Taste Engineering · 2 / 12

Where the AI Look Comes From

Ask AI to whip up a page and nine times out of ten you get the purple-gradient, frosted-glass, centered-headline trio. That sameness has a mathematical reason—once you see it, you know how to dodge it.

AI-look huntNoise checklistCarnival effectAverage trapSpecificity knob
Try it · Catch the "AI look" first

Below is a typical AI-generated landing page. It hides five high-frequency AI-look tells—tap them one by one. Each hit explains why that tell became AI's default move.

Tap the five AI-look tells on the page 0 / 5
Tap spots on the fake UI that scream "AI made this"
neuraflow-ai.example.com
NeuraFlowProductSolutionsPricingLog in
1 Purple gradient NEW · AI Workflow 2.0 is live
2 Centered headline
Unlock your infinite potential
One-stop intelligent platform for better work
Trusted by 10,000+ teams worldwide
3 Frosted glass
99.9%
Uptime
10k+
Teams
4.9
Rating
5 Gradient button Start for free
14-day free trial · No credit card
4 Three icon cards
Smart summaryPull key points from long docs
Data insightsReports from one sentence
Seamless collabConnects 200+ tools you use
Purple-gradient hero. After 2020, SaaS sites made purple gradients the industry uniform—training data is packed with them. You didn't specify a background, so the model lands where samples are densest.
Centered headline plus two gray lines. "Big headline + subcopy + centered" is the safest template layout in any industry. For the model, the layout that never fails shows up most—so it becomes the default.
Frosted-glass cards. In the glassmorphism years, tutorials and asset sites flooded the web with samples labeled "premium." When the model wants "refined," its first move is a frosted layer.
Three icon cards. "Three benefits, three icons" is the standard landing-page tutorial block—almost every template ships one. The model has seen it so often that feature sections snap into three columns.
Gradient primary button. In component libraries and asset sites, "important buttons" often go gradient to grab attention. The model bound the two: hit a CTA, slap on a gradient.
You got all five. Each piece alone looks decent; the problem is this combo shows up on thousands of pages. The AI look is the taste of high-probability training data: wherever you gave no instruction, it fills in the most common answer.
Try it · Visual-noise checklist

The AI look had an older academic name. In About Face 4 Chapter 17, Cooper calls this stuff visual noise: extra visual elements that yank attention off what actually carries information. He lists seven forms. This "team weekly admin" hits all seven—check them off against the list.

Seven kinds of visual noise—diagnose each 0 / 7
Tap a symptom on the checklist; the matching lesion on the UI lights up
weekly-report.example.com/admin
Over-decoration
T E A M W E E K L Y
Weak hierarchy
This week at a glance
23 tasks done, 4 delayed, 9 new requests
Data refreshes hourly; tap a card for details
Dense contrast
Done 82% · Delayed 17% · Satisfaction 4.2
Info-free 3D
Quarterly overview3D viz engine · Brand new
Too many colors
Export Filter New Share Archive Settings
Crowded elements
EngDesignMarketingOpsSupportFinanceAdminLegalAllMineDelayedThis week
Heavy separators
Payment gateway migration · Done
Support backlog · In progress
New-hire onboarding · Scheduled
All seven hit. This list is from About Face 4 Chapter 17—written before generative AI, yet every pathology still maps. Read the causes below and you'll see why. When accepting AI output, run this checklist first.
Spot it · The carnival effect

Among the seven noises, "too many colors" has a proper name. About Face 4: colors crowded like a palette overwhelm users—that's the carnival effect. Max out saturation and the noise doubles, stealing the scene from content. Same weekend-market flyer, two palettes—tap the one people can actually read.

Spot it: which side is better Pick a side
Same amount of info, two color paths. Tap your answer.
Plan A
riverside-market.example.com/a
This Sat–Sun · Riverside Park
Opens in 1 day 08 : 42 : 15
Weekend market is on
Craft Coffee Vinyl Plants Limited
Fifty vendors in
Stub draws for vinyl
Scan to join, daily check-in for blind boxes · 20 Moments likes for coffee
Get tickets See the map
Plan B
riverside-market.example.com/b
WEEKEND MARKET
Riverside Weekend Market
This Sat–Sun 10:00–18:00 · Riverside Park · Fifty vendors
Get a free ticketSee the map
CraftPottery, leather & handmade silver — 18 stalls
CoffeeLocal roast & pour-over — 12 stalls
VinylUsed-vinyl market · Live DJ on the hour
Free entry · Pet friendly · Rain or shine
Common beats good

When AI generates design, it picks the high-probability region of training data—the "average" of web design. Same mechanism as language hallucination: in conversation, fluent beats true; in design, common beats good. It ships the purple-gradient trio the way it invents a nonexistent book title with a straight face: both pick the answer that "most looks like it belongs here."

Average is mediocre: what AI gives by default is the median of web design.

Good news: the mechanism leaves a door open—the probability distribution shifts with input. The more specific your description, the narrower the model's options. Narrow enough, and you pull it off the average.

Twist it · The specificity knob

Same brief, three levels of specificity—watch the output change.

Prompt specificity, three levels Flip it
You tell AI
"Make a pretty page."
teamflow.example.com
Powered by a new AI engine
Enter a new intelligent era
Empower every team · Unlock creativity
99.9% uptime · 10k+ teams
Try nowLearn more
Simpler team collaboration
One workspace for projects and docs
Get startedBook a demo
Project management
Doc collaboration
Calendar sync
PROJECT TRACKER
A task system for engineering teams
Keyboard-first, millisecond response, grown into your codebase.
Get startedView docs
IssuesCreate in two keystrokes; auto-link commits
CyclesTwo-week loops; burndown auto-builds
RoadmapQuarterly rollups; live progress
Pick one · Which line pulls AI off the average
Want AI off its default aesthetic—which line works best? Single choice
Wrong answers come with an explanation. Keep going until you hit the right one
A"Make it look nicer"
B"More design sense, more premium"
C"Reference Linear's info density; one primary color sitewide; no gradients"
D"Use whatever style is trending now"
Key Takeaways

The AI look has a source: models default to high-probability training regions—the average of web design. The purple-gradient trio is that region's storefront.

Visual noise has a pathology list: over-decoration, info-free 3D, heavy separators, crowded elements, dense color/texture contrast, too many colors, weak hierarchy—seven from About Face 4. Colors jammed like a palette even have a name: the carnival effect.

Same mechanism as hallucination: in chat, fluent beats true; in design, common beats good. Wherever you gave no instruction, it fills in the most common answer.

The fix is cranking specificity: a reference plus checkable hard constraints (one primary color, no gradients). Each notch pulls output farther from the default look. Next time before AI generates a page, put these five tells on a ban list.

Source: Original to Xiaoshan Academy's Taste Engineering series; some design principles adapted from About Face 4, Chapter 17.