# Truvit — Designer study protocol

Prepared 21 September 2026. **Study not yet conducted.** This is a plan and an empty recording structure, not evidence of participant performance.

## Purpose

Find whether designers understand the decisions they own, can make a meaningful revision, and can identify which files and approvals are current. Evaluate comprehension separately from successfully clicking through the flow.

Recruit 3–5 people who have made campaign or visual communication work: practicing designers, design interns, or advanced design students. Record relevant experience and previous AI-tool use with anonymous participant IDs. This is exploratory research, not a representative sample or a benchmark of productivity.

## Setup

- Session length: approximately 30–40 minutes; remote screen sharing or an in-person session.
- Use the current private working prototype with the owner present. Do not share API credentials. Use supplied example material, not confidential client files.
- Prepare a fresh project for each session. Retain its project ID and code/deployment version in the private session record.
- Keep the original participant studies and all agent-operated technical tests separate from this study.
- Check the connection before the session. Record model latency separately from active task time. Stop at the prototype's 16-call pass limit; do not silently reset the limit or hide an error.
- Have the public walkthrough and exported files available if connectivity prevents execution. Label any session that uses them as a walkthrough review, not a completed live workflow test.

## Opening script

“We are testing this prototype, not you. Please describe what you expect and what you are trying to do. You may stop or skip a question at any time. I will avoid explaining the interface until you have had a chance to interpret it. May I take anonymous notes? May I record the screen for internal analysis? Those are separate choices. Nothing will be published as an attributed quote without your permission.”

Ask whether the participant has created campaign visuals and which tasks they normally perform. Record their words as notes; mark a quotation as verbatim only if actually captured and approved for that use.

## Scenario

“You are preparing a visual concept for Moreno Bath, the example brand in this prototype. The supplied photographs are design references; they are not evidence for new product specifications. Your client wants an expressive first-apartment campaign that invites viewers to explore the collection. Do not invent dimensions, materials, prices or performance claims. You need a master plus 4:5 and 9:16 drafts.”

Allow a participant to choose a specific visual voice. Do not require them to reproduce the existing cobalt or plum examples.

## Tasks and evidence

| Task | Neutral prompt | Observe | Completion evidence |
| --- | --- | --- | --- |
| 1. Understand the starting point | “Before you continue, tell me what the tool knows, what it assumes and what decision you need to make.” | Whether they distinguish source statements from assumptions; meaning of Moreno; requests for explanation | Their explanation and the exact screen used; confirming a checkpoint alone is insufficient |
| 2. Choose a direction | “Choose a direction you would develop for this brief. Tell me what you would keep or change.” | Comparison behavior, use of tradeoffs, expectation about what selection will generate | Chosen direction with their reason; record disagreement with the AI |
| 3. Change the master | “The client likes the direction but wants a shorter, more personal headline. Make that change while keeping the visual voice.” | Discoverability of direct editing; what they expect to stay; whether new review is understood | New file, changed copy, retained art direction, participant explanation of approval state |
| 4. Revise one format | After master acceptance and adaptation: “The 9:16 draft needs a stronger hierarchy. Change only that format. What should happen to the other files?” | Target selection, feedback specificity, whether the master and 4:5 are expected to remain | New 9:16 file or an accurately documented error; compare retained file IDs without exposing credentials |
| 5. Review and export | “Decide whether this package is ready to share as a draft. Explain any issue you would resolve first, then export if you consider it ready.” | Whether generated review is challenged, meaning of approval, file discoverability | Actual ZIP opened, expected files present, participant judgment recorded; declining approval can be an appropriate outcome |

Do not prompt a participant to accept a creative they consider unsuitable merely to achieve task completion. If they remain stuck for two minutes, ask “What would you expect to happen?” before providing help. Record the help and the task as assisted. Use a technical fixture only after separating it from the participant's own decision.

## Measures

Record per task: outcome (independent / assisted / not completed / declined appropriately / technical failure), observed errors, help given, active task time, system-wait time and the participant's explanation. Missing measures remain “not recorded”; never estimate them afterwards.

Ask for perceived ease on a 1–7 scale after each task, with 1 = very difficult and 7 = very easy. Treat it as self-report. With a small exploratory sample, show individual scores or counts with denominators, not a general population claim.

To inspect creative differentiation, show actual exports from two contrasting briefs at the end, without labeling one “better.” Ask what changed, which brief each seems to serve, and what remains too similar. Do not infer advertising effectiveness from these preferences.

## Closing questions

1. At which moment did you feel you were directing the work? Where did you feel constrained?
2. Which AI statement would you double-check, and how?
3. What do you believe is included in an approval? What happens if the headline changes later?
4. What would keep you from using this for an early client presentation?
5. What would you remove or simplify?

## Decision rules

- Any reproducible path exporting a changed file under an old approval is a correctness defect; fix before demonstrating that path again.
- Repeated inability to explain a checkpoint triggers a wording or interaction revision. A single critical misunderstanding is still worth investigating; counts are not the only priority signal.
- If participants can finish but cannot explain what changed, do not label the flow validated.
- A style preference is a design input, not proof of audience fit. No time-saving claim without a separately designed comparable baseline study.
- Retest the specific changed interaction with fresh or returning participants and identify which group was used.

## Empty session record — duplicate per participant

- Participant ID / date / study version:
- Relevant experience / AI familiarity:
- Consent for anonymous notes / screen recording / publication of quotes:
- Device and viewport / project ID / app version:
- Task outcomes and timings:
- Observed actions (facts):
- Participant explanation (paraphrase or approved verbatim quote):
- Facilitator assistance:
- Technical failures and model wait time:
- Interpretation (separate from observation):
- Proposed change and rationale:
- Retest status:

## Reporting template

“We tested [version] with [number and relevant backgrounds] on [dates]. [Count/denominator] completed [specific task] independently. We observed [behavior] and interpreted it as [limited inference]. We changed [specific interaction]. [Retest status].”

Do not fill in the brackets until real sessions provide the evidence.