Case study
We let an AI agent fix a demo page blind.
We built a throwaway SaaS page (loopline-demo.vercel.app), audited it cold, and handed the .md straight to an AI coding agent — no context beyond "fix these." First score: 43. One pass later: 79, with mobile up 55 points. Both reports below are raw and unedited.
Raw, unedited reports: before (43) · after (79) · the page itself: loopline-demo.vercel.app
What it caught
16 UX findings, 11 SEO, 7 performance. The gist, in its own words:
"Loopline has a polished, minimal aesthetic but fails the fundamentals: the headline is a generic slogan, the CTAs… don't stand out… the conversion form asks for far too much, and the missing viewport meta means mobile is likely broken."
What the agent shipped
“A better way to run your team” says nothing. Swapped for a line that names the actual product.
Two identical buttons became one bold CTA and one quiet text link.
Added the missing viewport meta — the single biggest win. Mobile 30 → 85.
Six fields down to name and work email. The rest was just friction.
Made-up “company” wordmarks out, named testimonials with faces in.
What moved
| Score | Before | After | Δ |
|---|---|---|---|
| Overall | 43 | 79 | +36 |
| Trust | 40 | 78 | +38 |
| Mobile | 30 | 85 | +55 |
| Call to action | 38 | 74 | +36 |
| First impression | 42 | 82 | +40 |
| Hierarchy | 55 | 80 | +25 |
| Readability | 58 | 68 | +10 |
| SEO | 48 | 88 | +40 |
What it still isn't happy about
Readability got the smallest bump (58 → 68). The re-run still flagged low contrast, friction reducers stranded at the bottom instead of by the CTA, and a header button fighting the hero. Same rubric for the fix as for the original — no grading on a curve.