There's a look. You land on a site, and before you've read a word, a quiet voice says: AI made this. The blue-purple gradient. Inter set in the same weights as ten thousand other landing pages. A centered hero promising to "build the future of work." Three equal cards. You've seen this exact site before, on a different domain, selling something else entirely.
That recognition is the "built with AI" tell, and in 2026 your prospects have learned to read it fast. Visitors form a credibility judgment about a website in roughly 50 milliseconds—about a third of a blink—and that snap verdict is driven overwhelmingly by visual appeal, not content (Lindgaard et al., Behaviour & Information Technology). When a site pattern-matches to "default AI output," the trust discount lands before your value proposition ever gets a turn. This guide names the twelve signals that trigger that discount, why AI tools default to each one, and what intentional design does instead.
Who this is for: Founders and marketers who shipped a site on Lovable, Bolt, v0, or Framer AI and suspect it reads as cheap—and anyone evaluating a vendor's output and wondering whether "looks fine" is actually costing them deals.
Key Takeaways
- Visitors judge a site's credibility in ~50ms, and visual appeal is the prime driver of that judgment—before they read your copy (Lindgaard et al.)
- AI builders converge on the same defaults because they train on the same popular templates—designers call the result "visual elevator music" (our AI-vs-custom breakdown)
- The tells cluster into four groups: visual defaults (the purple gradient, Inter, the 3-card grid), AI imagery, copywriting patterns (em-dash confetti, buzzword soup), and code/SEO boilerplate
- The most expensive tell isn't ugliness—it's sameness: looking like every other AI site erases the differentiation you're paying to establish
- None of these signals is individually fatal. The damage comes from stacking three or four, which is exactly what an untouched AI export does
- Fixing the tell is rarely a rebuild. It's intentional decisions—type, color, real imagery, specific copy—layered onto what the tool gave you
Why the "AI Tell" Costs You
The mechanism is the halo effect. A first impression formed on visual appeal in 50ms doesn't stay contained to "this looks nice"—it colors every downstream judgment: how credible the company seems, how much you'd pay, whether you trust the checkout. Lindgaard's research found these rapid judgments are highly stable; a longer look rarely overturns the snap verdict, it mostly rationalizes it.
So when a visitor's first impression is "this is a default AI site," that frame attaches to your pricing, your claims, and your competence. Design analysts at studios tracking the trend report that AI-default aesthetics now read as a low-effort signal to a growing segment of buyers—the same way a stock-photo handshake once did. Poor design and content drive away a meaningful share of visitors before conversion (Figma's 2026 research puts visitor abandonment from weak design/content at around 38%). The point isn't that AI output is broken. It's that recognizable AI output spends your credibility on looking like everyone else.
1. The Blue-to-Purple Gradient
The single most reliable tell. Open almost any AI-generated landing page and you'll find an indigo-to-violet gradient—on the hero background, the primary button, the icon accents. There's a structural reason: the default accent in the most-trained-on component systems (Tailwind's indigo-500 and its neighbors) propagated through the training data, so models reach for it by default. It's nobody's brand color; it's the absence of a brand color.
What it signals: "We accepted the default." A color system that maps to your actual brand—even one deliberate, slightly-off hue—instantly separates you from the gradient crowd.
2. Inter Everywhere, and Nothing Else
Inter is a genuinely good typeface. It's also the default in nearly every AI builder, component library, and UI kit—which is exactly the problem. When a site uses Inter (or a system-sans fallback) at a couple of weights with no pairing, no display face, no typographic point of view, that's a strong signal the type was never chosen at all.
What it signals: No typographic intent. A single considered pairing—a distinctive display face for headlines against a clean body—does more for perceived craft than almost any other cheap change.
3. The Centered Hero With a Headline That Fits Anyone
"Build the future of work." "Your all-in-one platform." "Scale without limits." AI ships hero copy that could belong to a CRM, a yoga studio, or a logistics startup with equal ease—because it was generated to fit the shape of a hero, not your specific product. Paired with the obligatory centered layout and a gradient behind it, it's instant wallpaper.
What it signals: You haven't said what you actually do. Specificity is the antidote: name the customer, name the problem, name the outcome in words only your company could write.
4. The Three Equal Cards and Identical Spacing
Below the hero: three cards of equal width, 16px corner radius, the same 24px of padding repeated down the whole page. AI defaults produce a rhythm so uniform it reads as flat—every section the same height, the same gap, the same card. Real design has hierarchy: things that matter get more room.
What it signals: No editorial priority—everything weighted the same because nothing was decided. Deliberate asymmetry and varied spacing signal a human made choices about what matters most.
