Detecting Ai Slop

---
name: detecting-ai-slop
disable-model-invocation: true
source: superpowers-plus
augment_menu: true
triggers: ["/sp-detect", "calculate slop score", "check for AI slop", "detect AI writing", "slop density", "is this AI generated", "writing definitions", "tooltip text", "prose for documentation", "writing prose", "documentation text", "review AI text", "check AI writing", "score this text", "analyze writing quality"]
anti_triggers: ["fix slop", "rewrite", "edit this writing", "remove AI slop", "improve AI draft"]
description: Use when analyzing text to calculate a slop score (0-100) that measures AI slop density. Read-only analysis — does NOT rewrite text (use eliminating-ai-slop for rewrites). Invoke for CVs, cover letters, marketing copy, drafts, tooltip definitions, documentation prose, or any text where you need to quantify machine-generated patterns before deciding whether to edit.
summary: "Use when: analyzing text for AI slop score. Use for CVs, cover letters, documentation prose."
coordination:
  group: writing
  order: 1
  requires: []
  enables: ['eliminating-ai-slop']
  escalates_to: []
  internal: false
composition:
  consumes: [markdown-content]
  produces: [slop-score-report]
  capabilities: [analyzes-writing, scores-quality]
  priority: 30
---

# Detecting AI Slop

> **Guidelines:** See [CLAUDE.md](../../CLAUDE.md) for writing standards.
> **Last Updated:** 2026-09-11
> **See also:** [reference.md](./reference.md) (pattern dictionary), [examples.md](./examples.md) (usage examples)
>
> **Wrong skill?** Rewriting to remove slop → `eliminating-ai-slop`. Profanity/inappropriate language → `professional-language-audit`.

## Runtime Enforcement Tool

**`tools/slop-check.sh`** is the executable implementation of the pattern catalog in `reference.md`. It is the shared gate used by all prose publication paths — wiki writes, PHR gates, Linear comment gates, and recruiting docs — so patterns are enforced consistently and only need to be maintained in one place.

| Caller | Invocation |
|--------|------------|
| `wiki-content-check.sh` (write gate) | `slop-check.sh --content <file> --mode full` |
| PHR pre-sentinel gate | `slop-check.sh --content <file> --mode summary` |
| Linear comment gate (silent) | `slop-check.sh --content <file> --mode silent` |

```sh
# Blocking gate (exits 1 on violations)
slop-check.sh --content FILE

# Silent gate (no output, just exit code)
slop-check.sh --content FILE --mode silent

# Summary mode
slop-check.sh --content FILE --mode summary
```

Exit codes: `0` = clean (warnings only), `1` = blocking violations, `2` = usage error.
Weak intensifiers (`very`, `extremely`, etc.) are always advisory -- they appear in output but never set exit code 1.
Advisory-tier buzzwords and fillers (`interplay`, `meticulous`, `intricate`, `vibrant`, `at the heart of`, and the rest marked advisory in `reference.md`) behave the same way. `slop-check.sh --list-patterns` prints the full catalog with each entry's tier.

When adding a new banned pattern, add it to `reference.md` and `tools/slop-check.sh` in the same commit.

## Detection Approach

This skill analyzes text and produces a **slop score** (0-100) with detailed breakdown by detection dimension. Use it to quantify AI slop before deciding whether to rewrite.

**Core principle:** Detection is read-only. This skill flags patterns but does not rewrite. Use `eliminating-ai-slop` for active rewriting.

## When to Use

- Score a CV or resume for AI-generated content
- Analyze cover letters for generic patterns
- Audit marketing copy for slop density
- Review your own AI-assisted drafts before editing
- Compare before/after versions of edited text
- Triage documents: which need the most cleanup?

## Content Type Detection

The skill auto-detects content type from context:

| Content Type | Detection Signals |
|--------------|-------------------|
| Document | Default fallback |
| Email | "email", "to:", "subject:" |
| LinkedIn | "linkedin", "post", "connections" |
| CV/Resume | "resume", "cv", "experience" |
| Cover Letter | "cover letter", "dear hiring" |
| README | Filename is "README" |
| PRD | "requirements", "PRD", "product" |
| Commit Message | Conventional-commit prefix, bare or scoped (`feat:`, `fix(api):`, `refactor:`), or explicitly labeled "commit message" |
| Code Review Comment | Explicitly labeled "review comment", or inline diff context is present |

**Override:** "Analyze this as a [type]: [text]"

## Output Format

```text
Slop Score: 73/100

Breakdown:
├── Lexical:      28/40  (14 patterns in 500 words)
├── Structural:   18/25  (formulaic intro, template sections)
├── Semantic:     12/20  (hollow examples, absolute claims, and other Semantic findings — see Top Offenders)
└── Stylometric:  15/15  (low sentence variance, flat TTR)

Top Offenders (showing 10 of 23):
 1. Line 12: "incredibly powerful" [Generic booster]
 2. Line 34: "leverage synergies" [Buzzword cluster]
 3. Line 56: "it's important to note" [Filler phrase]
 4. Line 78: "Framed through Growth Mindset: People, Process, Technology" [Framework name-dropping — Semantic, listed despite dimension cap]
 ...

Stylometric Measurements:
├── Sentence length σ: 7.3 words (target: >15.0) ⚠️
├── Type-token ratio: 0.48 (target: 0.50-0.70) ⚠️
└── Hapax rate: 31% (target: >40%) ⚠️

Verdict: Heavy slop. Substantial rewrite needed.
```

