# The AI Design Working Toolkit

UX Design Coach · October 11, 2026

Adapt these prompts to your project. AI output is not participant evidence. Keep confidential data in approved systems.

## Idea validation ledger

Turn a promising idea into claims you can investigate.

### Audience and problem
Who experiences the problem, in what context, and what do they do today?

Your notes:


### Riskiest assumption
What would make this idea unworkable if it were false?

Your notes:


### Current evidence
List sources, dates and contradictory observations. Separate facts from assumptions.

Your notes:


### Test and decision rule
What will you do, with whom, for how long? What result would change your mind?

Your notes:


### Result and next decision
What happened? What can the evidence support, and what remains uncertain?

Your notes:


## Research evidence log

Keep sources, observations and interpretations separate.

### Question and method
Which decision does the study inform? Who participated, and what are the limits?

Your notes:


### Evidence rows
Source ID | Observation or exact excerpt | Context | Interpretation | Contradiction

Your notes:


### Candidate finding
Which sources support it? Which sources challenge it? What is your confidence and why?

Your notes:


### Human review
Who checked the original material? What changed during review?

Your notes:


### Decision
What will change, who owns it and what must be tested next?

Your notes:


## Design quality scorecard

Review eight dimensions and keep blockers outside the average.

### Artifact and task
Version, date, audience, primary task, intended fidelity and reviewers.

Your notes:


### Eight dimensions
Task clarity; content integrity; interaction completeness; visual hierarchy; brand specificity; accessibility; responsive behavior; maintainability.

Your notes:


### Ratings and evidence
For each: 0 blocking/missing, 1 major revision, 2 minor revision, 3 meets brief, or not evaluated. Add the observation and source.

Your notes:


### Release blockers
List serious defects independently of scores. Assign an owner and evidence needed to resolve each.

Your notes:


### Comparison conditions
Inputs, model/tool, settings, revision budget, review effort and unresolved issues. This rubric is not a certification or scientific benchmark.

Your notes:


## User testing plan

Prepare a study that helps a real decision.

### Decision and question
What needs to be learned, and which method fits?

Your notes:


### Participants and data
Relevant criteria, access needs, recruitment, consent, recording, approved processing and retention.

Your notes:


### Neutral tasks
Write realistic scenarios without naming the controls. Define success, assistance and failure.

Your notes:


### Pilot and observation
Check prototype limits. Log session ID, action, context, quotation and interpretation separately.

Your notes:


### Findings and limits
Attach sources, include contradictions, state sample limits and choose a next action.

Your notes:


## Art direction brief

Create a specific visual point of view before generating.

### Purpose and audience
What should the person understand or do? What content must remain exact?

Your notes:


### References and permissions
Reference | Quality to study | What not to copy | Rights or permission status.

Your notes:


### Visual rules
Type relationships, color roles, spacing, image treatment, composition and motion.

Your notes:


### Distinct directions
Propose three different concepts, not three color variations. Record why you chose one.

Your notes:


### Review
Check content, cropping, coherence, responsive behavior, accessibility and editability.

Your notes:


## Automation and AI team contract

Define a bounded workflow the team can understand and stop.

### Trigger and owner
What starts the job, and who is accountable?

Your notes:


### Inputs and access
Required fields, approved sources, allowed tools, permissions and data limits.

Your notes:


### Output and acceptance
Artifact format, validation rules, reviewer and quality criteria.

Your notes:


### Failure and recovery
Missing inputs, invalid output, retries, duplicate prevention, error reporting and stop control.

Your notes:


### Role handoff
Assistant role | Input | Output | Reviewer | Stop condition. Model agreement is not independent user evidence.

Your notes:


### Evaluation
Baseline time, processing cost, review time, accepted output, errors, rework and next review date.

Your notes:


## Client offer canvas

Turn expertise into a clear, credible service.

### Buyer and problem
Who can use this service, and what observable problem triggers the need?

Your notes:


### Deliverable and scope
What is included, excluded, required from the client, and reviewed together?

Your notes:


### Proof
Sample artifact or permitted case study. Label fictional examples and avoid unsupported results.

Your notes:


### AI and method
Where does AI assist? Who reviews? Are real participants involved? How is data handled?

Your notes:


### Invitation and measurement
Fact-checked reason to connect, next step, qualified conversations, proposals and wins.

Your notes: