← Files HumanWritingARCHIVED FILE

skills/human-writing/references/natural-writing-checklist.md

5.94 KB · Oct 3, 2026 · 06:30 UTC

↓ Download file

# Natural-writing diagnostic checklist

This checklist adapts observations from Wikipedia's “Signs of AI writing”
advice page:

https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing

Source reviewed July 26, 2026. The page is a descriptive field guide, not a
prescriptive style guide. Its examples come largely from Wikipedia, so apply
only the patterns that make sense for the current genre.

Wikipedia text is available under the Creative Commons
Attribution-ShareAlike 4.0 International license:

https://creativecommons.org/licenses/by-sa/4.0/

This checklist modifies and expands the source material for a general writing
workflow.

## How to use this checklist

Review a draft for clusters of patterns. Do not ban a word or construction
merely because it appears below. A single phrase proves nothing, and awkward
attempts to hide every possible signal usually make the prose worse.

## Content patterns

### Inflated significance

Check for unsupported claims that an ordinary fact:

- marks a pivotal moment;
- reflects a broader shift;
- underscores lasting importance;
- leaves an enduring legacy;
- shapes an evolving landscape; or
- stands as a testament to something larger.

Keep such interpretation only when it is the writer's real point and the
evidence supports it. Otherwise, state the fact and let the reader judge.

### Superficial analysis

Look for an `-ing` phrase attached to a factual sentence to manufacture
importance: “highlighting,” “ensuring,” “reflecting,” “symbolizing,”
“contributing,” “fostering,” or “enhancing.” Replace it with a supported causal
statement or delete it.

### Promotional gloss

Question travel-brochure and press-release language such as “vibrant,” “rich,”
“renowned,” “groundbreaking,” “nestled,” “showcases,” “boasts,” or
“commitment to.” Use evaluative language only when the genre calls for it and
the evaluation is defensible.

### Vague authority

Replace phrases such as “experts believe,” “critics argue,” “observers note,”
or “industry reports suggest” with a named, relevant source. If no source is
available, remove or clearly label the writer's own judgment.

### Template-shaped argument

Watch for:

- a generic “challenges and future prospects” ending;
- a final paragraph that recites the headings;
- a forced optimistic outlook;
- boilerplate about balancing opportunities and risks; or
- a lead that awkwardly defines the title rather than addressing the topic.

End when the message, argument, or story has reached its natural stopping point.

## Language and syntax patterns

### Dense “AI vocabulary”

Words often seen in formulaic output include “additionally,” “align,”
“bolster,” “crucial,” “delve,” “enduring,” “enhance,” “foster,” “garner,”
“highlight,” “interplay,” “intricate,” “key,” “landscape,” “meticulous,”
“pivotal,” “robust,” “showcase,” “tapestry,” “testament,” “underscore,”
“valuable,” and “vibrant.”

These words are not forbidden. Replace them when a shorter, more exact word
would sound more natural in context.

### Elaborate substitutes for simple verbs

Prefer a simple copula or verb when accurate:

- “is” instead of “serves as” or “stands as”;
- “has” instead of “boasts,” “features,” or “offers”;
- “wrote” instead of “authored”;
- “used” instead of “utilized”;
- “moved” instead of “relocated”; and
- “tried” instead of “attempted.”

Retain the longer form when it carries a real distinction.

### Repeated rhetorical frames

Check for too many:

- “not only X, but also Y” constructions;
- “not X, but Y” reversals;
- “X rather than Y” contrasts;
- three-adjective strings;
- three-part lists that imply false completeness; or
- parallel clauses used mainly for drama.

Keep the structure when it sharpens the thought. Change it when it is merely a
default rhythm.

### Elegant variation

Do not cycle through synonyms just to avoid repeating the correct noun. Stable
terms improve clarity, especially in technical, legal, and executive writing.

## Formatting patterns

Review whether the draft overuses:

- title case in headings;
- bold labels on every bullet;
- vertical lists where a sentence would be easier;
- em dashes for routine pauses;
- emoji as structural markers;
- small tables with little comparative value;
- thematic breaks between every section; or
- many headings for a short piece.

Match formatting to the medium. An operating plan may need bullets and a table;
a personal email usually does not.

## Chatbot residue

Remove text addressed to the requester when it does not belong in the
deliverable, including:

- “Of course” or “Certainly”;
- “Here is a detailed breakdown”;
- “I hope this helps”;
- “Would you like me to...”;
- placeholder instructions;
- references to the prompt or supplied materials;
- knowledge-cutoff boilerplate; and
- speculation that missing information is private, scarce, or unavailable.

State an actual limitation briefly when it matters. Do not speculate beyond the
evidence.

## Positive human-writing cues

Natural prose often includes:

- simple `is` and `has` constructions;
- ordinary verbs;
- a mix of direct statements and justified qualifiers;
- occasional wordiness where the speaker naturally emphasizes something;
- asymmetry, with more space given to what actually matters;
- specific details that could not be pasted into an unrelated document; and
- choices the writer can explain.

Do not add typos, grammatical errors, slang, fake personal details, arbitrary
sentence fragments, or random punctuation. Those are disguises, not voice.

## Limits of detection

Writing style cannot establish authorship. Human writers use many of these
patterns, and AI detectors have meaningful false-positive and false-negative
rates. Judge the draft by clarity, specificity, truthfulness, and fit for its
audience—not by a detector score.

SHA-256: 818370cef4d9809b4aa59009647ee6ea5df34a606d208995d9bd57774b0db254