Language and Grammar Patterns (7-13) in Humanizer: Detecting AI-Writing Tells

Pattern 7 ("Repeated sentence openings") targets prose where consecutive sentences start with identical subjects or phrases—a rhythmic habit of language models that Humanizer collapses into varied, natural sentence structures.

Humanizer is an open-source tool that identifies and rewrites recognizable patterns of AI-generated text. The project, hosted at blader/humanizer, defines 13 distinct language and grammar patterns that distinguish mechanical prose from human writing. This article covers patterns 7 through 13, all detailed in the repository's SKILL.md file and referenced in README.md.

Pattern 7: Repeated Sentence Openings

Pattern 7 belongs to Humanizer's "Rhythm by rule" category. It flags sequences where multiple sentences begin with the same subject—typically she, he, or identical phrases.

The Detection Logic

According to SKILL.md lines 51-58, the rule watches for:

• Sentence series opening with identical subjects
• Mechanical repetition where a human writer would vary structure
• Unintentional mirroring rather than deliberate stylistic choice (e.g., "She came. She saw. She conquered." is intentional and preserved)

The Rewrite Strategy

Humanizer advises three transformations: merge sentences, change the subject, or begin with the action. The tool does not ban the repeated word entirely—remaining sentences may still start with the same term when contextually appropriate.

# Example transformation captured in SKILL.md

# Before:

She noted the door. She noted the lock on it. She filed both away.

# After Humanizer rewrite:

She noted the door and its lock, then filed both away.

Additional documented examples:

# Before:

He opened the file. He read the contents. He saved the changes.

# After Humanizer rewrite:

He opened the file, read the contents, and saved the changes.

# Before:

She arrived early. She prepared the room. She greeted the guests.

# After Humanizer rewrite:

She arrived early, prepared the room, and greeted the guests.

Pattern 8: List Without a Point

Pattern 8 identifies information-dumping lists that lack argumentative purpose. AI models frequently enumerate details without framing why they matter.

• Watch for – disconnected facts presented as bulleted or serial information
• Problem – readers cannot determine the list's relevance to the surrounding argument
• Rewrite – either delete the list, convert selected items into evidence for a specific claim, or add explicit framing that establishes the point

Pattern 9: Thesaurus Substitution

Pattern 9 catches inappropriate synonym swaps where a language model replaces a precise term with a near-equivalent that subtly shifts meaning.

• Watch for – words that sound more sophisticated but fit awkwardly in context
• Problem – precision loss; the substituted term carries connotations or denotations that distort the intended message
• Rewrite – restore the original, contextually precise term unless the substitution genuinely improves clarity

Pattern 10: Generic Transition

Pattern 10 targets rote transitional phrases applied without structural necessity—another rhythm-by-rule artifact.

Common flagged openings include Furthermore, Moreover, Additionally, and In conclusion when they connect ideas that follow logically without explicit signaling. The rule preserves transitions that genuinely clarify relationships, but strips those that merely pad word count or simulate academic register.

Pattern 11: Hedge Stacking

Pattern 11 detects excessive qualifying language that undermines confident statements.

• Watch for – accumulations of somewhat, rather, quite, fairly, relatively, generally, typically, often, tends to, can be, may be, might, could potentially
• Problem – layered hedges create vague, non-committal prose that reads as algorithmically cautious
• Rewrite – remove redundant hedges; retain single qualifiers only when genuine uncertainty or scope limitation is intended

Pattern 12: Abstract Modifier Bloat

Pattern 12 flags strings of nominalizations and abstract modifiers that distance action from actors.

AI writing frequently produces constructions like "The implementation of the optimization of the system resulted in an improvement in performance" rather than "We optimized the system and improved performance."

The rewrite converts nominalizations to verbs and restores agents where contextually appropriate.

Pattern 13: Conclusion Without Synthesis

Pattern 13 identifies section or document conclusions that merely restate preceding points rather than synthesizing them into new insight.

• Watch for – summary paragraphs beginning with In summary, To conclude, Overall that repeat claims already made
• Problem – wasted vertical space; missed opportunity to advance the argument
• Rewrite – either delete the pseudo-conclusion or transform it to identify implications, limitations, or connections not previously articulated

Implementation in the Humanizer Codebase

All seven patterns are defined in SKILL.md at the repository root. The file uses a consistent schema:

1. Numbered section header (e.g., ### 7. Repeated sentence openings)

2. Problem statement explaining the AI-writing tell

3. Watch list of specific indicators

4. Rewrite guidance with explicit do's and don'ts

5. Before/after examples demonstrating transformation

The README.md file provides pattern inventory and usage instructions but defers to SKILL.md for complete specifications.

Humanizer consumes these pattern definitions through its core detection engine. While the repository does not expose pattern 7-13 logic as individually callable functions, the SKILL.md structures serve as the authoritative specification that both human editors and automated tooling reference.

Summary

• Pattern 7 (Repeated sentence openings) – collapses redundant subject-led sentences into varied structures
• Pattern 8 (List without a point) – demands argumentative framing for information dumps
• Pattern 9 (Thesaurus substitution) – restores precision over pseudo-sophistication
• Pattern 10 (Generic transition) – strips mechanical transitional padding
• Pattern 11 (Hedge stacking) – eliminates redundant qualification
• Pattern 12 (Abstract modifier bloat) – nominalization-to-verb conversion
• Pattern 13 (Conclusion without synthesis) – advances restatement to genuine synthesis

Each pattern in the 7-13 range shares a core principle with patterns 1-6: identifying rule-governed rhythm in AI output and replacing it with ear-governed variation.

Frequently Asked Questions

How does Pattern 7 differ from intentional stylistic repetition?

Pattern 7 explicitly excludes deliberate rhetorical repetition. The SKILL.md documentation cites "She came. She saw. She conquered" as protected repetition—parallel structure serving stylistic purpose. The rule targets only unintentional mechanical mirroring where variation would improve prose.

Can Humanizer automatically apply these pattern rewrites?

The repository provides pattern definitions in SKILL.md as specifications rather than automated transformations. Implementation requires integration with the detection engine or manual application of the documented rewrite strategies. The before/after examples demonstrate intended outcomes rather than API-callable functions.

Where are patterns 7-13 documented in the source code?

Primary documentation resides in SKILL.md lines 51-58 (Pattern 7) and subsequent sections through line 130+ (Patterns 8-13). The README.md offers overview tables but defers to SKILL.md for complete specifications including watch lists, problem statements, and rewrite guidance.

Do patterns 7-13 overlap with earlier patterns 1-6?

Patterns 7-13 continue the "Rhythm by rule" methodology established in patterns 1-6 but target distinct AI-writing tells. Pattern 7 shares thematic territory with structural repetition patterns (1-3) but focuses specifically on sentence-initial subject repetition rather than broader structural mimicry.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}