DETAILED_PATTERNS.md

Detailed AI-Writing Patterns

All of these are drawn from Wikipedia:Signs of AI writing, with additions from anti-artificial style guides. They are the things to strip or replace when humanizing text.

1. Significance / Legacy / Broader Trends

AI templates:

  • "stands/serves as" → "is"
  • "is a testament/reminder" → delete or "shows"
  • "a vital/significant/crucial/pivotal/key role/moment" → be specific or delete
  • "underscores/highlights its importance/significance" → delete
  • "reflects broader" → delete
  • "symbolizing its ongoing/enduring/lasting" → delete
  • "contributing to the" → delete
  • "setting the stage for" → "before" or delete
  • "marking/shaping the" → "when" or "this"
  • "represents/marks a shift" → "became" or "then"
  • "key turning point" → delete
  • "evolving landscape" → delete
  • "focal point" → "center" or "where people went"
  • "indelible mark" → delete
  • "deeply rooted" → "old" or "long-standing"
  • "founding/establishment represented a significant shift" → "then"
  • "part of a broader movement" → delete
  • "More than just [category]" → delete the framing, describe the thing directly (e.g., "Python is more than just a language" → "Python is a language with a strong community")

AI also overstates ecosystem/environment connections for biology topics and conservation status even when unknown.

2. Canned Notability / Attribution / Media Coverage

Templates:

  • "independent coverage" → name the source
  • "local/regional/national/[country] media outlets" → name outlets
  • "music/business/tech outlets" → name them
  • "profiled in" → "was in" or "[Publication] wrote about"
  • "written by a leading expert" → name them or delete
  • "active social media presence" → "posts on X/Instagram" or delete

AI tends to list coverage instead of saying what the coverage said. Summarize the content, not the existence of the article.

3. Superficial Analyses

Participial phrases appended to sentences:

  • "... , creating a lively community."
  • "... , further enhancing its significance."
  • "... , contributing to the socio-economic development."
  • "... , reflecting its continued relevance."
  • "... , highlighting their historical significance."
  • "... , demonstrating the ongoing relevance."
  • "... , illustrating its lasting influence."
  • "... , emphasizing their critical view."

Recast: either fold the idea into the main clause, make it a standalone sentence, or delete if it's redundant.

"Has generated debate about" / "prompted broader reflection on" / "shaped emerging policy discussions" — these are almost always unsupported claims. Delete or substantiate with a specific source.

4. Promotional / Ad-like Language

Words to kill or replace:

  • "boasts a" → "has"
  • "vibrant" → specific color/activity or delete
  • "rich" → specific
  • "profound" → specific or delete
  • "enhancing" → "improving" or delete
  • "showcasing" → "showing" or delete
  • "exemplifies" → "is an example of" or just say the thing
  • "commitment to" → "tries to" or "works on"
  • "natural beauty" → describe it
  • "nestled" → "in"
  • "in the heart of" → "in"
  • "groundbreaking" → "new" or describe what changed
  • "renowned" → "known for [X]"
  • "featuring" → "has"
  • "diverse array" → "several" or list them
  • "unlock" / "harness" / "empower" → describe the actual mechanism or benefit
  • "elevate your experience" → say what actually changes for the user

5. Vague Attributions / Weasel Words

  • "Industry reports suggest" → name the report or delete
  • "Observers have cited" → who?
  • "Experts argue" → which experts? Cite or delete.
  • "Some critics argue" → same.
  • "It is important to note..." → just say the thing
  • "It is widely understood that..." → prove it or delete
  • "several sources/publications" (when only a few cited) → be honest about quantity
  • "such as" before exhaustive lists → if the list is complete, drop "such as"

6. False Balance / Hedging

AI is RLHF-conditioned to sound "balanced" even when a topic isn't genuinely ambiguous:

  • "On the one hand... on the other hand..." structure when the writer actually has a view → pick a side
  • "While X has benefits, Y also presents valid points" → say what you actually think
  • Avoid forced neutrality. Real humans have biases and aren't afraid to show them.

7. Outline-like Conclusions

Formulas to break:

  • "Despite its [positive words], [subject] faces several challenges..."
  • "Despite these challenges, [subject] continues to thrive..."
  • "Challenges and Future Directions" / "Future Outlook" / "Challenges and Legacy" — section titles
  • "The future of X lies in its ability to adapt..."
  • "In conclusion" / "In summary" / "Overall" — delete

8. Knowledge-Gap Filler

When AI doesn't know something, it wraps the gap in confident-sounding filler:

  • "While specific details are limited/scarce..."
  • "not widely available/documented/disclosed"
  • "based on available information"
  • "in the provided/available sources"
  • "as of my last knowledge update"

Humans say "I don't know," "records are missing," or just skip the point entirely. Don't paper over gaps.

9. High-density AI Vocabulary

Era-agnostic overused words:

  • additionally (especially at sentence start)
  • align with
  • boasts (meaning "has")
  • bolstered
  • crucial
  • delve
  • demystify
  • emphasizing
  • enduring
  • enhance
  • fostering
  • garner
  • highlight (as verb)
  • interplay
  • intricate/intricacies
  • key (as adjective)
  • landscape (abstract)
  • mastery / masterclass
  • meticulous/meticulously
  • navigate (metaphorical)
  • pivotal
  • realms (of possibility etc.)
  • resonate
  • robust
  • showcase
  • tailored (a tailored approach)
  • tapestry (abstract)
  • testament
  • underscore (as verb)
  • valuable
  • vibrant
  • concrete (when defending against AI accusations)

These words aren't banned — humans use them. The tell is density. If a paragraph contains four or more, strip at least half.

10. Copula Avoidance

AI avoids "is"/"are" in favor of:

  • "serves as"
  • "stands as"
  • "marks"
  • "represents"
  • "refers to" (especially in leads)
  • "features"
  • "offers"
  • "maintains"

Replace with simple "is"/"has" unless there's a real reason not to.

11. Negative Parallelisms

Patterns to break:

  • "Not only X, but (also) Y" → "X and Y"
  • "It is not just X, it's Y" → direct statement
  • "no X, no Y, just Z" → direct statement
  • "not A but B" across multiple sentences → pick one
  • "More than just [category]" → describe directly

12. Lexical Diversity / Elegant Variation

AI avoids repeating words, so it cycles synonyms unnaturally:

  • "constraints" → "obstacles" → "limitations" → "restrictions" in successive sentences when the same thing is meant
  • "community" → "collective" → "group" → "circle"

Humans repeat words. It's fine. Pick one term and stick with it.

13. False Ranges

AI uses "from X to Y" when listing unrelated items, as if they form a spectrum:

  • "topics ranging from biology to medieval history" → "topics including biology and medieval history"
  • "from startups to Fortune 500s" → only valid if it's literally every size in between

14. Arbitrary Quantification

AI loves to number things that don't need numbers:

  • "3 Key Pillars," "5 Core Strategies," "The 4 Pillars of X"
  • If there aren't exactly three distinct points, don't force them into a numbered list.

15. Formatting Tells (for humanization)

These matter when the output format is visible to the reader (not plain text passed through):

  • Boldface overuse → remove most of it; humans bold very sparingly
  • Em dash overuse → replace with commas or -- ; use maybe once per 200 words
  • Curly quotes → replace with straight quotes " ' unless the context demands typographic quotes
  • Vertical lists with bold inline headers → convert to plain paragraphs or simple bulleted lists
  • Thematic breaks (----) before every heading → remove
  • Markdown (# headings, **bold**) → convert to the target format's native syntax
  • Bullet characters (•) → actual list syntax or flowing prose