skills/write-like-me/references/ai-texture-catalogue.md

# AI Texture Catalogue

Use this as writing hygiene, not an AI detector. A pattern is evidence of weak or generic writing only in context. Do not accuse the user of AI authorship.

No ordinary word, punctuation mark, sentence shape, or rhetorical move is banned outright. Flag a pattern only when it weakens this draft through emptiness, repetition, false drama, vagueness, or mismatch with the writer's purpose. Preserve deliberate usage and distinctive voice.

## Triage before rewriting

Classify each weak passage.

### Rewordable

The passage has a real claim but generic expression. Preserve the claim and replace abstraction with an actor, behaviour, decision, cost, mechanism, or consequence.

### Hollow

The passage has no usable claim. Ask for the missing point when it is central. Delete it when it is disposable. Do not hide it behind better phrasing.

### Unsupported

The passage introduces a fact, statistic, quotation, attribution, credential, personal story, or lived experience that the supplied material does not support. Remove or flag it. Never make it more plausible by adding detail.

## High-value signals

### Generic framing

- `In today's fast-paced world`
- `In a rapidly evolving landscape`
- `Now more than ever`
- `It is important to note`
- `This underscores the need`
- `At the intersection of`
- `The future belongs to`
- staged honesty such as `Let me be completely clear` when nothing personal follows;
- lone-expert framing such as `The thing everyone overlooks` when the claim can stand unaided;
- rhetorical staging such as `Imagine if...` or a self-answered question that only delays the point.

Start with the actual tension, claim, scene, or consequence.

### Inflated and abstract language

Watch for clusters of words such as transformation, innovation, alignment, enablement, leverage, impact, framework, ecosystem, robust, seamless, pivotal, transformative, and game-changing.

One word may be right. A cluster often hides who does what and what changes.

Named risks:

- **Importance inflation:** announcing that an ordinary event is historic, pivotal, vital, or proof of excellence instead of showing its consequence.
- **Vague authority:** using unnamed experts, reports, research, or consensus. Name the source, qualify the claim, or ask for evidence.
- **Inflated verb phrase:** replacing `is`, `has`, `does`, or a precise action with ceremonial language. Prefer the clearest accurate verb, not automatically the shortest one.
- **Precision theatre:** inserting a mechanism, number, or example that the source never supplied. Specificity must come from evidence, not invention.

### Formulaic movement

- repeated `not just X, but Y` or `less X, more Y` constructions;
- false concessions where both sides are vague;
- repeated rule-of-three lists;
- rhetorical questions that merely delay the answer;
- obvious signposting such as `Zooming out`, `The key takeaway`, or `What leaders should do next`;
- a conclusion that restates every paragraph without adding a landing.
- **Colon reveal:** a setup phrase followed by a colon and a supposedly dramatic lowercase revelation. Keep colons for genuine lists, labels, quotations, or syntactic clarity.
- **Trailing pseudo-analysis:** an `-ing` clause that gestures at significance without explaining a mechanism or consequence.
- **Negative ladder:** several fragments saying what something is not before finally saying what it is.
- **Synonym rotation:** renaming the same actor or object in successive sentences merely to avoid repetition. Repeat the accurate term when clarity benefits.

Keep a strong contrast when it earns its place. Remove repetition, not every rhetorical device.

### Over-structured rhythm

- every paragraph has nearly the same length;
- every sentence has similar cadence;
- headings or bullets make thin ideas look complete;
- short social writing is broken into one-line paragraphs by default;
- body paragraphs can be rearranged without changing the argument.
- **Dramatic fragment stack:** multiple clipped fragments or repeated conjunction openings used to manufacture intensity.
- **Mechanical symmetry:** several adjacent sentences or sections share the same grammatical skeleton even though the ideas do not require it.

Preserve useful structure while restoring dependency between ideas and natural variation.

### Structural defaults

Use these only when the draft's underlying decisions, not merely its wording, weaken the piece. Read [authorial decisions contract](authorial-decisions-contract.md) before changing structure.

