Blog
12 posts on what actually separates a hand-built page from a generated one — the reasoning behind the scanner's signals, written out at length instead of compressed into a weight and a one-line reason.
How to make an AI-generated website look hand-builtWe took the page from our first scan run that came back highest, fixed it one change at a time, and rescanned after every fix. 75 to 4, with the exact reason for each drop.
Does UI/UX Pro Max actually improve AI-generated design? Our before-and-after testWe gave the same model the same brief twice, once cold and once with an open-source design skill loaded, then scored both pages with this site's own scanner. 97 to 11. Here is what moved, and the four things the skill did not fix.
The best AI coding tools in 2026Claude Code, Codex, Cursor, Copilot, Antigravity and the prompt-to-app builders, compared on price, control, how hard each one is to actually drive, and the metric nobody publishes: how much of the tool ends up shipping to your visitors.
Every AI builder leaves a fingerprint. Here's where each one hides.Lovable, v0, Bolt, Replit and Base44 all sign their work, none of them on purpose. What each one leaves in the markup, which artefacts survive a migration, and why an asset host is not a confession.
Five signals that fire on Stripe, Linear and GitHubThe most frequently triggered detectors in the set don't fire on AI builders. They fire on some of the most carefully built pages on the internet, and that tells you how to read any score, including your own.
What the hand-built sites actually doEleven signals on this site subtract from the score instead of adding to it. Here's which ones the 28 worked examples actually earn, which ones almost nobody does, and the one that literally no site in the set manages.
The credentials sitting in your page sourceThirty points and a hard score floor, because it's the one finding here that costs real money. What a single HTML fetch can prove about leaked keys, why Firebase's API key is deliberately not flagged, and what the Lovable CVE actually taught everyone.
The site that invented the look fails its own vibe checktailwindcss.com scores 75 out of 100 here, the second-highest in the set. Working out why it deserves that and also doesn't is the best available lesson in what this kind of scanner can and cannot know.
The nine sections every landing page has nowHero, logo bar, three feature cards, stats, testimonials, pricing, FAQ, CTA, footer. Where that running order came from, why generators reproduce it exactly, and how to tell a structure from a habit.
Leftovers: the scaffolding nobody went back to clean upDead links, placeholder images, a missing favicon. Individually trivial, collectively the clearest signal that a page shipped the moment it rendered, and the cheapest category on the whole list to clear.
Why every AI-built site has the same purple gradientIndigo to violet, top-left to bottom-right, behind a headline nobody chose the color for. Where the palette comes from, why generators keep reaching for it, and what it actually proves.
Why AI writing can't stop reaching for emojiReward models score decorated answers higher than plain ones, so the models learned to decorate. The research behind the habit, the April 2025 case where everyone watched it happen live, and the three detectors here that catch it.