This is the internal guide I use before publishing anything. Every draft gets a pass against these rules. If a word is on the banned list, it comes out. If the voice sounds confident and polished in a way no real person ever sounds, it gets rewritten. Sharing it here because a few of you asked after the AI Voice Audit post.

Part 1: Voice Style

The goal is to sound observational and exploratory, not confident and salesy. Real practitioners share what they're noticing. They don't declare what everyone should do.

The core swap

Sounds like AI

Sounds like a person

"I know this works."

"I've noticed this tends to work."

"I do this every time."

"I've been testing this for a few months."

"The only framework you need."

"One framework I keep coming back to."

"Never do X."

"X often backfires, in my experience."

"Every marketer should..."

"Many marketers I've talked to..."

Absolutes are a red flag

Words like the only, never, every, always, everyone are AI tells. Real people hedge. They've seen enough edge cases to know the world is messier than that.

Swap: absolutes to soft quantifiers.

  • "The only tool you need" becomes "One of the tools I keep using"
  • "Every founder makes this mistake" becomes "A lot of founders I talk to run into this"
  • "Never use X" becomes "X hasn't worked well for me, but your mileage may vary"

CTAs: invitations, not instructions

Prescriptive (AI)

Invitation (human)

"Do this today."

"Worth trying if you're seeing the same thing."

"Stop doing X immediately."

"I've been moving away from X - curious if others have too."

"Follow this framework."

"Here's how I've been thinking about it."

Authority: earned, not declared

Authority anchors make AI content feel hollow because they're stated, not shown.

  • "15 years taught me X" becomes "Something I keep noticing after 15 years in enterprise marketing is X"
  • "I've run this 40+ times" becomes "I've been running this for a few months"
  • "As an expert in..." becomes just showing the thing you know

How to end posts

End with uncertainty markers and a genuine question. Not a fake engagement bait question. An actual thing you're curious about.

Good endings:

  • "Curious what others are seeing with this."
  • "Still figuring out if this holds at scale - would love to hear if you've tested it."
  • "Not sure this is the right call for everyone, but it's been working for me."

Bad endings:

  • "What do you think? Drop a comment!"
  • "Share your thoughts below!"
  • "I'd love to hear your perspective on this."

Signature phrases worth keeping

These are human-sounding openers that signal observation, not declaration:

  • "Something I've been noticing..."
  • "A pattern I keep seeing..."
  • "I'm still figuring this out, but..."
  • "This might just be my experience, but..."
  • "I've been testing something and the early signal is..."

Part 2: Formatting Rules

The non-negotiable

Never use em dashes (—)

Not in LinkedIn posts. Not in newsletters. Not in tweets. Not anywhere.

Replace with:

  • A hyphen (-) for a casual aside
  • A colon (:) when introducing something
  • A period when the thought is complete enough to stand alone
  • A comma when you just need a pause

Run a final Ctrl+F for — before posting anything.

Other formatting to avoid

  • Triple line breaks between every single sentence - AI loves to make every sentence its own paragraph. It reads like someone nervously hitting enter. Group related thoughts.
  • Bolding every third word - Bold should mean something. If everything is bold, nothing is.
  • Lists of exactly 5 things, all the same length - Real insight is uneven. Some points are one sentence. Some need a paragraph. Let them be different sizes.
  • Headers on every section of a short post - If your LinkedIn post has H2 headers, it's a blog post that lost its way.

Part 3: Banned Words

These words are automatic rewrites. No exceptions.

The master list

Category

Banned

Overused AI verbs

delve, harness, unlock, unleash, foster, showcase, highlight, emphasize, underscore, empower, streamline, elevate, reimagine, redefine, accelerate, democratize

AI-flavored adjectives

cutting-edge, groundbreaking, innovative, pioneering, trailblazing, transformative, seamless, robust, scalable, intuitive, holistic, immersive, predictive, dynamic, visionary, disruptive, unprecedented, unparalleled

Marketing jargon

leverage, synergy, game-changer, paradigm, paradigm-shifting, landscape, ecosystem, realm, tapestry, testament, findings, optimize, enhance

Intensity words

crucial, pivotal, meticulously, intricate, vibrant, state-of-the-art, next-generation, hyper-personalized, AI-powered

Vague buzzwords

revolutionize, reinvent, future-proof, thought leader, best-in-class

Plain substitutions

This is the part that actually matters. The banned word is never the problem on its own - the problem is reaching for it instead of the plain word.

Instead of...

Use...

leverage

use

crucial / pivotal

important

groundbreaking / innovative

new

enhance / optimize

improve

revolutionize

change

showcase

show

foster

build

unlock

open up / give access to

delve

look at / dig into

harness

use / apply

empower

help / let

streamline

simplify

elevate

improve / lift

seamless

smooth / easy

robust

strong / solid

scalable

works at scale

intuitive

easy to use

transformative

significant / meaningful

unprecedented

new / first

accelerate

speed up

dynamic

fast-moving / active

Part 4: The Pre-Publish Checklist

Run this before every post, newsletter, or long-form piece.

Step 1: Em dash scan Ctrl+F for — and replace every instance.

Step 2: Banned word scan Ctrl+F for the top 10 most likely culprits: leverage, crucial, delve, showcase, enhance, innovative, transformative, seamless, robust, groundbreaking.

Step 3: Voice check

  • Does any sentence start with an absolute? ("The only", "Never", "Every", "Always")
  • Does the CTA give an instruction instead of an invitation?
  • Does the opening line sound like a throat-clear? ("In today's world...", "As AI continues to evolve...")
  • Does the conclusion hedge appropriately or does it declare a verdict?

Step 4: Format check

  • Is every sentence its own paragraph? Group them.
  • Are there 5 bullets all the same length? Break the symmetry.
  • Is there bolding that doesn't mean anything? Remove it.

Step 5: Read aloud If you would not say it out loud to a smart colleague over coffee, rewrite it.

Part 5: A Note on What This Is For

This guide is not about hiding that you use AI.

It's about not sounding like you outsourced your thinking.

There's a difference between using AI to write faster and using AI to write instead. The tells in this guide mostly signal the second one. When a post uses "delve" and ends with a symmetrical 5-point list and a "What do you think?" CTA, the reader doesn't think "oh they used AI." They just feel vaguely bored and unconvinced, and they scroll past.

The goal is content that sounds like a person who has something to say and used AI to say it faster.

That's it.

Guide by @rananjayraj | The AI Driven Marketer For more AI-powered marketing workflows — follow on LinkedIn

And if you want more AI + marketing workflows like this delivered to your inbox every week, join the newsletter below.

LinkedIn | The AI Driven Marketer | Website | Github | Youtube