- **Thematic over-determination:** the draft demonstrates an implication, states it, and then explains the lesson again instead of trusting the reader where the purpose allows inference.
- **Single-track movement:** every paragraph advances one frictionless line, with no supplied uncertainty, counterpressure, competing cause, or meaningful dependency.
- **Insight-resolution default:** a tension ends in personal understanding, acceptance, or a neat lesson even though the source does not support that resolution.
- **Performed embodiment:** generic bodily sensations, sensory detail, or setting-as-emotion substitutes for a supported event, feeling, or consequence.
- **Vague world contact:** diffuse references to books, research, culture, places, or experience replace a supplied named source or conceal that no source was provided.
- **Unearned closure:** the ending resolves uncertainty, moral tension, or causality more completely than the evidence permits.

Do not reverse these patterns mechanically. Do not add disorder, counterarguments, time jumps, brands, direct address, ambiguity, roughness, or an open ending merely to look less like AI writing.

### Performed humanity

- unsupported `Let me be honest` or `I have been thinking a lot about`;
- generic emotional claims without an event or consequence;
- trendy relatability such as `this hit different` without earned detail;
- confident first-person opinions added when the user supplied none;
- fake vulnerability, struggle, clients, meetings, lessons, or memories.

Personal writing must come from the user, not from a humanizing technique.

### Generic endings

- `Thoughts?`
- `Agree?`
- `What do you think?`
- `Let me know in the comments`
- `Those who embrace this shift will thrive`
- **Manufactured profundity:** a final metaphor, aphorism, or mic-drop line that is broader and grander than the evidence.
- **Recap ending:** a last paragraph that merely repeats the argument the reader just finished.

End with the actual implication, a useful next move, a grounded observation, or a question specific to the argument.

### Formatting tells

- long dashes or double hyphens used as a default rhythm device;
- excessive bolding;
- emoji headings without a real reason;
- excessive bullets for prose thinking.

Apply the long-dash authority order in [output contracts](output-contracts.md). Do not treat a long dash as evidence of weak writing by itself or strip punctuation that belongs to quoted/source text.

### Answer-wrapper leakage

- `Here is a more human version`;
- `I have matched your voice below`;
- routine explanations of what was changed when the user asked only for paste-ready prose;
- generic offers to revise, shorten, or adjust the writing after the requested piece.

Keep process commentary outside the requested writing and omit it by default. Retain a note only when the user requests explanation or an unresolved semantic, factual, or evidence limitation genuinely needs attention.

## Context safeguards

- Do not rewrite quoted text, code, cited examples, or deliberate demonstrations of bad writing unless asked.
- Preserve technical terms when they are the accurate terms.
- Allow more structure in documentation and less in personal or social prose.
- Preserve a deliberate rhetorical question, triad, or unusual word when it works.
- Do not replace the user's natural irregularity with forced slang, fragments, or opinions.
- Do not use a thesaurus to manufacture vocabulary diversity.
- Do not manufacture structural diversity. A linear argument, explicit conclusion, or tidy resolution may be correct for the task.

## Second pass

After rewriting, check:

1. Did the meaning, polarity, and uncertainty survive?
2. Did any unsupported personal or factual material appear?
3. Is the result specific, or merely smoother?
4. Did one AI formula survive in several forms?
5. Did the repair flatten the user's strongest wording?
6. Does the ending belong to this argument?
7. Did the writer make the central decisions, or did the edit merely disguise a generic structure?

## Influence and attribution

The named-pattern audit and minimum-effective-edit emphasis were strengthened after reviewing Peter Yang's MIT-licensed [No AI Slop](https://github.com/petergyang/no-ai-slop) skill. The distinction between surface texture and underlying authorial decisions was sharpened after reviewing Russell et al.'s [StoryScope](https://arxiv.org/abs/2604.03136) research on long-form fiction. Write Like Me independently expresses and contextualises these ideas, rejects blanket vocabulary and structural bans, and adds its own evidence model, portable personal profile, correction learning, sample isolation, and deterministic rewrite verification.