AI website design tells: a 13-point checklist (with how to check each one)
The exact 13 checks behind the Slop Score, heaviest first. Each comes with what counts as a hit and a quick way to check it yourself, no tools needed beyond your browser.
"It looks AI-generated" is a feeling. This page turns it into a list. These are the 13 tells our Slop Score scanner looks for, in order of how many points each one costs. Score your own site by hand with it, or use it as a review checklist before you launch.
How to score yourself
Start at 100. For each tell you find, subtract up to its points. Our scanner scales some deductions by how strongly a tell shows up; for a hand check, subtract the full amount. If your site uses custom typography and no stock Tailwind palette, multiply your total deductions by 0.6. When the fundamentals are yours, the smaller tells matter less. Then read the result: 60+ already looks designed, 35–59 has a few giveaways, and under 35 has lots of room to improve.
Open your site, then open DevTools (F12, or Cmd+Option+I on a Mac). Most checks take seconds.
The checklist
1. Purple/indigo gradient: −20
Hit if: any background gradient uses purple, indigo or violet hues.
Check: in the Console, run [...document.querySelectorAll('*')].filter(e=>getComputedStyle(e).backgroundImage.includes('gradient')).length. If it's above 0, inspect those elements for purple.
Fix: a flat brand colour, with a gradient used for one moment at most.
2. Generic AI copy: −16
Hit if: the page uses words like unlock, supercharge, seamless, elevate, empower, effortless, revolutionize, all-in-one or "in seconds".
Check: Ctrl+F each word.
Fix: say what it does, for whom, in plain words.
3. Default Tailwind palette: −14
Hit if: three or more stock Tailwind colours (indigo-500, slate-900, gray-50…) or heavy purple accents.
Check: inspect a few elements and look for class names like bg-slate-900 or text-indigo-600, or computed colours like rgb(99, 102, 241) (indigo-500).
Fix: five custom hex tokens. Palettes here.
4. Inter / system-font only: −12
Hit if: over 90% of text is set in Inter, system-ui, Geist, Poppins, DM Sans, Manrope or similar defaults.
Check: inspect the h1 and a paragraph and look at the computed font-family. If the heading and body use the same default face, it's a hit.
Fix: pair a characterful display face with a readable text face. 12 pairings.
5. Emoji used as icons: −12
Hit if: two or more emoji stand in for icons in headings, cards or buttons.
Check: just look, especially at feature cards and buttons.
Fix: one consistent icon set, or none.
6. Uniform icon-card feature grid: −12
Hit if: there are rows of 3, 4 or 6 identical cards, each with an icon and a heading.
Check: scroll to "Features".
Fix: one lead feature with a real screenshot, varied sizes, and the rest as a list.
7. Glassmorphism: −10
Hit if: any surface uses backdrop-filter: blur().
Check: [...document.querySelectorAll('*')].filter(e=>getComputedStyle(e).backdropFilter.includes('blur')).length
Fix: solid surfaces with a clear border or shadow.
8. Builder badge left on: −10
Hit if: an "Edit with Lovable", "Made with Bolt" or "Made with v0" badge is visible.
Check: open the live URL in a private window and look at the corners.
Fix: remove it before you share the link.
9. rounded-2xl everything: −8
Hit if: more than 45% of boxed elements (cards, buttons, inputs, images) have a radius of 12px or more.
Check: inspect a card, a button and an input and read border-radius.
Fix: a deliberate radius scale (2–4px works) used consistently.
10. Default Lucide icons: −8
Hit if: there are four or more stroke-2, 24px Lucide-style SVG icons.
Check: document.querySelectorAll('svg.lucide, svg[stroke-width="2"]').length
Fix: fewer icons. When you do use them, pick a distinctive set or custom glyphs.
11. Mobile horizontal overflow: −8
Hit if: the page is wider than a 390px phone viewport.
Check: turn on device mode at 390px wide, then run document.documentElement.scrollWidth > innerWidth.
Fix: find the element that overflows (usually a wide image, a table or a long unbroken string) and constrain it.
12. Centered-everything hero: −6
Hit if: the h1 is text-align: center.
Check: getComputedStyle(document.querySelector('h1')).textAlign
Fix: left-align the hero and give it an asymmetric layout.
13. Low type-scale contrast: −6
Hit if: the h1 is less than 2.6x the body font size.
Check: parseFloat(getComputedStyle(document.querySelector('h1')).fontSize)/parseFloat(getComputedStyle(document.body).fontSize)
Fix: a dramatic scale, with display text at least 3.5x the body size.
Which tells are most common?
Across the 30 sites on our public Slop Leaderboard (29 Lovable, Bolt and v0 showcase apps, plus linear.app as a reference), the most common tells as of 2026-10-17 are:
Share of the 30 scanned sites showing each tell. Average score: 46. Source: unslop-me.vercel.app/leaderboard.
Notice that the purple gradient doesn't make that top-eight list, even though it costs the most points. It's rarer than people think, but when it shows up it drags the score down hard. The tells that are almost everywhere are the quiet ones: icons, the badge, the card grid and the font.
Turning the checklist into fixes
Fix in this order for the best score-per-minute: the badge (one setting), emoji, fonts, colours and gradient, glass and radius, then the hero, the card grid and the copy. If you're in Cursor or Claude Code, put the checklist in a rules file so the model stops reintroducing defaults (see Cursor rules for UI design). You can also connect our free MCP server, which exposes this exact list as unslop_tells(). Details are on the agents page.
For a full launch QA beyond the visual tells (meta tags, favicon, trust details), use the 30-point pre-launch checklist.
FAQ
How can you tell if a website was made with AI?
Look for stacked defaults: a purple or indigo gradient, Inter as the only font, three identical icon cards, frosted glass panels, everything heavily rounded, a centered hero, words like 'seamless' and 'supercharge', and a 'Made with' badge. One alone means little; five or six together is a strong signal.
Is there a tool that checks for AI design tells automatically?
Yes. Unslop Me's free scanner runs these 13 checks on any public URL and returns a 0–100 Slop Score, the evidence for each tell and fix prompts for Lovable, Bolt, v0 and Cursor.
Which AI design tell costs the most?
The purple/indigo gradient, at up to 20 points, followed by generic AI copy at 16 and the default Tailwind palette at 14.
What score should I aim for?
60 or above reads as designed. Under 35 means several defaults are stacking up.