These four prompts cover the parts of a cognitive walkthrough where AI saves the most time without giving up rigor: simulating a first-time user against screenshots with the four prescribed questions, generating the action sequence and persona from a prototype, stress-testing the design against multiple personas in parallel, and clustering the failure points into a prioritized fix list. Each prompt is meant to be filled in with your own product, persona, and screenshots, then run in Claude, ChatGPT, Gemini, or any multimodal LLM with a long enough context window. The prompts assume a human team in the loop — the model produces a first pass, the workshop confirms or overrides it.
Prompt 1: Run a multimodal cognitive walkthrough on screenshots
You are an experienced UX researcher running a Cognitive Walkthrough on a [product type — e.g., healthcare check-in tablet, B2B onboarding flow, AI feature in a productivity app] for [persona — e.g., a brand-new patient in their fifties who has never used this app before].
Persona details (knowledge, motivation, prior experience):
[describe the persona in 3-5 sentences — what they know about the domain, what they know about similar products, what they want to accomplish, what they expect to find]
Task scenario:
[one sentence — what the user is trying to do]
Starting state: [what is on screen at the start]
Goal state: [what success looks like]
Action sequence (the correct path the user would need to take):
1. [action 1]
2. [action 2]
...
I will paste/attach screenshots of each step in order.
For every action in the sequence, answer the four Cognitive Walkthrough questions from the persona's point of view:
1. Will the user try to achieve this result? (Yes / No / Maybe + one-sentence rationale)
2. Will the user notice that the correct action is available? (Yes / No / Maybe + one-sentence rationale)
3. Will the user associate the correct action with the result they want? (Yes / No / Maybe + one-sentence rationale)
4. After the action is performed, will the user see that progress is being made? (Yes / No / Maybe + one-sentence rationale)
End each action with a verdict (Pass / Hesitate / Fail) and tag which of the four questions broke down.
Stay in the persona's voice. Do not slip into "any designer knows" or "I would just" — speak as the imagined first-time user.Prompt 2: Generate the action sequence and persona from a prototype
You are helping me prepare a Cognitive Walkthrough. I will paste a description (or screenshots) of a prototype below. Produce a draft persona, task scenario, and step-by-step action sequence I can use as the starting document for a workshop.
Prototype description / screenshots:
[paste link, description, or attach screenshots]
Product context: [what is this product, who is the intended user]
Critical task to evaluate: [the one task you want to walk through]
Please produce:
1. A one-page persona for a brand-new user of this product. Include: prior knowledge of the domain, prior experience with similar products, motivation for being on this screen, context (location, device, emotional state), and 2-3 expectations they would bring from previous experiences.
2. A task scenario: one-sentence description, starting state, goal state.
3. An action sequence — the correct path the user would need to take, one row per click or interaction, in order. Be specific about the element they would interact with on each step.
4. A list of 3-5 alternative paths the user might take by mistake, for the team to discuss during the walkthrough.
5. A list of any prior knowledge or assumptions the design seems to depend on, that a first-time user might not have.
Flag any step where you are uncertain about the correct action or where the persona's likely behavior is hard to predict.Prompt 3: Stress-test the design against multiple personas in parallel
I need to run a Cognitive Walkthrough on the same task scenario against 5 different personas in parallel, to see which failure points are persona-specific and which are universal.
Task scenario:
[one sentence]
Starting state: [what is on screen]
Goal state: [success state]
Action sequence:
[paste the action list]
Personas to run in parallel:
1. Novice: [3-sentence description]
2. Returning user: [3-sentence description]
3. Low-literacy: [3-sentence description]
4. Non-native speaker (English as second language): [3-sentence description]
5. Accessibility user (e.g., screen reader): [3-sentence description]
Screenshots: [paste]
For each persona, run the four Cognitive Walkthrough questions on every action, and produce one row per persona-action pair with the verdict and rationale.
At the end, build a matrix:
- Rows: actions
- Columns: personas
- Cells: Pass / Hesitate / Fail
Highlight:
1. Failure points that are universal across all personas (the most urgent fixes)
2. Failure points that are specific to one or two personas (smaller but persona-critical fixes)
3. Any persona that fails more than 50% of the steps (the design may not be viable for that group at all)
4. Any step where the personas split in surprising ways and the team should investigate furtherPrompt 4: Cluster failure points and draft the prioritized fix list
I have a Cognitive Walkthrough log of [N] failure points across [M] actions for [task] in [product]. The team has roughly [engineering and design capacity] for the next sprint.
Walkthrough log:
[paste each failure point with: action ID, screen, which of the four questions broke down (goal formation / action visibility / action labeling / feedback), severity (fail / hesitate), and a one-paragraph rationale]
Please:
1. Cluster the failure points by which of the four questions broke down. Report the count per cluster and identify the dominant failure type for this design.
2. For each cluster, propose a category of fix (onboarding, copy, layout, affordance, system message, etc.) and explain why it addresses that specific question type.
3. Score each individual failure point on three dimensions (1-5 each): severity (does the user completely fail or just hesitate), frequency (how often a real user would hit it), and effort to fix (1 = small, 5 = large).
4. Compute a priority score (severity + frequency – effort) and sort the failure points from highest to lowest priority.
5. Recommend the top 5-10 failure points to fix in the next sprint, with a concrete proposed fix and a rough effort estimate per item.
6. Identify any cluster of 3+ failure points on the same screen — these may deserve a redesign of the screen rather than incremental patches.
7. Draft a 5-sentence executive summary the lead can use as the opening of the readout brief, anchored on the dominant failure type and the top recommended fix.