AI Slop Detection Is Here — Your Content Strategy Needs to Change

For the past two years, the dominant content strategy for a lot of sites has been simple: generate a lot, publish a lot, let some percentage rank. That strategy is dying, and it’s dying faster than most teams have noticed.

Search Engine Land recently flagged something worth taking seriously: platforms have built classifiers that detect low-effort AI-generated content — “slop” — with better than 90% accuracy. This isn’t a vague policy statement about “quality content.” It’s a working detection system, and it’s already being used to demonetize and suppress pages, not just on Google, but across Pinterest and other platforms that depend on content quality to keep users around.

Why detection got this good, this fast

AI slop has a signature. Low-effort, mass-produced AI content tends to repeat sentence structures, recycle the same rhetorical moves, lean on emotionally manipulative hooks, and avoid saying anything specific enough to be wrong. That’s exactly the kind of pattern a classifier is good at catching — it doesn’t need to understand your industry, it just needs to see the shape of the text.

This matters because the economics that made slop attractive are collapsing. The entire pitch of AI content at scale was: production cost drops to nearly zero, so you can afford to publish 10x and let volume compensate for quality. If a chunk of that output gets filtered before it ever reaches an audience, the volume play stops paying for itself. Cheap to produce no longer means cheap to rank. It means cheap to get quietly excluded from distribution.

Watermarking is coming from the model side too

Anthropic has already implemented content watermarking, and under the EU AI Act, disclosure requirements for AI-generated content are moving from optional to expected. Other model providers are expected to follow. Combine that with platform-side detection and you get pressure from both directions: the content itself may carry a traceable signal, and the platform is independently screening for the patterns anyway.

For sites serving EU audiences — which covers a lot of B2B SaaS and e-commerce brands — this isn’t a distant compliance issue. It’s a content production issue that needs a policy now, not after enforcement starts.

What actually holds up

None of this means AI tools are off the table. It means unedited AI output published as-is is now a liability, not a shortcut. The content that survives detection and keeps ranking has a few consistent traits:

• A named author with actual expertise in the subject, not a generic “Team” byline
• Specific claims, numbers, and examples that couldn’t have come from a generic prompt
• Evidence of human judgment — a point of view, a disagreement with conventional wisdom, a real example from the author’s own work
• No padding. If a paragraph exists just to hit a word count, cut it.

This is the same list that’s always defined good content. The difference is that platforms can now check for it programmatically, at scale, and act on the result automatically. E-E-A-T stopped being a ranking philosophy you could gesture at and became something closer to a technical filter you either pass or don’t.

The practical shift for content teams

If your process is “AI drafts, someone skims it, it publishes,” that process needs a checkpoint. AI-assisted drafting is fine — most competent teams use it somewhere in the pipeline — but there has to be a stage where a real subject-matter expert adds something the model couldn’t: a specific data point, a client anecdote, a correction to something the draft got subtly wrong. If you can’t point to what a human added, a classifier increasingly can’t either, and neither can a reader.

The same logic applies to volume targets. Publishing cadence built around “how many articles can we generate this month” is the wrong metric now. The better question is how many articles this month say something a competitor’s AI tool couldn’t have produced on its own. That’s a smaller number, and it should be.

If you’re not sure where your existing content falls on this spectrum, or you want a second opinion on what a review process should catch before publishing, that’s a good starting point for an SEO audit.