Editorial Guidelines: Kill the AI Speak | Kale Realty

The Words That Scream "An AI Wrote This."

This is the list I run every piece of AI-generated content against before it goes out under my name. Free. Bookmark it.

Why I built this.

The voice print prompt (on the other page) makes ChatGPT and Claude write in your voice. This page is the safety net. It's the checklist you run after the AI hands you a draft. Because even with a great voice profile loaded, the AI defaults leak through.

This is the working list. The non-negotiables. The words that, if you spot one in your draft, you stop and rewrite. I pulled it from 761 blog posts and roughly 3 million words of my own writing, plus everything I've watched AI get wrong.

(Use it as-is, or use it as the template to build your own. Either works.)

The eleven rules.

These are mine. Yours will differ. Use them as a starting point.

  1. First-person always. "I" is the default. Never "one might" or "we as professionals" or passive constructions. Even industry content filters through personal experience. "I've seen agents who…" not "Many agents tend to…"
  2. Short sentences. Target an 11 to 15 word average. Max 25 words per sentence. One-word sentences welcome.
  3. Parenthetical asides. 2 to 4 per piece. Involuntary thoughts that leak onto the page. They undercut or add color to whatever sentence they're inside of.
  4. At least one digression per piece over 500 words. Go somewhere unexpected. Flag it ("Got a little off track there" / "Anyway"). Then return. The digression should be entertaining on its own.
  5. Self-deprecation that's specific. Not "I'm such a mess." Instead: "I put down an entire box of Entenmann's by noon." Self-deprecate, then pivot to a joke or insight. Never wallow.
  6. Rhetorical questions. 3 to 5 per piece. At least one absurd. At least one self-answered. "Is there such a thing as pot roast? There must be."
  7. Setup-subversion jokes. At least two per piece. Immediate correction ("I cared about three things. Okay, two things."). Parenthetical undercut. "Just kidding" pullback. Delayed reveal.
  8. Vulnerability stated plainly. "I was lonely." Not "I was so incredibly lonely that my heart ached." State it once, don't editorialize, move on.
  9. Contractions mandatory. "Don't," "can't," "I'm," "it's," "they're." Always. Never "do not," "cannot," "I am" in anything spoken or casual.
  10. Cold opens only. First sentence drops into action or makes a declaration. No "Hello everyone" or "Today I want to talk about" or "Let me start by saying."
  11. Short landings. End with a line or two. Sometimes one word ("Fin."). Never a moral. Never "And that's when I realized…"

The banned list. (Copy this. Use it.)

These are the patterns AI tools default to that break voice authenticity on contact. If your draft has any of these in it, the AI wrote it, and a reader can feel it.

dive in / dive into / let's dive
delve / delve into
unpack (as metaphor)
robust
seamlessly / seamless
transformative
unlock (as metaphor)
pivotal
empower / empowering
landscape (as metaphor)
holistic
cutting-edge
leverage (as verb)
synergy
ecosystem (outside of actual biology)
at the end of the day
here's the thing (as standalone opener)
I'm passionate about
as a [role], I believe
game-changer
let's be real / let's be honest (as opener)
without further ado
in today's world
it goes without saying
that being said (use "but" or "look")
needless to say
first and foremost
absolutely / exactly / totally (as agreement)
great question
I appreciate you sharing

Tone, in five lines.

  • Warm, not saccharine.
  • Self-aware, not preachy.
  • Funny from honesty and specificity, not from trying hard.
  • Vulnerable but never fragile.
  • Smart but never intellectual. Mock yourself if you use a big word.

How the voice shifts by surface.

Real estate and recruiting. Filter through personal experience. "I've interviewed 700 top producers and the thing that keeps coming up is…"

Podcast scripts. More digressive than written. Let yourself wander. The co-host reels you back.

Video scripts (reels and short-form). Compress hard. Cold open in three seconds. One aside, one subversion, done.

Email and outreach. Keep the warmth. Lose the digressions. It should feel like a text from a friend who runs a brokerage.

Social media. Blog-title energy. Specific, surprising, makes someone want to know more.

Run this on every draft.

The five-second checklist. If 3 or more items fail, send the draft back to the AI to fix.

  • First person throughout
  • Average sentence under 18 words
  • 2 or more parenthetical asides
  • 1 or more digression with a self-aware return (if 500+ words)
  • 1 or more specific self-deprecating moment
  • 2 or more rhetorical questions
  • 2 or more setup-subversion jokes
  • Zero banned AI-speak words
  • Zero corporate jargon (unless ironic)
  • Zero motivational-speaker tone
  • Vulnerability stated plainly
  • Contractions everywhere
  • Cold open (no throat-clearing)
  • Short landing (no moral)
  • Reads like something said out loud

This is the system we share with every Kale agent.

The voice print prompt. The editorial guidelines. The AI prompt library. None of it locked behind a paywall, because the agents already at Kale wanted it shared. If you want to see what running an AI-equipped real estate practice looks like from the inside, the door's open.

Every AI-assisted output should clear your local association rules, your brokerage's AI policy, and Fair Housing before it goes out. The chatbot doesn't know the rules. You do.

D.J. Paris · Kale Realty · Chicago · 312.238.9796