How To Write *Without* Sounding Like AI

I enjoy writing with AI’s support. It’s allowed me to research a lot deeper than I ever had been able to, and create drafts a lot faster! Then, I go back and forth with the AI to expand or refine ideas. Finally, I edit the piece to more authentically express my voice.

This new flow has allowed me to produce higher quality ideas, and I hope it has given better value to my readers.

At least, I’ve noticed that my Ai-assisted writing has gotten at least the same amount of engagement — sometimes more — than my human-only writing ever did. And since it’s a fun process for me, I’m going to keep at it :)

Yet I also notice that the result can sometimes feel too polished or AI-ified… aka “ai slop”! I don’t always have the energy to edit as deeply as I’d like… to humanize it as much as I want to.

As a reader you can probably feel it… often within a sentence or two.

Sadly, that might make an ideal reader prejudge a piece too quickly, and scroll or tap away, without ever getting the value of all the work I put in.

So I decided to do some research into how we might use AI to more easily remove those AI-isms 🤣 …so that we can free up our writing time for more of the thinking, less of the editing!

Sometimes you might want to write simply as an exercise of reflection and thinking. Other times, you are writing for the purpose of serving your audience — to gift them the best ideas and inspiration that you can. This blog post is lovingly dedicated to those of us who write to be of service — coaches, healers, teachers, soulpreneurs.

The First Wave of AI-isms:

The first wave of AI tells were such words, which you might recognize show up often in ChatGPT, such as “delve”… “tapestry”… “navigate the landscape.” Researchers found that some of these jumped more than 50% in published articles right after ChatGPT came out! They became signatures, because models overused them at rates no human writer ever would.

Most of those have faded now. Newer models have moved on to other words… and those will, eventually, be replaced too. Which is why creating a “banned words” list isn’t the best answer.

What’s more evergreen is the structure underneath, the patterns readers feel even when they can’t name them:

• Sentences that all land in the same length range. Nothing short, nor surprising.
• Paragraphs that follow the same little template (topic sentence, explanation, example, summary, again and again…)
• The polite middle distance, where the writing never really takes a position.
• Lists of three, everywhere!
  
  

Why this matters to me:

A lot of writing about this topic frames it around fooling detection tools. How to slip past GPTZero or score well on Originality.ai…

I’m not writing essays for school. Probably neither are you. Let’s return to the purpose of why we write:

To explore our ideas. And, as importantly, to serve our ideal reader.

Also interesting to note is that independent testing which looked at 16 popular “humanizer” tools found that only 2 actually worked. The rest mangled the meaning, dropped up to 20 percent of the original content, and produced text that still got flagged! Google’s own spam policy explicitly names “automated synonymizing, paraphrasing, or obfuscation techniques” as a violation. Which is exactly what those tools do…

So the shortcut isn’t really a shortcut. The longer path (the one where we actually shape the writing ourselves) is also the better path.

A Few Things I’ve Found Helpful:

These are the practices that tend to serve human readers better:

Vary the rhythm. Mix very short sentences with longer ones. Let some be five words and other sentences be thirty, and others even a single word. Real thinking doesn’t fit into neat boxes. Real writing shouldn’t either.

Get specific. “A recent study” becomes the actual study, and “many people” becomes a real number, if there is one. “Useful tools” becomes the names of those tools. Specificity is one of the clearest signals that a real person was paying attention. Of course, AI can help you find those studies and numbers. Just be sure to double-check the web — surf to the actual sites that AI quotes. This all makes writing more grounded.

Take a position, even if small. Even a single sentence where you say what you actually think. (AI tends to hedge and not take an brave stance.) That moment of taking a stand does more to wake a reader up, than a large amount of polished paragraphs.

Cut the smoothing words. “Furthermore,” “moreover,” “it’s worth noting,” “in conclusion.” These exist mostly to make writing look organized. Often, the sentence is stronger without them. However, I still make this mistake often ;-)

Use the words you’d actually say out loud. Read a paragraph aloud. If you’d never speak it that way to a friend, client, colleague, change it and make it sound more like you.

