Fixing AI-speak

A style-guide to detect and remove obvious AI-isms in text. To be used as a “edit out all of this” style guide for my donor profiles.

ANTI-STYLE GUIDE

How to Edit Out the Tells of AI-Generated Writing

Negative Reference for Final Editorial Passes

INTRODUCTION

This is not a style guide for how to write well.
It’s a reference for what to delete when a piece of writing feels synthetic.
It is meant for use near the end of a writing process—once your structure, content, and argument are solid—and you’re ready to ensure the work feels like a human wrote it. This is your “AI washout pass.” You’re not polishing; you’re subtracting.

Use this document as a checklist for:

  • Donor profiles
  • Proposals
  • Blog posts
  • Essays and open letters
  • Email campaigns
  • Public-facing writing that must be legible, persuasive, and human

1. STRUCTURE TELLS

1.1 Paragraph Uniformity

AI sign: Every paragraph is the same length (3–5 lines), structurally identical, with neat intros and conclusions.
Fix: Vary paragraph shape and pacing. Allow for fragments, one-liners, or intentionally short emphasis blocks. Let longer paragraphs run when the thought warrants it.

If every paragraph can be swapped for another without changing the feel, something’s wrong.

1.2 Section Symmetry

AI sign: Sections are evenly sized and similarly built: three bullets here, three bullets there, mirror headers throughout.
Fix: Break symmetry when needed. Let some sections be tighter or messier. Natural writing breathes unevenly.

1.3 Repetitive Substructures

AI sign: Within sections, every point starts with the same phrase structure (“One reason is… Another is… Finally…”).
Fix: Vary entry points. Lead with consequence, not enumeration. Not everything needs scaffolding.

1.4 Template Openings and Conclusions

AI sign:

  • Intro starts with “In today’s world…” or “As we navigate…”
  • Conclusion uses “In summary…” or “Looking ahead…”
    Fix: Cut generic intros and mechanical wrap-ups. Start in the middle of the action. End when the thought resolves.

Human writers often begin abruptly or finish naturally. AI prefers a padded landing.

2. RHYTHM AND PACING TELLS

2.1 Sentence Rhythm Consistency

AI sign: Every sentence follows a tight subject-verb-object form or ends with a phrase that mirrors the one before it.
Fix: Mix sentence length and shape. Let some trail off. Let some punch early. Read it aloud: if it sounds like a speech generator, break the pattern.

2.2 Em Dash Overload

AI sign: Sentences peppered with em dashes—constantly—used to extend thoughts that should’ve stopped.
Fix: Use dashes sparingly. Favor periods. Don’t emulate “smart” style with punctuation alone.

2.3 Excessive Threes (Rule of Three Fatigue)

AI sign: Lists of three appear constantly: “X, Y, and Z.” Also, triple adjectives: “strategic, ethical, and effective.”
Fix: Break the spell. Use two when that’s all there is. Use one strong example. Use four when it’s honest.

AI overuses three because it feels safe, complete, and rhythmic. Humans vary out of instinct.

3. LANGUAGE & VOCABULARY TELLS

3.1 Buzzword Clumping

AI sign: Clustered adjectives like pivotal, vital, intricate, innovative, dynamic, ever-evolving. Especially when paired.
Fix: Cut inflation. Use specific, neutral language. Describe what’s happening, not how important it is.

Instead of “a pivotal and vibrant community initiative”, try “a new tenant union formed after a rent hike.”

3.2 Overuse of Hedging Phrases

AI sign: Constant use of “It’s important to note…”, “One could argue…”, “may prove useful…”, “seeks to…”, or “aims to…”
Fix: Take a position. Say what something is. Don’t overqualify unless the nuance matters. Remove filler.

3.3 Inflated Verbs

AI sign: Simple actions are dressed up: “serves as a platform for…” instead of “is a place for…”, or “represents a shift toward…” instead of “marks a change.”
Fix: Favor short, declarative verbs. Use “is,” “has,” and “does” when they’re what you mean.

3.4 Meaningless Adjectives

AI sign: Descriptors that sound fine but say nothing: robust, impactful, innovative, vibrant, dynamic.
Fix: Replace with evidence or remove entirely. If you can’t prove it or show it, don’t describe it.

3.5 Elegant Variation for No Reason

AI sign: The subject is renamed repeatedly: “the company,” “the firm,” “the tech giant,” “the innovator” in one paragraph.
Fix: Stick to consistent terms unless variety adds clarity. Repetition is better than confusion.

3.6 Hallmark Phrases

High-risk phrases that almost always signal AI unless carefully placed:

  • “In today’s fast-paced world…”
  • “Stands as a testament to…”
  • “Unlocking the potential of…”
  • “An ever-evolving landscape…”
  • “A rich tapestry of…”
  • “Driving innovation through…”

Fix: Delete on sight. Replace with what’s actually happening.

4. CONTENT AND INSIGHT TELLS

4.1 Generality Over Specificity

AI sign: The text gestures toward ideas without examples: “This trend is changing how we live and work…”
Fix: Add names, places, numbers, causes, effects. Tell what happened. Don’t just hint.

4.2 Repetition Without Development

AI sign: Same idea restated across multiple paragraphs with slight variations.
Fix: Advance the argument. Each paragraph must add or pivot.

4.3 No Personal Texture

AI sign: Complete absence of voice, position, narrative, or grounded point of view.
Fix: Ask: could this have been written by anyone? If yes, you need to inject perspective—direct, implied, or structural.

4.4 Forced Balance

AI sign: Every sentence hedges: “It may not be perfect, but…”, “While some argue X, others argue Y.”
Fix: You don’t need to show both sides unless that’s the point. Pick a lane. Show your judgment.