## Scoring Algorithm

| Dimension | Max Points | Calculation |
|-----------|------------|-------------|
| Lexical | 40 | `min(40, 2*(pattern_count - dash_count) + 3*dash_count)`, where `dash_count` is the number of em-dash instances (en-dashes are legitimate punctuation and never count) (each worth +3 instead of +2; see reference.md typographic-tells section) and `pattern_count` includes those dash instances |
| Structural | 25 | `min(25, 5 * structural_pattern_instances + sum(style_tell_weights))` — count each matched instance of a row marked Structural in the Structural & Semantic Patterns table (5 pts per instance, not per pattern type); style-level tells (random bolding, one-sentence paragraphs, etc.) use variable weights from `reference.md` |
| Semantic | 20 | `min(20, 5 * semantic_pattern_instances)` — count each matched instance of a row marked Semantic in the Structural & Semantic Patterns table (5 pts per instance, not per pattern type); each instance scores once, on its stated dimension only |
| Stylometric | 15 | `min(15, 5 * stylometric_flags)` |

**Total:** Sum of dimensions, capped at 100. Count each occurrence once, even when it matches more than one dictionary entry or pattern row (see *One occurrence, one hit* in reference.md).

### Score Interpretation

| Score | Interpretation |
|-------|----------------|
| 0-20 | Clean: minimal AI patterns detected |
| 21-40 | Light: some patterns, minor editing needed |
| 41-60 | Moderate: noticeable AI fingerprint, edit recommended |
| 61-80 | Heavy: significant slop, substantial rewrite needed |
| 81-100 | Severe: text reads as unedited AI output |

## Stylometric Thresholds

Based on StyloAI (Opara, 2024) and Desaire et al. (2023) research.

| Metric | Flag If | Target |
|--------|---------|--------|
| Sentence length σ | σ < 15.0 | σ > 15.0 |
| Paragraph length SD | SD < 25 | SD > 25 |
| Type-Token Ratio | TTR < 0.50 or TTR > 0.70 | 0.50 ≤ TTR ≤ 0.70 |
| Hapax legomena rate | Below user baseline | At or above baseline (default when no user baseline: flag if rate < 40%) |

## Structural & Semantic Patterns (capped by the Structural 25 + Semantic 20 dimension maxima)

Each pattern below scores +5 on its stated dimension.

| Pattern | Description | Dimension |
|---------|-------------|-----------|
| Formulaic Introduction | Rephrasing topic → importance → overview | Structural |
| Template Sections | Overview → Key Points → Best Practices → Conclusion | Structural |
| Over-Signposting | "In this section...", "As mentioned earlier..." | Structural |
| Staccato Paragraphs | >50% are 1-2 sentences | Structural |
| Symmetric Coverage | Equal weight to all options without prioritization | Structural |
| Hollow Specificity | "Many companies have seen improvements" (which?) | Semantic |
| Absent Constraints | Absolute claims without limitations | Semantic |
| Balanced to a Fault | Every pro has matching con of equal weight | Semantic |
| Circular Reasoning | Rephrases thesis without new evidence | Semantic |
| Structural Contrast | "It's not X; it's Y" and subject-varying forms ("The goal isn't X; it's Y"), elevation forms ("not just X — it's Y", "not merely X — it's Y"), plus hedged concessions ("a minor X, but a real one") (see Cat. 9 in reference.md) | Structural |
| Copula Avoidance | "stands as" or "serves as" in place of a plain "is" that says the same thing (see Cat. 2 in reference.md); scores here, not also as a Lexical hit | Structural |
| Gerund-Tail Commentary | A factual clause followed by ", highlighting...", ", underscoring...", or ", reinforcing..." that adds no new evidence ("The job retried three times, highlighting the importance of idempotency"); exempt when the gerund clause names a new concrete detail (metric, file, mechanism) | Structural |
| Framework Name-Dropping | Framework invoked with no concrete claim attached (see Semantic Fabrication in reference.md) | Semantic |
| Fabricated Open Questions | "Open questions"/"next steps" invented for closed or decided topics | Semantic |
| Process Metrics as Results | Activity/funnel counts standing in for the actual outcome | Semantic |
| Verification Theater | Completion or certainty claims ("all tests pass", "no regressions", "fully tested") with no linked command, CI run, or test name (see Semantic Fabrication in reference.md) | Semantic |

**Cap behavior:** 8 rows above are tagged Structural; that dimension saturates at 25 points once any 5 of them are found (5 × 5 = 25), before style-tell weights are even added — same ceiling logic as below. 8 rows above are tagged Semantic; the dimension saturates at 20 points once any 4 of them are found (4 × 5 = 20) — this is a scoring ceiling, not a count of how many Semantic patterns exist. Fabrication findings (framework name-dropping, fabricated open questions, process metrics as results, verification theater) are factual defects, not style defects: always list them in Top Offenders even when the dimension is already capped.

