Deslopify: AI Slop & Trope Removal #

Procedures, editorial standards, and de-slopification workflows for purging text of recognizable Large Language Model (LLM) structural patterns, formulaic clichés, and conversational tropes to make technical writing sound authentically human, grounded, and engaging.

Reference Catalog of AI Tells #

Consult references/tropes.md for the exhaustive catalog of AI tells, syntactic anti-patterns, and overused vocabulary.

Quick Reference: Common AI Tells & Direct Alternatives #

Category	Overused AI Pattern	Human-Sounding Alternative
Magic Adverbs	quietly, deeply, fundamentally, remarkably, arguably	Cut entirely or replace with concrete metrics and verified facts.
Pompous Vocabulary	delve, tapestry, landscape, robust, seamless, leverage, harness, testament	Use simple, concrete words: explore, look at, system, reliable, fast, use.
The "Serves As" Dodge	serves as a reminder, stands as a testament, marks a pivotal moment	Use direct copulas: is, reminds us, shows, routes, was built in.
Negative Parallelism	"It's not X — it's Y", "Not because X, but because Y"	State the point directly without staging a theatrical contradiction.
Dramatic Countdowns	"Not a bug. Not a feature. A design flaw."	Combine into a single direct statement: "This is a design flaw."
Self-Answering Rhetoric	"The result? Devastating.", "The scary part? Nobody noticed."	State facts without self-answering drama: "Nobody noticed the failure."
Filler Transitions	"It's worth noting that...", "Importantly...", "Notably..."	Cut the preamble. If it's worth noting, state the fact directly.
Pedagogical Tones	"Let's break this down", "Here's the thing", "Here's the kicker"	Eliminate teacher-mode signposting; present the technical facts directly.
Fractal Summaries	Summarizing every section at the end of the section	Let the section content speak for itself; eliminate redundant sub-conclusions.

Before & After Transformation Matrix #

Use these concrete technical writing transformations to convert formulaic AI slop into crisp, peer-to-peer engineering prose:

AI Slop Anti-Pattern	Raw AI Draft (Slop)	Human Engineering Rewrite	Key Editorial Changes
Pompous Vocabulary + Adverbs	"The system quietly leverages a robust orchestration pipeline to seamlessly deliver unprecedented throughput across the entire distributed landscape."	"The system uses a distributed pipeline to process 50,000 requests per second."	Cut buzzwords (quietly, leverages, robust, seamlessly, landscape); added verified metrics.
Negative Parallelism + Self-Answering Drama	"It's not just a caching layer — it's a fundamental reimagining of state. The result? Instantaneous queries."	"Keeping index pages in memory reduced query latency from 80ms to 2ms."	Removed theatrical negation ("not just X — it's Y") and rhetorical drama ("The result?"); stated technical cause and effect directly.
Dramatic Countdown	"Not a slow disk. Not network jitter. A deadlock in the connection pool."	"A connection pool deadlock stalled all worker threads."	Eliminated faux-suspense countdown; stated root cause upfront.
Pedagogical Preambles & Fillers	"It's worth noting that before we delve into the implementation, let's break down why this matters."	"Here is the connection pool architecture:"	Cut teacher-mode signposting ("delve", "let's break down", "it's worth noting").
The "Serves As" Dodge	"The proxy server serves as a crucial gateway, standing as a testament to modular design."	"The proxy routes incoming traffic and terminates TLS."	Replaced pompous copulas (serves as, stands as a testament) with active technical verbs (routes, terminates).
Superficial Present-Participle Analysis	"We enabled HTTP/3, highlighting the team's forward-looking approach and underscoring our commitment to performance."	"Enabling HTTP/3 eliminated head-of-line blocking on packet loss."	Cut hollow puffery ("highlighting...", "underscoring..."); explained concrete technical benefit.

4-Stage Deslopification Workflow #

Follow this procedure when reviewing or rewriting any text:

graph LR
    A[1. Scan & Tag Tells] --> B[2. Deconstruct AI Structure]
    B --> C[3. Direct Active Rewrite]
    C --> D[4. Cadence & Rhythm Audit]

1. Stage 1: Scan & Tag Tells #

Scan the input text against the patterns in references/tropes.md.
Tag every occurrence of:
- Magic adverbs (quietly, deeply, fundamentally).
- Pompous vocabulary (delve, tapestry, landscape, robust, seamless).
- Rhetorical questions and false suspense (The catch?, Here's the kicker).
- Negative parallelism (It's not just X, it's Y).

2. Stage 2: Deconstruct AI Structure #

Strip out formulaic LLM structures:
- Fractal summaries: Delete sub-conclusions under intermediate headings (e.g., "In summary, this step showed...").
- Signposted conclusions: Replace "In conclusion", "Wrapping up", and "The bottom line" with direct summaries or actionable next steps.
- Bold-first bullet fatigue: Convert long lists of bold-leaded pseudo-bullets into connected, narrative paragraphs where appropriate.
- Excessive em-dashes: Limit em-dashes to at most one per document; use parentheses, commas, or separate sentences instead.

3. Stage 3: Direct Active Rewrite #

Replace inflated vocabulary: Simplify grandiose descriptors to plain, active verbs and concrete nouns.
Break formulaic symmetry: Vary sentence lengths dramatically. Follow a long, nuanced sentence with a short, punchy one.
Remove false suspense: State the takeaway upfront (Inverted Pyramid style) rather than holding back information for a staged reveal.
Ground claims in specifics: Replace vague hype ("provides an incredibly powerful mechanism") with concrete technical specifics ("executes within 5ms on a single core").

4. Stage 4: Cadence & Rhythm Audit #

Read the final draft to verify:
- Does it sound like a knowledgeable human engineer speaking with a peer?
- Are there any lingering rule-of-three triplets (tricolons)?
- Is the sentence length varied naturally rather than stuck in uniform 15-word rhythms?
- Is the tone authentic, conversational, and grounded in real-world engineering nuance?

deslopify