Start with one task card and its private completion notes.

Sources and Further Reading