What “slop” actually means
Slop isn’t about whether AI was used. It’s about whether anybody read what came back.
A slop site has a hero, three feature cards in a grid, four testimonials in the same voice, a stock photo of someone looking at a laptop, and a CTA that says “Get started today.” You’ve seen it. You know.
The signals are concrete, not vibes. Templated section ordering. Headings of identical length. Hedge-and-explain rhythm in every paragraph. Stock imagery with the same gradient overlay on every hero. Marketing-speak in every register. None of these alone is decisive. Humans use stock photos. Humans write listicles. Humans say “leverage,” sometimes ironically. It’s only a real call when three or four of these stack up.
A slop detector reads the page through that lens and tells you which axes are leaning, with the specific phrases or sections that triggered the read.
The four axes we read
Content
We count common LLM-marker phrases in the body text: “leverage,” “seamlessly,” “robust,” “unlock the power of,” “in today’s fast-paced world.” A few matches is normal. A dense cluster across many distinct phrases is what tips the score.
Structure
Hero, then 3-icon feature grid, then testimonials, then CTA, then footer. That’s the canonical AI-builder template. We read heading regularity and paragraph length distribution, and we detect the builder itself where we can: Framer, Durable, Mixo, Lovable, v0. The builder name is a high-signal input on its own.
Imagery
Filename patterns, alt-text repetition, the visible artifacts diffusion models leave behind. Stock photography is fine on its own. Stock photography with identical gradient overlays on every hero is part of the signal.
Tone
Voice consistency across sections, register shifts, emotional flatness. A distinct first-person voice or a joke that actually lands moves the score the other direction. Slop’s defining quality is that it has no author.
What you’ll see in the report
- An overall AI score 0–100 with three verdict bands: Mostly human, Mixed signals, Mostly AI-generated.
- A category breakdown across the four axes with one-line signal sentences.
- Up to three findings, each with verbatim quotes pulled from the page, where they appeared, and a one-sentence fix.
- A what’s working block. The strongest human-authorship signals on the page. The score is a fingerprint, not a moral judgment.
- The detected tech stack if an AI page-builder fingerprint shows up.
- A counted list of LLM-marker phrases with hit counts.
What isn’t slop
The score is automation-shaped, not quality-shaped. A few things worth knowing before you panic:
- Templated landing pages aren’t automatically slop. A well-built marketing site can use the same section ordering as a slop site. If the content has a real voice and the imagery is custom, the verdict tends to land at “Mixed signals” rather than “Mostly AI-generated.”
- Drafting with AI is fine. The edit pass is what separates a working page from slop. Drafted-then-edited content lands in the middle of the score band.
- Stock photos alone aren’t slop. Plenty of human-built sites use stock imagery. The signal is uniformity of treatment plus other signals pointing the same way.
Run a free slop scan
Paste a URL on the homepage. Under 30 seconds for most pages. Every scan creates a public report you can send to a co-founder or post anywhere.
Common questions
Is “slop” the same as AI-generated?
Not quite. AI-generated is a question of authorship: was a model in the loop. Slop is a question of effort: did anyone edit the model’s output before publishing. You can have AI-drafted content that isn’t slop (someone edited it) and you can have human-written content that reads as slop (templated, hedged, no voice). The score reads the second axis, not the first.
Will my site score high just because I used a page builder?
Possibly on the structure axis. If your copy has a distinct voice and your imagery is custom, the overall verdict will sit in the “Mixed signals” band rather than “Mostly AI-generated.” The detected builder shows up in the report so you can see exactly what triggered the structural score.
What do I do if the scan says my site is slop?
Read the findings. Each one carries verbatim evidence (the specific phrases or section labels we matched) and a one-sentence fix recommendation. Start with the highest-severity finding, apply the fix, rescan in a few days. If your edit moved the signal, the next scan will reflect it.