When we shipped Kiro AI auto-review, the first question every designer asked was reasonable: "What does it actually check, and is it just pattern-matching against trends?"
Fair question. Here's the full list of what auto-review evaluates on a design or web canvas, what it does well, and what it doesn't.
What auto-review checks.
1. UX heuristics.
Nielsen's ten heuristics, applied to the canvas: visibility of system status, match between system and real world, user control and freedom, consistency, error prevention, recognition over recall, flexibility, aesthetic minimalism, error recovery, and help/documentation. Auto-review surfaces the top 3 most likely issues by severity.
2. Contrast ratios.
Every text element vs its background, calculated against WCAG 2.1 AA (4.5:1 normal, 3.0:1 large) and AAA (7.0:1 normal, 4.5:1 large). Fails are flagged with the actual ratio, the rule violated, and the element selector.
3. Touch targets and clickable area.
On mobile viewports: any clickable element under 44x44 CSS pixels gets flagged. Apple's Human Interface Guidelines minimum, Material's accessibility recommendation. This catches a lot of small icons-as-buttons that are technically tappable but practically miss-prone.
4. Copy and microcopy.
Reading level (Flesch-Kincaid), passive voice density, jargon flags, and CTA verb strength. "Submit" and "Click here" get flagged for being weak. "Sign up" or "Start free" don't.
5. Visual hierarchy.
Heading vs body size ratios, contrast between primary and secondary text, button prominence vs surrounding elements. Flags pages where the most important action isn't visually the most important thing.
6. Brand consistency (if you've uploaded a brand kit).
Color matches against your brand palette, font matches against your typeface stack, logo placement against your spec. Off-brand red? Flagged. Wrong font weight in a heading? Flagged.
What the score actually means.
Auto-review produces a score from 0 to 10. Scoring isn't a vibe — it's a weighted sum of severity-tagged issues. A 7/10 means: a few moderate issues, no critical ones, the work is shippable but improvable. A 4/10 means: there are critical issues that should not ship as-is. A 9/10 means: minor polish remains; this is essentially ready.
The score is calibrated against a reference set of canvases that designers we trust have rated themselves. It's not perfect. It is useful.
What auto-review does NOT check.
Whether the work is on-brief. Whether the concept is good. Whether the client will like it. Whether the project is the right project. Whether you should have taken a different angle. None of these are pattern-matching problems — they're judgment calls Kiro AI doesn't try to make.
Auto-review is a senior designer doing a quick QA pass on craft. It's not a creative director. It catches the things that are objectively suboptimal and stays out of the things that are subjectively yours to decide.
Free plan includes 3 reviews/month.
Run auto-review on your work