Defend 'AI Slop' Patterns by Auditing Rhythm
Banned patterns like rule of three, em dashes, and binary contrasts are rhetorical tools—measure perplexity, burstiness, and entropy to spot autopilot repetition vs. intentional craft, then build an AI detector.
Rhythm Metrics Separate Alive Writing from Flat Prose
Great writing syncopates like Stravinsky's Rite of Spring, breaking predictable 4/4 time. Use three metrics to diagnose:
- Perplexity: Surprising word choices defy predictions. Low perplexity yields generic prose (e.g., overusing delve, leverage, tapestry feels flat only if unchosen). High perplexity, from multilingual brains or century-spanning vocab, creates voice—readers revolt pleasurably before brains catch up.
- Burstiness: Vary sentence lengths for impact. Joan Didion mixes long winds, short slaps, medium breaths; AI clusters medium sentences (3-4 lines per paragraph). Fake burstiness (overdone one-word punches) returns to monotony. Vary to sustain attention—visual paragraph lengths signal thought units, turning walls into landscapes.
- Information entropy: Pack new thinking per sentence. Low entropy restates known ideas; high delivers density. Voice guides alone fail—rhythm underpins style.
These metrics flag metronomic drafts from AI or humans, enabling intentional choices that grab readers.
8 Flagged Patterns Work When Chosen, Fail on Autopilot
Internet bans ignore linguistic norms; defend patterns with diagnostics:
- Inanimate agency: Native to English (Peter Master's study of 3,000 subject-verb pairs shows it outpaces passives). Autopilot stacks four ('The framework reveals...'); chosen: one precise use ('Thermometer measures temperature'). Ask: Does a human belong here?
- Binary contrasts: English merges German's aber/sondern. Autopilot fakes insight ('Not harder, smarter'); chosen corrects beliefs ('Music wasn’t wrong. It was too right'). Ask: Does it negate a real reader assumption?
- Wh-openers (clefts): Front-load old info, emphasize new. Autopilot delays ('What makes this interesting is constraint'); chosen resets after buildup. Ask: Does pre-'is' add meaning?
- Colon reveals: Cataphoric signposts build models. Autopilot vaguens ('Here’s the thing: consistency'); chosen compresses ('Fatal flaw: forgot mobile'). Ask: Does pre-colon contribute?
- Negative listing (apophasis): Suppresses propositions. Autopilot wastes cognition ('Not tutorial, listicle...'); chosen corrects ('Didn’t quit from failure/tiredness—boredom'). Ask: Were readers assuming negations?
- Rule of three (tricolon): Aristotle's completeness (one=power, two=comparison, three=pattern). Autopilot fills ('Speed, efficiency, innovation'); chosen breaks ('God created humanity. Humanity AI. AI religion'). Ask: Does third surprise or complete?
- Uniform paragraphs: Kills visual burstiness. Autopilot: identical 3-4 sentence bricks. Chosen: rare, for syncopation—one-sentence punches amid immersion.
- Parallel kickers: Habituation dulls repeats. Autopilot: every section mic-drops; chosen: one punch amid flats. Ask: Can readers predict endings?
Em dashes rhythmically pause—banning flattens without replacement. AI edits insert 5 patterns in 20 seconds (e.g., stacking inanimates, empty colons), erasing human agency.
Build AI Content Rhythm Analyst in One Prompt
Paste this prompt into Claude/GPT/Gem for 9 files: 8 pattern refs (definitions, autopilot/writer examples, questions) + INSTRUCTIONS.md. Upload to Claude Project (add Voice Profile). Paste drafts for audits: pattern flags, 1-10 burstiness score (1=metronomic, 10=Stravinsky). Flags repetition—you judge choice vs. accident. Doesn't deem 'good/bad', human/AI, or fix structure/emotion. Premium kit skips setup. Result: permanent editing ears, turning bans into intentional rhythm.