4.5 Hollow Significance

AI sign: Every fact is inflated: “This marked a major turning point…”, “a critical shift in the landscape…”
Fix: Let big things feel big through detail. If it matters, prove it. If it doesn’t, shrink it.

5. FORMATTING & TYPOGRAPHY TELLS

5.1 Uniform Sentence Length

AI sign: Every sentence spans roughly the same length—12 to 16 words.
Fix: Mix in short bursts. Let long ideas run. Break the rhythm.

5.2 Obvious Template Formatting

AI sign: Repetitive subheadings (“The Importance of X”, “The Role of Y”, “Looking Ahead”).
Fix: Make headings meaningful. Use phrasing that belongs only to this piece.

5.3 Punctuation Tells

AI sign: Em dashes in every paragraph. Curly quotes (“ ”) where the system usually shows straight quotes.
Fix: Clean dashes. Normalize quotes. Use em dashes sparingly and only for genuine interruption.

6. TONE & VOICE TELLS

6.1 Over-Politeness and Deference

AI sign: Sentences feel like a customer support rep wrote them: “It’s worth noting that…”, “Hopefully this helps…”, “Thank you for reading!”
Fix: Remove anything that sounds like it was written in a chat window.

6.2 No Real Author

AI sign: The piece avoids commitment, avoids “I,” avoids specificity, avoids style.
Fix: Ask yourself: Could this have been written by someone I know? If not, it probably wasn’t.

6.3 Too Smooth, Too Safe

AI sign: Every sentence flows easily. Nothing interrupts. There’s no jagged edge, no roughness.
Fix: Add friction. Include something strange, personal, or unresolved. Let it breathe.

If the piece is technically perfect but doesn’t feel alive—it isn’t.

7. DECLARED COMPRESSION (THE “SHORT ANSWER” PROBLEM)

AI sign:
Explicit framing like:

  • “Short answer:”
  • “In brief:”
  • “Put simply:”
  • “The core idea is this:”
  • “The key takeaway:”

Followed by 2–4 tight, aphoristic sentences.

Why this is a tell:
Humans don’t announce abstraction levels. They move into them.
This move exists to reassure the reader that the argument is under control — a classic AI impulse.

It also collapses tension too early.

Fix:
Delete the label and the block.
If the idea matters, let it emerge through explanation, example, or consequence — not declaration.

If it only works when labeled, it doesn’t belong.

8. APHORISM STACKING (PRE-PACKAGED WISDOM)

AI sign:
A sequence like:

  • declarative sentence
  • contrast sentence (“X isn’t Y — it’s Z”)
  • closing punchline (“That’s the gap.” / “That’s the problem.”)

Why this is a tell:
This is a rhetorical unit optimized for memorability, not thinking.
It reads like something meant to be quoted, not something discovered.

Humans rarely land insights this cleanly unless they’re writing slogans.

Fix:
Break the unit.

  • Merge sentences so one does too much work
  • Add friction, specificity, or consequence
  • Or cut the punchline entirely and move on

If it feels like a slide headline, it’s wrong.

9. OVER-RESOLUTION (ENDING TOO NEATLY)

AI sign:
Sentences that explicitly name closure:

  • “That’s the gap.”
  • “That’s the tension.”
  • “That’s the challenge.”
  • “That’s why this matters.”

Why this is a tell:
AI prefers to name significance.
Human writers often trust the reader to feel it — or leave it unresolved.

Explicit resolution reads as instructional, not authored.

Fix:
Delete the sentence.

If the point holds, it will hold without being labeled.

10. PERFORMATIVE CLARITY

AI sign:
Writing that is:

  • perfectly balanced
  • rhetorically elegant
  • immediately graspable
  • hard to argue with
  • and emotionally inert

Why this is a tell:
This is “good AI writing.”
It explains without risk. It clarifies without cost.

Real thinking usually leaves residue:

  • a slightly awkward sentence
  • a turn that overreaches
  • a claim that invites pushback

Fix:
Introduce texture.

  • Add an example that narrows the claim
  • Add a consequence that sharpens it
  • Or let the sentence run longer than is polite

If it feels done, it’s probably dead.

11. FINAL FAIL-SAFE TEST (NON-NEGOTIABLE)

Before publishing, ask:

  • Would I say this out loud to a specific person?
  • Would this sound strange if read in a room?
  • Does this feel like it’s explaining to the reader instead of thinking with them?
  • Does this section feel like it could be lifted intact into a deck or carousel?

If yes: cut it.

Even if it’s right.
Especially if it’s right.

LAST RULE (PRINT THIS)

If a passage feels quotable on first read, delete it and rebuild the idea the long way.

This document exists to strip the synthetic finish off good ideas.
That includes ideas that look finished before they’ve earned it.

FINAL TEST: SPOT-CHECK PROMPTS

Use these as last-stage sniff tests:

  • Could this paragraph appear in a hundred other articles with no changes?
  • Do any sections feel “pre-written,” like they were waiting for a topic to be slotted in?
  • Does every paragraph build, or are some just summarizing?
  • Does this read like something you’d forward to someone and say “this nails it”?
  • Is there anywhere the reader might stop and say, “Hmm” or “Wait, that’s interesting”?
  • If you removed all subheadings, would the argument still move?

If the answer to too many of these is no, there’s probably still residue.

TAGLINE VERSION (for proposal or content team headers):

This document exists to strip the synthetic finish off good ideas.
Use it to break the spell, restore the texture, and get back to writing things that sound like you.

Leave a comment