AiSlopTest [principle] v1.0.0

The final quality bar for any AI-generated interface is whether a viewer would immediately assume 'AI made this.' A distinctive interface makes someone ask 'how was this made?' — not 'which AI made this?'

Implications

• Run this mental test before shipping: if someone could mistake it for every other 2024-2025 AI output, iterate.
• The DON'T rules (no Inter, no cyan-on-dark, no gradient text, no glassmorphism) are literal fingerprints — each one is a tell.
• Diversity across generations is required: the same aesthetic commitment becomes a new fingerprint when repeated 1000 times.
• Bold and wrong is recoverable; safe and generic is already a failure.

AiSlopTest [principle] v1.0.0

The final quality bar for any AI-generated interface is whether a viewer would immediately assume 'AI made this.' A distinctive interface makes someone ask 'how was this made?' — not 'which AI made this?'

Implications

• Run this mental test before shipping: if someone could mistake it for every other 2024-2025 AI output, iterate.
• The DON'T rules (no Inter, no cyan-on-dark, no gradient text, no glassmorphism) are literal fingerprints — each one is a tell.
• Diversity across generations is required: the same aesthetic commitment becomes a new fingerprint when repeated 1000 times.
• Bold and wrong is recoverable; safe and generic is already a failure.

Sources

Rationale

Generic visual choices — Inter, rounded cards, purple-on-dark, centered hero layouts — are the statistical average of AI training data. Models optimize for safe, inoffensive outputs that score well in RLHF but read as machine-generated at a glance. The AI Slop Test operationalizes this: show the output to someone and say 'AI made this' — if they immediately believe you, the design has failed to establish any distinctive voice. The test is binary: either the design surprises or it doesn't.