ViewDAO
Introducing Slop Check: AI Pattern Detection for Content.
Most AI-generated writing follows predictable patterns. The same adjectives appear in the same positions. Sentences alternate between short and long in mechanical rhythm. Paragraphs open with claims, follow with evidence, close with implications. Readers notice. They just don't have a way to measure it.
Slop Check is a detection tool built on one principle: bad AI writing has a fingerprint. It's not random. It's not natural variation. It's the statistical residue of training data compressed into decision trees and large language model outputs. The tool scans for 200+ signals across vocabulary, syntax, structure, and rhetoric. Redundant adjectives. Fake certainty markers. Symmetric takes. Copula abuse. Filler nouns that drift between paragraphs without anchoring. When these signals cluster, the text registers as higher slop probability.
The system doesn't flag any single phrase as "AI" (single phrases aren't dispositive). Instead it looks for density and distribution. A newsletter with five "Indeed" markers scattered across eight hundred words might score 0.4. The same word appearing three times in two hundred words might score 0.7. Context matters. Specificity matters. Clustering matters. This is why human judgment still screens the final call, but the heuristics do real filtering work. Early testers report catching roughly seventy percent of obvious LLM output and fifty percent of moderately laundered AI writing. False positives run around five percent on clearly human-written pieces.
The tool came from a more specific problem. Content platforms were being flooded with acceptable-looking but hollow posts. Not obviously wrong. Just filler with better grammar than human filler usually has. A designer at an analytics firm noticed her Slack was clogged with AI-written updates that satisfied no one but technically met the requirement to "update stakeholders." A finance writer got pitched by a PR firm using a template so mechanical you could see the sentence shuffler's work. An AI safety researcher watching governance discussions online noticed the same three takes repackaged in slightly different word order, each claiming to be original analysis.
Slop Check won't replace reading. It will replace the part of reading that's pattern-matching for hollowness. Right now that work is invisible and unpaid. You've done it a thousand times. You open something, skim two paragraphs, feel the flatness, close the tab. The tool makes that feeling explicit. It says: "This text has 47 pattern matches with AI-generated content in your training sample. 12 appear in the opening two paragraphs." You then decide if that matters for your use case. Sometimes it won't. A medical explainer generated by Claude might score high on slop markers but still be useful. A CEO memo that scores high might be a red flag. The probability is data. The judgment is yours.
The core insight isn't that AI writing is always bad. It's that undetected AI writing is almost always being used dishonestly. Someone is trying to pass generated content off as human-written analysis. Or they're too lazy to edit. Either way, the reader deserves to know what they're reading. Slop Check makes that visible. It won't end AI writing in professional contexts. But it will make it harder to hide.