Human-Eye QA for AI-Built Public Pages

AI can generate a page that passes syntax checks but still looks careless. Human-eye QA is the release gate for public Rocky pages, especially wiki articles, project pages, and client-facing previews.

What scripts catch

• HTTP status.
• Broken images.
• Missing titles and descriptions.
• Duplicate slugs.
• Horizontal overflow.
• Console errors.
• Canonical and sitemap mistakes.

What humans catch

• Headings that wrap into lonely words.
• Buttons with uneven padding.
• Tiny tags or chips.
• Images cropped in a way that looks accidental.
• Repeated hero images that make pages feel cloned.
• Internal scaffolding copy that should never reach readers.

Screenshot set

• Desktop home/index state.
• Desktop article or detail page.
• Tablet article page.
• Mobile index/search state.
• Mobile long article.
• Any image-bearing page.

Pitfalls

• Reviewing only the code diff.
• Taking screenshots before images finish loading.
• Ignoring one small issue that repeats across every card.
• Deploying before checking the mapped domain, not just localhost.

Verification steps

• Create contact sheets from representative screenshots.
• Scan for clipping, spacing, overlap, weak contrast, and awkward headings.
• Fix shared CSS causes.
• Rerun screenshots after the fix.
• Verify the live mapped domain after deployment.