Pattern Catalog

Pattern Catalog

The public AI Writing Patterns 1–220 document is an editorial taxonomy. It gives writers and contributors shared language for patterns that can make prose feel generic, inflated, evasive, or over-structured.

Read it as a judgment aid

The catalog is not a detector, a plagiarism system, or proof that a person used AI. Its examples show patterns worth inspecting, but each entry includes a legitimate exception because a phrase can be appropriate in a specific voice, genre, or factual context.

Use this five-part test before editing a flagged pattern:

1. Pattern: Is the named shape actually present?
2. Function: What is it doing in this specific sentence?
3. Cost: Does it make meaning less direct, precise, or human?
4. Exception: Is there a real contextual reason to retain it?
5. Repair: Can a smaller, clearer change improve the line?

If the cost is not real, leave the sentence alone.

Catalog versus executable rules

The catalog intentionally exceeds the CLI ruleset. It is broader because editorial thinking needs nuance; the executable ruleset is smaller because automation needs reliability.

Layer	Purpose	Scope
Catalog	Review, training, diagnosis, and shared vocabulary	220 documented patterns, including exceptions and escalation guidance
AI Editor rules	Local deterministic findings and scores	Only the patterns visible through npx holdyourvoice patterns

This boundary is non-negotiable. Do not say the tool detects all 220 patterns unless the executable rules and tests actually do.

Using the catalog in a review

Choose an entry only after you have noticed an actual sentence-level problem. Read its spotting test and exception. State the edit in concrete terms—what it changes and why—rather than saying the copy “sounds AI.” Keep the writer’s chosen roughness if it carries voice, and never use a catalog entry to erase a deliberate style choice without a real reader benefit.

Contributing to the taxonomy

Public additions must be generic, independently useful, and free of client examples, private corpus material, or proprietary training data. Preserve the catalog’s ordered numbering and write the exception as carefully as the diagnosis. A catalog addition does not automatically authorize a code rule.

For the full provenance and safety boundary, read the repository’s pattern taxonomy document.