SuperAPI AI Directory

human-write

Writing & research

View repository

Updated 10-15 22:14

About this skill

Human Write

Asking AI to help polish your writing? Avoid the following words!

Based on two 2025 studies analyzing 27.5 million PubMed articles, identifying English words favored by AI.

Trigger Timing

  • The user asks you to polish an English article
  • The user asks you to write an English academic/professional article
  • The user wants to check whether an article contains AI-sounding English words
  • The user says "human-write" or "scan AI words"
  • For Chinese writing, please refer to chinese-patterns.md

Top 10 Most Severe AI Words

Based on two 2025 studies analyzing 27.5 million PubMed articles, identifying English words favored by AI.

Top 10 Most Severe AI Words

After writing or polishing, search these 10 first. If any appear, think twice:

Rank Word Why suspicious Replacement direction
1 delve Highest 2024 Z-score, AI's favorite explore, examine, look into
2 underscore AI uses it to replace emphasize emphasize, highlight, stress
3 meticulous AI's favorite adjective careful, thorough, precise
4 intricate AI uses it to replace complex complex, detailed, elaborate
5 tapestry AI's favorite metaphorical term mix, combination, array
6 commendable AI's praise term impressive, notable, solid
7 showcase AI uses it to replace show/demonstrate show, demonstrate, present
8 pivotal AI uses it to replace important/key key, important, central
9 bolster AI uses it to replace support/strengthen support, strengthen, back up
10 surpass AI uses it to replace exceed/outperform exceed, beat, outperform

Complete AI-preferred vocabulary list

AI preference More natural alternative
delve (into) explore, look into, examine
underscore emphasize, highlight, stress
showcase show, demonstrate, present
bolster support, strengthen, back up
elucidate explain, clarify, make clear
garner gain, attract, earn, get
leverage use, take advantage of, build on
unveil reveal, introduce, present
scrutinize examine, inspect, review
surpass exceed, beat, outperform
boast have, feature, offer
foster encourage, promote, support
facilitate help, enable, make easier
harness use, apply, put to use
navigate handle, manage, deal with
streamline simplify, improve, speed up
illuminate show, reveal, shed light on
spearhead lead, drive, champion
embark (on) start, begin, launch
endeavor try, attempt, work toward
encompass include, cover, span
catalyze trigger, spark, drive
augment add to, increase, supplement
delineate describe, outline, define
juxtapose compare, contrast, set side by side
transcend go beyond, rise above, exceed
AI preference More natural alternative
meticulous careful, thorough, precise
intricate complex, detailed, elaborate
pivotal key, important, central, crucial
commendable impressive, notable, solid, good
groundbreaking new, innovative, first-of-its-kind
transformative significant, game-changing, major
nuanced subtle, complex, layered
comprehensive full, complete, thorough
robust strong, solid, reliable
multifaceted complex, varied, diverse
noteworthy notable, worth noting, significant
paramount most important, top, critical
holistic overall, whole, complete
overarching main, broad, general
unprecedented never before seen, first, new
salient key, main, important
burgeoning growing, rising, expanding
AI preference More natural alternative
primarily mainly, mostly, largely
notably especially, in particular
predominantly mostly, mainly, largely
compellingly convincingly, strongly
profoundly deeply, greatly, strongly
moreover also, besides, in addition
furthermore also, and, in addition
importantly (often deletable — just state it)
crucially (often deletable — just state it)
remarkably surprisingly, strikingly
substantially greatly, significantly, a lot
inherently by nature, naturally, fundamentally
AI preference More natural alternative
tapestry mix, combination, range
realm area, field, domain
landscape field, area, scene, picture
synergy teamwork, cooperation, combined effect
paradigm model, approach, framework
nexus link, connection, center
interplay interaction, relationship
cornerstone foundation, basis, key part
underpinning basis, foundation, support
discourse discussion, debate, conversation
trajectory path, direction, trend
myriad many, numerous, a wide range of
plethora many, plenty, a lot of

Usage

Mode 1: Automatically avoid when polishing

When polishing English for the user,proactivelyavoid all the above words. If the original text does not contain these words, they should not appear after polishing either.

Rules:

  1. Do not add AI words that were not in the original text — polishing is improvement, not embellishment
  2. Replace AI words already in the original text — use more natural alternatives, but ask the user first
  3. Preserve the user's voice — if the user already likes to use a certain word, do not replace it for them

Mode 2: Scan detection

Use a built-in script to scan the article and generate an AI vocabulary density report:

python3 ~/.claude/skills/human-write/scan-ai-words.py <file>
# 或從 stdin
cat essay.txt | python3 ~/.claude/skills/human-write/scan-ai-words.py

Output format

═══ AI Word Scan Report ═══

Score: 3.2 / 10  (moderate AI flavor)

Found 8 AI-flavored words in 1,200 words (0.67% density)

  TIER 1 (top flagged):
    delve ×2  (lines 5, 23)
    meticulous ×1  (line 14)

  TIER 2 (commonly flagged):
    comprehensive ×3  (lines 8, 31, 45)
    furthermore ×2  (lines 12, 38)

Suggestions:
  Line 5:  "delve into the mechanisms" → try "explore the mechanisms"
  Line 14: "meticulous analysis" → try "careful analysis"
  ...

Scoring criteria

Score Density Interpretation
0-2 < 0.2% Natural, almost no AI traces
3-4 0.2%-0.5% Slight, a small replacement is enough
5-6 0.5%-1.0% Moderate, recommend checking one by one
7-8 1.0%-2.0% Obvious, requires extensive rewriting
9-10 > 2.0% Strong AI flavor, rewrite it

Research background

Two key papers

  1. Matsui K. (2025). Delving Into PubMed Records: How AI-Influenced Vocabulary has Transformed Medical Writing since ChatGPT. Perspectives on Medical Education, 14(1), 882-890.
  2. Kobak, D., et al. (2025). Delving into LLM-assisted writing in biomedical publications through excess vocabulary. Science Advances, 11(27), eadt3813.

Core findings

  • Analyzed 27.5 million PubMed records (2000-2024)
  • Of 135 AI-preferred words, 103 showed a statistically significant surge in 2024
  • The trend started from 2020(two years before ChatGPT's release)
  • Most of the 84 standard academic terms in the control group remained stable

Two-phase pattern

  • First phase (2020-2022): Early AI tools such as DeepL and Grammarly began to influence word choice
  • Second stage (2023-present): The ChatGPT explosion, and AI vocabulary was amplified on a large scale (β = 0.655, p < 0.001)

Non-native English-speaking researchers

  • In papers from China, South Korea, Taiwan, and elsewhere, the proportion of AI vocabulary use is about 20%
  • English-speaking countries about 5%
  • This does not mean "improper use," but it does mean more attention needs to be paid to post-review editing

Important reminder

  • The "appearance" of vocabulary does not necessarily mean it was definitely written by AI
  • The key isdensity: It is normal for humans to occasionally use delve, but using it in every paragraph is not normal
  • AI did not "create" these usages, but greatly "amplified" them
  • At the same time, some traditional academic terms have instead decreased: purpose of, hypothesis, results suggest

Notes

  • This list is based onmedical literatureresearch, but AI-preferred vocabulary is common across fields
  • It is not saying these words "cannot be used," but rather that before using them you should confirm it is your own choice
  • It is entirely reasonable for non-native English speakers to use AI for polishing, as long as they edit it themselves one final time
  • Human writing has "bumps, dents, obsessions, and flaws"; these are features, not shortcomings
  • AI is a drafting machine, not a ghostwriter