Yes, AI can do the first pass at removing AI-isms :)

Even though I do the final editing with my own hands, I still like to use AI to do the first editing out of the ai-isms. So I’ll share with you the prompt that I’ve been testing out. You can paste it into any AI model (Claude, ChatGPT, Gemini, etc).

You are an experienced editor helping me make an AI-assisted draft sound less robotic, more human. The goal is not to fool a detector. The goal is to respect a human reader’s time and attention so that they don’t scroll away thinking that it’s ai slop.

Please work in three passes. Don’t skip any of them. Thank you!

PASS 1 — AUDIT

Read the source text below. List every place you find:
- Overused AI vocabulary (see the small list further down)
- Sequential sentences that all land in the same length range
- Sequential paragraphs following the topic-sentence / explain / example / wrap pattern
- Hedging or “both-sides” framing where a real position would instead be clearer
- Triads (groups of three adjectives or phrases) used for rhythm
- Smooth transition words that exist mostly to look organized
- Generic phrases where a concrete detail would land better

PASS 2 — CRITIQUE

For each thing you found, write one sentence about what’s wrong with it and what the writing needs instead. Don’t rewrite yet. Just notice for now.

PASS 3 — REWRITE

Now rewrite the full text with these positive principles in mind:

VOICE

- Sound like a thoughtful human sharing what they’ve learned, not a brand voice or a corporate report.
- Take real positions. If the source hedges, find the actual point underneath and say it gently but clearly.
- Use contractions naturally (it’s, don’t, you’re, we’ll).
- Active voice unless passive genuinely sounds better.

RHYTHM

- Vary sentence length a lot. Mix short ones (3 to 8 words) with longer ones (25 or more) in the same paragraph.
- Don’t let three sentences in a row have similar length or shape.
- It’s fine to start a sentence with “And” or “But.”
- Fragments are okay. Use them sparingly.
- Ellipses (…) are welcome where a gentle pause helps.

SPECIFICITY

- Replace generic claims with concrete ones from the source. “A recent study” becomes the actual study. “Many users” becomes a number if there is one.

- Cut filler that doesn’t carry information.

A SMALL LIST OF THINGS TO AVOID

Never use the following words or phrases:
- what no one else is talking about / what nobody tells you
- loop, signal, cascade, drift / drifting
- quiet / quietly
- tapestry, realm, landscape (as metaphor)
- in conclusion, ultimately, to summarize
- unlock, unleash, harness
- framing of: “not just X but Y” … or “it’s not X, it’s Y”
- Em-dashes (use commas, parentheses, or just split the sentence)

WHAT TO PROTECT

- Keep every fact, figure, name, and claim from the source. Don’t add new information or examples. Don’t drop any either.
- Match the original length within about 10%.

OUTPUT

After the rewrite, please add one short paragraph noting the biggest changes you made and why. That part is for me, not for the reader.

Thanks for your excellent work.

SOURCE TEXT:

[paste the AI-generated text here]

A few notes on how to use the prompt:

The list above is just a starting point. Modify the list to your taste! If you’ve never been bothered by something listed above, remove it. If some ai writing pattern starts to grate on you, add it to the list! Be aware that if the list gets much longer, it might make the output worse, so experiment with your prompt length.

Also, the prompt can be made much better if you give it small samples of your own preferred writing voice. Research shows that even 2 to 5 samples of an author’s voice produces dramatically better results, not just general instructions.

Run the prompt twice if needed. The first pass usually helps a lot. A second pass on the same text can help a little more.

Don’t expect a perfect result on its own. The prompt is helpful, but the best writing still needs your own human work. The small touch of a personal story, a moment of honest opinion, a sentence only you would write… those are the things no model can add, and what human readers feel most.

Your Turn:

Try generating a piece of AI writing based on a topic that interests you.

Then give the above prompt a try.

Did it make the writing better? What would you have further removed, or changed? Modify the prompt. Then run it again. Doing this iteratively will improve your prompt over time.

I’d love to hear what you learn through this process.

May this process free up our time and energy for deeper reflection and the necessary human editing. May it help us produce better writing that helps more people, more deeply!

If you’d love to learn how to use AI with a group of fellow soulpreneurs, consider joining my AI training group: https://georgekao.com/aigroup