## Pattern Category Quick Reference

For the complete pattern dictionary, see [reference.md](./reference.md). **Dimension** shows which scoring bucket each category feeds (see Scoring Algorithm above) — this table spans Lexical, Structural, and Semantic, not Lexical alone (Stylometric patterns are measured directly, not via this category dictionary).

| Category | Examples | Dimension | Action |
|----------|----------|-----------|--------|
| Generic Boosters | incredibly, extremely, very | Lexical | Delete or replace with metrics |
| Buzzwords | robust, seamless, leverage, elevate, harness, pivotal, impactful | Lexical | Replace with plain language |
| Filler Phrases | "It's important to note that", "In today's ever-evolving world" | Lexical | Delete entirely |
| Hedge Patterns | of course, arguably, seems to | Lexical | Commit or remove |
| Sycophancy | "Great question!", "Happy to help!" | Lexical | Delete |
| Transitional Filler | Furthermore, Moreover, Additionally, However, Indeed | Lexical | Use sparingly or cut |
| Vague Abstraction | the frame, the lens, the narrative, the space | Lexical | Replace with the specific noun |
| Structural Contrasts | "It's not X; it's Y", "[Subject] isn't X; it's Y", "not just X — it's Y", "not merely X — it's Y" | Structural | State Y directly; drop the negation frame |
| Style Tells | one-sentence paragraphs, random bolding, abstract noun stacking | Structural | Restructure |
| Typographic Tells | em-dash (—), smart quotes | Lexical | Replace with standard punctuation (each em-dash scores +3 pts, not +2). En-dash (–) is legitimate punctuation and is never flagged |
| Clichés | state of the art, at the end of the day, paradigm shift, move the needle, think outside the box (see reference.md Cat. 10 for full list) | Lexical | Replace with the specific claim. Score once per instance; if a phrase also matches a pattern in another Lexical category (e.g. "at the end of the day" also matching Filler Phrases Cat. 3, or "state of the art" also matching Buzzwords Cat. 2), the earlier-numbered category takes precedence, do not double-count. |
| AI Jargon | failure mode, failure class, failure pattern, failure category, error class, defect pattern (singular and plural; full list and exemptions in reference.md Cat. 11) in free-running prose | Lexical | Flag at 2 pts per instance. See reference.md Cat. 11 for full exemption rules; do not flag structural section contracts, external quotations, code/API contexts, or a concrete singular use that names how the system fails (a process lapse does not count). Default replacement: name the actual problem. Full replacement guidance in eliminating-ai-slop. |
| Semantic Fabrication | framework name-dropping, fabricated open questions, process metrics as results, verification theater | Semantic | Ground in a source or delete |
| Resurrected Corrected Claims | reintroducing a phrasing the author already struck earlier in the document/session | Semantic (unscored — requires session context, no scoring-table row) | Sweep prior corrections before each edit pass |

## Dictionary Integration

This skill reads from `.slop-dictionary.json` if present in workspace root.

- Custom patterns are included in detection
- Exceptions are skipped during detection
- Weight affects scoring: `score = base_score * weight`

**Note:** This skill reads from the dictionary but does not write. Use `eliminating-ai-slop` to add patterns or exceptions.

## Semantic Quick Tests

Use these when reviewing AI text qualitatively (merged from `reviewing-ai-text`):

| Test | Slop Signal | Real Signal |
|------|-------------|-------------|
| **Specificity** | "Use appropriate caching strategies" | "Use Redis with 5-minute TTL for session data" |
| **Asymmetry** | "Both options have merits" | "Use Postgres unless >10M writes/day" |
| **Constraint** | "Implement microservices for scalability" | "Microservices add 3x ops overhead. Stay monolith unless dedicated platform team." |
| **First-Person** | Generic enough to apply anywhere | Grounded in specific context |

## Companion Skills

- **eliminating-ai-slop**: Active rewriting to remove detected patterns
- **professional-language-audit**: Profanity and inappropriate language detection
- **readme-authoring**: README generation
- **incorporating-research**: Score research quality before incorporating

## Example

```bash
# Score text for AI patterns (read-only analysis)
echo "Check for: hedging ('It is worth noting'), filler ('In order to'),
  superlatives ('incredibly powerful'), and vague claims ('comprehensive')"
```

## Failure Modes

- **False positives on domain jargon:** Flagging legitimate technical terms (e.g., "robust" in a load-testing context) as slop
- **Score inflation:** Giving a passing score to text with subtle but pervasive AI patterns
- **Detection without action:** Scoring text as sloppy but not invoking `eliminating-ai-slop` to fix it