5. Motion With No Purpose
Generic fade-ins on scroll, all firing with identical timing and easing. Hover states that don't actually respond. Animation bolted on as decoration rather than tied to a state change or used to direct attention. AI tools add motion because "modern sites have motion," not because the motion means anything.
What it signals: Polish theater. Purposeful motion—feedback on interaction, transitions that orient the user—reads as craft; uniform decorative fades read as a checkbox.
6. The Plastic AI Illustration
AI-generated illustrations and 3D blobs share a quality designers describe as "slightly too smooth, slightly too symmetrical"—a plastic sheen, uncanny lighting, gradients that don't occur in real materials. Generated photography has its own tells: mangled text on signage, hands with the wrong number of fingers, backgrounds that dissolve under inspection.
What it signals: Even the imagery is synthetic. Custom illustration, real photography, or genuine 3D work signals investment that a generated asset can't fake.
7. No Real Proof
The stock cliché evolved but didn't die: the diverse team smiling at a laptop in an impossibly well-lit office. What's missing is the tell—no real product screenshots, no actual team photos, no customer logos, no specific numbers. AI fills space with plausible-looking generic imagery because it has no access to your real proof.
What it signals: Nothing here is verifiable. Real screenshots, named customers, and specific results are the cheapest trust upgrade available, and AI can't generate them for you.
8. Em-Dash Confetti and the Telltale Openers
Em-dashes scattered through every paragraph—like confetti—where a human would use one sparingly. Sentence openers that have become AI fingerprints: "In today's fast-paced world," "a rich tapestry of," "the landscape of [your industry]." The em-dash is a fine tool; the density is the signal, alongside those stock transitions.
What it signals: The copy was generated, not written. (Yes, this very article uses em-dashes—the tell is using them in every sentence as connective filler, not for genuine asides.)
9. Buzzword Soup
"Seamless." "Robust." "Cutting-edge." "Future-ready." "Best-in-class." "Unlock value." "Elevate your [noun]." AI reaches for superlatives because they're statistically common in marketing copy and commit to nothing. A page can be full of these words and still leave a reader unable to say what the product does.
What it signals: Hot air. Replace every superlative with a concrete claim—a number, a named capability, a before/after—and the page stops sounding generated.
10. Hedging and Emoji Bullets
"May help you." "Can potentially streamline." Feature lists led by emoji (🚀 ✨ 💡), every conceivable benefit hit, no stance taken. AI hedges because it's optimized to avoid being wrong, and it covers every point because it can't prioritize. The result is copy that's comprehensive and forgettable.
What it signals: No conviction. Brands people trust take positions; they say "this is for you, and this isn't."
11. Boilerplate Meta and Generic Alt Text
Open the source. Duplicate or templated meta descriptions across pages. A default OG image (or none), so the link unfurls as a gray box on social. Alt text that reads image or is simply absent. AI scaffolds the page structure but rarely fills the metadata layer with anything specific.
What it signals: Nobody finished the job. Real meta descriptions, a custom OG image, and descriptive alt text are invisible until shared or searched—and their absence quietly signals the same low effort the visible design does.
12. Templated, Flaw-Dense Code
Under the hood, untouched AI exports tend toward repeated component structure, dead utility classes, and accessibility gaps. Independent analyses of AI-generated code report markedly higher issue density than human-written code—roughly 1.7× more total issues in some studies, and a substantial share carrying a security or design flaw (figures vary by study; treat as directional, not gospel). The visitor never sees the code, but they feel its symptoms: layout shift, broken focus states, sluggish interaction.
What it signals: The foundation was never reviewed. Performance, accessibility, and clean structure are where a studio's involvement shows even when the design looks superficially similar.
The Fix: Where Custom Design Pays Off
Notice what almost none of these fixes require: a full rebuild. The tell is rarely that AI built something broken—it's that AI built something generic, and generic is a series of un-made decisions. Choosing a real color system, one deliberate type pairing, specific copy, and genuine imagery removes most of the tells in days, not months.
The threshold is the same one we draw in our AI-vs-custom decision guide: an AI export is fine while your site is a placeholder. The moment your website is your first impression—when you're losing deals to better-looking competitors, raising, or rebranding—the sameness becomes the liability. Premium custom web work spans a wide ladder, from roughly $5K–$15K for a focused brand site up to $100K–$200K+ for flagship interactive experiences; the right rung depends on what your brand has to outrun. For teams shipping fast on AI builders, the highest-leverage move is usually not abandoning the tool—it's layering intentional design onto the MVP once perception starts to matter, the same way you'd brief a branding and design partner for any other strategic asset.
