De Slopify

De-Slopify — Remove AI Writing Artifacts

Core Insight: You can't do this with regex or a script. It requires manual, systematic review of each line. Slop has a mechanism: the text becomes simultaneously less specific and more exaggerated — rare concrete facts get smoothed into generic important-sounding claims. So the fix is never just deleting tics; it's reinstating the concrete detail the slop displaced. Human writers use most of these devices intentionally and sparingly; the problem is density and reflex, not any single occurrence.

THE EXACT PROMPT — Full De-Slopify

I want you to read through the complete text carefully and look for any telltale
signs of "AI slop" style writing; one big tell is the use of emdash. You should
try to replace this with a semicolon, a comma, or just recast the sentence
accordingly so it sounds good while avoiding emdash. This applies to emdashes
used as prose punctuation only: leave dash/minus characters alone when they are
doing non-prose work, such as in math formulas, code, CLI flags, numeric ranges,
or tables.

Also, you want to avoid certain telltale writing tropes:
- Setup-reversal copy ("Four steps, and only one of them is yours." /
  "Everything about billing changed. Your invoice didn't.")
- Negative parallelism ("It's not [just] XYZ, it's ABC")
- Clickbait lead-ins ("Here's why", "Here's the thing")
- AI vocabulary clusters (delve, pivotal, tapestry, intricate, underscores,
  seamless, robust, vibrant, leverage) and copula avoidance ("serves as",
  "marks a", "boasts" instead of plain "is"/"has")
- Significance inflation ("marks a pivotal moment", trailing "-ing" commentary
  clauses like "highlighting its importance")
- Bullet lists of "- **Snappy Label:** text" where prose would read better
- Filler: paragraphs that say in four sentences what fits in one
Basically, anything that sounds like the kind of thing an LLM would write
disproportionately more commonly than a human writer and which sounds
inauthentic/cringe.

And you can't do this sort of thing using regex or a script, you MUST manually
read each line of the text and revise it manually in a systematic, methodical,
diligent way. Use ultrathink.

THE EXACT PROMPT — Quick Version

Review this text and remove AI slop patterns: emdashes used as prose
punctuation (not dashes in math, code, or ranges), setup-reversal copy, "It's not X, it's Y" formulas, "Here's
why" lead-ins, AI vocabulary (delve/pivotal/tapestry/underscores/seamless),
"serves as" instead of "is", trailing "-ing" significance clauses, bold-label
bullet lists, and filler. Recast sentences to sound naturally human; restore
concrete specifics where slop replaced them. Use ultrathink.

High-Signal Tells (fix on sight)

Judgment Calls (fix when dense, keep when earned)

• Filler: a paragraph whose content fits in one sentence, or that has no discernible claim. Compress or delete; this is the essence of slop.
• Generic analogies: plausible but unconsidered metaphors ("every chord is a puzzle piece that clicks into a song"). Delete or make them specific.
• Hedges ("It's worth noting...", "Keep in mind..."): delete unless the caveat changes reader behavior.
• Formatting slop: bold scattered without emphasis logic, emoji bullets (✅🎯🚀), Title Case Headings, heading-only sections, rules between every section, tables for prose, heavy bullet density.
• Monotone rhythm: every sentence the same length, unnaturally uniform tense/POV. Vary it.
• Elegant variation: rotating synonyms for a key term. Technical writing should repeat the term.
• Weasel attribution: "Experts argue", "Industry reports suggest". Name the source or cut the claim.
• The challenges formula: "Despite X, faces several challenges... Future outlook..." Cut the arc; keep one real limitation.

Red Herrings (do NOT "fix")

• Perfect grammar and absent typos — humans have spell-check too
• Absence of contractions — may be house style or non-native phrasing
• Academic vocabulary alone — "ascertain" in one spot isn't slop; a cluster of AI-vocabulary words is
• Dash characters outside prose — minus signs in math formulas, dashes in code or CLI flags, en dashes in numeric ranges, table syntax. Only prose-punctuation emdashes are the tell
• AI-detector tool output — these tools false-positive constantly; the review is manual judgment, in both directions

Before/After Examples

Setup-reversal copy:

Before: Ten features. Zero headaches.
After:  Includes ten features for managing billing, invoicing, and usage tracking.

Emdash + negative parallelism:

Before: It's not just a linter—it's a complete code quality system.
After:  A code quality system that covers linting, formatting, and CI checks.

"Here's why":

Before: We chose Rust for this component. Here's why: performance matters.
After:  We chose Rust for this component because performance matters.

Significance inflation (restore the specific fact):

Before: The v2 release marks a pivotal moment, underscoring our commitment to reliability.
After:  v2 fixes the three crash bugs reported since March.

Copula avoidance:

Before: The CLI serves as the primary entry point and boasts a robust plugin system.
After:  The CLI is the primary entry point and has a plugin system.

Forced enthusiasm:

Before: Let's dive in! We're excited to help you get up and running.
After:  Install the tool and run your first command in under a minute.

Workflow

• Read the entire text once for register: does it sound like ad copy or a lecture?
• Line-by-line pass: mark every high-signal tell (table above)
• Density pass: judgment-call patterns — fix only where they cluster
• Specificity pass: wherever you cut inflated claims, reinstate concrete facts
• Reread the result; it should sound like a competent engineer explaining, not a brand

Why Manual Review is Required

1. Context matters — sometimes an emdash or triad is actually the right choice
2. Recasting sentences — often the fix isn't substitution but rewriting
3. Restoring specifics — a script can delete "pivotal" but can't supply the missing fact
4. Tone consistency — need to maintain one voice throughout

When to De-Slopify

• Before publishing a README
• Before releasing documentation
• After AI-assisted writing sessions
• During documentation reviews

What NOT to Fix

• Technical accuracy — don't sacrifice correctness for style
• Necessary structure — headers and lists that organize real content are fine
• Clear explanations — being thorough isn't slop; filler is saying nothing at length
• Code examples — focus on prose, not code
• Earned devices — a device used once, deliberately, for real effect can stay

Reference Index