"AI slop" is the new term for visual elements that scream this was generated by a tool, not designed by a person. Sparkle emojis on every CTA. Gradient text. Soft-glow blue blobs behind every hero. Identical 4-up stat cards that say "256K Names" and "21 Countries". The specific tells are everywhere now because every no-code site builder, every AI design tool, and every "make me a landing page" prompt outputs roughly the same handful of moves.
This started mattering for two reasons. First, Google has begun quietly down-weighting AI-generated content and imagery in search rankings — image search especially. Second, conversion drops measurably when the design looks like a Canva template instead of like a real product. Visitors don't trust it.
Last weekend I went through four of my own sites — babyna.me, DinnerDash, FactFacts.com, and the marketing landing for KidCart — and removed the slop. Here's the punch list of what I found and fixed, in case it helps you spot the same things on yours.
1. The sparkle emoji on every CTA
Hero badges that said ✨ 256,000+ Names · Free Forever. Buttons with ✨ Generate Names. The ✨ emoji is the single most reliable AI-output tell of the last two years — it's what ChatGPT defaults to when asked for "make this feel premium." Removing it is the cheapest visible win.
Fix: Drop the emoji. The text alone is fine. If you really want a visual mark, use a small inline SVG icon that inherits currentColor.
2. Soft-glow radial gradient hero blob
Every AI-generated landing page has a 600–800px circular gradient blob centered behind the H1, fading to transparent. It's a SaaS landing-page cliché going back to 2020. CSS for the offending block usually looks something like:
.hero::before{position:absolute;width:800px;height:800px;border-radius:50%;background:radial-gradient(circle,rgba(91,140,255,.06) 0%,transparent 70%)}
Fix: Delete the entire ::before rule. The page reads cleaner immediately. If your hero feels empty without atmosphere, that's a sign it needs better content, not better lighting.
3. Identical 4-up stat cards
Big number, small label, repeated four times. 256,000+ Names. 21 Countries. 145 Years of Data. 50 US States. This is the SaaS "hero metric template" — every reviewer recognizes it.
Fix: Inline the stats as a sentence. "21 countries · 145 years of birth data · 50 US states · real government records" reads better and takes up less vertical space.
4. Flag emojis as country selectors
21 country tiles, each with a flag emoji and a country name. Looked fine in Chrome on macOS — until I checked Windows. Windows 10 displays flag emojis as the regional indicator letter pair, so 🇺🇸 became "US" in a colored box, 🇬🇧 became "GB", and so on. The selector looked broken on half my visitors' machines.
Fix: Build a tiny inline SVG flag sprite. ~6KB raw, ~1.7KB gzipped, served once and cached site-wide via <use href="/assets/flags.svg#f-us"/>. Renders identically on every device and looks intentional rather than emoji-default.
5. Gradient text on the logo
The wordmark used background: linear-gradient(135deg, #5B8CFF, #FF6FB1); -webkit-background-clip: text;. It looked fine on screenshots. It looks like a 2018 SaaS template anywhere else.
Fix: Solid color or a hard two-tone (e.g., BabyNa in indigo, .me in pink as separate spans). Cleaner, more designed, and easier to hit accessibility contrast.
6. Pink-vs-blue gendered rank pills
The "Top Boy Names" list used blue rank pills, "Top Girl Names" used pink. The data is gendered, sure — but the visual hammers it. Drop the pink/blue bind, use a single neutral pill with the rank number. The same data still reads.
7. The "→" trailing arrow on every CTA link
Every View all link had a →. Every section had a Read more →. By the fourth one in the page, the arrow becomes meaningless. Use it sparingly — once per page if at all — and let the link affordance carry it the rest of the time.
8. The default dark-blue palette
Default dark mode for almost every AI-generated site is #0b1020 background plus an electric blue accent like #5B8CFF. It's the "observability tool, but for [your category]" reflex. For a baby-name site? It's wrong. The honest scene — a parent on a couch at 9pm scrolling for ideas with their partner — doesn't force navy blue. It forces something warmer.
Fix: Switch the dark palette to a warm brown #1A1614 with a cream #F5F0E8 text. Replace the electric-blue accent with a slate indigo and a dusty pink. Suddenly the whole site reads as designed, not generated.
9. Polaroid "fan of three" preview cards
Three cards stacked at slight angles — rotate(-4deg), scale(1.04), rotate(4deg). The stagger was novel in 2021. By 2024 every AI design template defaulted to it. If you're showing samples, show one good sample at full size, not three at half-size pretending to be a fan of polaroids.
10. The robot 🤖 emoji on the AI search page
The AI Search hero said 🤖 AI Baby Name Search & Generator. Robot emoji on the page about AI search is the visual equivalent of writing "AI" in the title four times. Drop it. The headline carries the meaning.
11. AI-generated article hero images of babies
Eighteen articles each had a 240×120 AI-generated header image — sleeping babies in moonlit cradles, plates of italian food rendered by Stable Diffusion, baby clothes floating in golden bokeh. They looked fine in isolation. Side-by-side as a grid, they looked like a stock-image bin.
Fix: Replace AI-rendered header art with actual data viz built from the article's data. A line chart of the name "Mary" rising and falling 1880–2024. A frequency bar chart of Spanish compound prefixes. Real charts beat stock-AI imagery on three dimensions: they're unique to your site (Google rewards), they reinforce the article's argument, and they look designed rather than generated.
12. Floating animated background "orbs"
The site had a JavaScript routine spawning a translucent pastel circle every 3 seconds, drifting upward over 12–32 seconds, fading in and out. The pastels matched the brand palette. It looked like atmosphere. It also looked like every other "AI tool" landing page, plus it spawned ~1,200 DOM nodes per hour of open tab.
I left this one in for now — the customer liked it. But if I see it on a third-party project I'm reviewing, it's the first thing I cut.
What I'd do if I were starting fresh today
I'd write a one-page personal style guide before the first line of CSS. Pick one display font that has actual character (Playfair Display, Fraunces, IBM Plex Serif — anything other than another geometric sans). Pick a warm or cool tonal palette and commit to it. Say no to the entire emoji-as-icon pattern from day one.
The thing about AI design tools is they don't make ugly sites. They make average sites. Average looks the same everywhere. The win is in the small specific decisions that AI tools default away from.
The four sites I cleaned up still aren't perfect. But they look like four different products built by a person now, not one product made four times by a template.
If you want to do this yourself: open your homepage, scroll to the hero, and ask "would a designer at Linear or Stripe ship this?" Be honest. Then start cutting.