About Utsubo
Utsubo is a creative studio building custom web experiences—interactive, immersive, and built around a brand's actual identity rather than a template's defaults. We work in Three.js, WebGPU, and motion design for brands that need their first impression to do more than pass. When a site needs to read as deliberate at a glance, that's the work.
Let's Talk
Suspect your site is giving off the tell? We help brands move from default to distinctive—often without starting over.
If you're exploring a partnership, let's discuss:
- Which tells your current site is triggering, and which actually cost you
- Whether the fix is a polish pass or a rethink
- What level of custom work (animated, interactive, full 3D) fits your goals and budget
Checklist: Is Your Site Giving Off the AI Tell?
- Your accent color is an actual brand color, not the default indigo/violet gradient
- Your typography is a deliberate pairing, not Inter at default weights alone
- Your hero headline names what you specifically do—no one else could use it
- Your layout has hierarchy; not every section is an equal three-card row
- Motion responds to interaction or directs attention—it isn't decorative fade-in
- Illustrations and imagery are custom or real, not plastic AI renders
- You show real proof: product screenshots, team photos, named customers, numbers
- Copy uses em-dashes sparingly and avoids "In today's fast-paced world" openers
- Superlatives are replaced with concrete, specific claims
- The page takes a position instead of hedging and emoji-bulleting every feature
- Meta descriptions, OG image, and alt text are written and specific
- Code has been reviewed for performance, accessibility, and clean structure
FAQs
How can you tell if a website was made with AI?
Look for stacked defaults rather than any single flaw: a blue-to-purple gradient, Inter as the only typeface, a centered hero with generic copy ("build the future"), three equal feature cards with identical spacing, plastic AI illustrations, and copy heavy with em-dashes and superlatives. One of these alone proves nothing—real designers use gradients and Inter too. It's the cluster of three or four untouched defaults that marks an unedited AI export.
Does an AI-looking website actually hurt conversions?
Indirectly but measurably. Visitors judge credibility in about 50 milliseconds, driven mainly by visual appeal, and that judgment colors everything after it via the halo effect (Lindgaard et al.). When a site reads as default AI output, the trust discount applies to your pricing and claims before they're evaluated. Figma's 2026 research attributes roughly 38% of visitor abandonment to poor design and content. The cost is rarely a single broken element—it's the sameness eroding the differentiation you're paying to build.
Why do all AI-generated websites look the same?
Because they train on the same data. The most popular component libraries, templates, and high-traffic sites dominate the training set, so models converge on their defaults—Tailwind's indigo accent, Inter, card grids, centered heroes. Designers call the result "visual elevator music": competent, inoffensive, and indistinguishable from the next one. Sameness is the structural outcome of optimizing for "what usually works," not a bug.
Is it bad to build a website with AI tools like Lovable or Bolt?
No—AI builders are excellent for MVPs, placeholders, and validating an idea fast. The problem starts when "temporary" becomes your permanent public face. The fix usually isn't abandoning the tool; it's layering intentional design—real color, type, copy, and imagery—onto what it generated once your brand needs to compete on perception.
What's the fastest way to make an AI-built site look custom?
In rough order of impact: replace the default gradient with a real brand color; choose one deliberate type pairing instead of Inter alone; rewrite the hero to say specifically what you do; add real proof (screenshots, team photos, named customers); and swap plastic AI illustrations for custom or real imagery. These are decisions, not rebuilds, and they remove most tells in days.
Are em-dashes really a sign of AI writing?
The em-dash itself isn't—it's a legitimate, useful piece of punctuation that good writers have always used. The tell is density: AI tends to scatter em-dashes through nearly every sentence as connective filler. Combined with stock openers like "In today's fast-paced world" and "a rich tapestry of," frequent em-dashes become a recognizable fingerprint of generated copy. Used sparingly and intentionally, they're fine.
Do these AI tells affect SEO and AI search visibility?
They can. Boilerplate meta descriptions, missing alt text, and default OG images weaken both traditional SEO and how well your pages surface in AI-driven search. Generic, low-specificity content also tends to earn fewer citations from AI overview tools, which favor distinctive, verifiable information. The same lack of intent that shows up visually shows up in the metadata layer—and search systems notice.
When should I invest in custom design instead of an AI builder?
When your website stops being a placeholder and starts being your first impression: you're losing competitive evaluations where site quality matters, raising funding, rebranding, or your brand has outgrown the template. Below that threshold, an AI builder is the smart, economical choice. Above it, the sameness becomes a measurable liability. See our AI-vs-custom decision guide for the full framework.
Technology-First Creative Studio
