Do a Heuristic Evaluation
Run Nielsen's 10 usability heuristics against any interface using Claude as your evaluation partner. Get a prioritised findings report you can act on.
The situation
An interface needs a usability review before it ships. Your gut says something is wrong. You need specific findings with severity ratings, not a list of impressions, and something a PM can actually prioritise.
You'll produce a full heuristic report against all 10 Nielsen heuristics, grouped by severity, in about 25 minutes.
What you walk away with
A full evaluation against all 10 Nielsen usability heuristics with a specific violation example for each
Every finding rated: Critical, Major, Minor, or Cosmetic
A prioritised report grouped by severity tier, ready for sprint planning or a stakeholder review
The difference one prompt makes
Don't
Review this design for usability problems.
Do this
Evaluate this 4-field payment form against Nielsen's Error Prevention heuristic only. List each place where the user could make an irreversible mistake without a confirmation or undo option. Format: [screen element] + [risk] + [severity: Critical / Major / Minor].
Describe the interface you are evaluating
Paste the template below and fill in your interface. Claude will confirm it has enough context and ask whether to run all 10 heuristics at once or one by one. Answer "all 10 at once" — that is the prompt for step 2.
# Interface context template:
"I am running a heuristic evaluation using Nielsen's 10 usability heuristics.
Interface: [What it is: app, website, feature]
Primary user: [Who uses it, what they know, what device they use]
Core task: [The main thing a user is trying to accomplish]
Screens to evaluate: [List the screens or flows]
Description: [Describe the screens, layout, labels, interactions]"Run the evaluation
Paste the evaluation prompt. Claude works through all 10 heuristics and returns a verdict for each — comply, partial, or violation — with a specific example and a severity rating. At the end you will have a list of violations with counts: something like "1 Critical, 3 Major, 3 Minor." That list is what step 3 turns into a report.
# Prompt to run the evaluation:
"Evaluate this interface against all 10 of Nielsen's usability heuristics.
For each heuristic, state: comply, partially comply, or violation.
For any violation: describe it specifically, give an example from the interface,
and rate severity (Critical / Major / Minor / Cosmetic)."Generate the prioritised findings report
Paste the consolidation prompt. Claude groups all violations by severity — Critical down to Minor — and adds a one-line fix for each. What comes back is paste-ready: you can drop it directly into a Jira ticket, share it with your PM, or bring it to sprint planning.
# Prompt to generate the report:
"Consolidate all findings into a prioritised report.
Format: severity tier, heuristic violated, specific issue, one-sentence fix.
Group by severity."Common mistakes
Four failure modes in heuristic evaluations.
- Describing the interface too vaguely. Claude evaluates what you describe. "A form with some fields" produces generic heuristic observations. "A 4-field payment form with amount, phone, PIN, and a confirm button on Android mobile" produces specific, actionable findings. Put the time in on the description.
- Conflating Minor with Cosmetic. Minor friction slows users down or causes recoverable errors. Cosmetic issues are visual preferences that have no functional impact. Labelling a Minor issue as Cosmetic means it gets deprioritised and never fixed.
- Skipping heuristics where the interface "obviously" complies. The most dangerous assumption in a heuristic evaluation is that a heuristic does not apply. Check all 10, including the ones that look fine. Violations in "passing" heuristics are easy to miss.
- Presenting the evaluation without the severity grouping. A list of 10 raw findings is hard to act on. The report step groups by Critical, Major, Minor so stakeholders know what to fix before the next release versus what to consider next quarter.
What's next?
Prepare your Figma for AI handoffNew guides, when they ship
One email, roughly weekly. CLAUDE.md templates, workflows I actually use, and the cut-for-length stuff that does not make the public guides. One-click unsubscribe.
Or follow on Substack