“It’s Not This. It’s This.”

Tips for Avoiding “AI Speak”

AI chatbots can dramatically reduce time from inspiration to publication for writers. They can organize your notes, create an outline, generate and edit drafts, and identify weaknesses in your reasoning.

They can also discredit you in under a minute.

In a sea of AI-generated or AI-assisted content, how do you distinguish yourself?

For one thing, make sure anything you write with AI assistance doesn’t sound like you didn’t bother to read it. AI chatbots use certain stylistic elements and phrases repeatedly. For a skeptical audience, these conventions signal “generic content, generated without any human involvement.”

Will the increasing prevalence of these conventions ultimately shift perception of “good” writing? Perhaps.

In the meantime, here are some suggestions to preserve your authentic voice and avoid undermining credibility with your intended audience.

1. Don’t Use “It’s Not X, It’s Y” Construction.

In the past month, this ChatGPT stylistic convention seems to have found its way into two-thirds of the LinkedIn posts I see. Two explanations are possible: (1) People have been posting like this for some time, and LinkedIn has only recently adjusted its algorithm to channel them into my feed. (2) The LLM has fed upon AI-generated content containing these words and now predicts them as the most likely next words, creating a self-perpetuating loop.

The next time you read a social media post, blog, or marketing email, count how many times this word pattern appears. There are a few variations. For example:

• It’s not X, it’s Y.

• This is not X—it’s Y.

• Not X. Not Y. Just good plain Z.

Once you see it, you can’t unsee it.

2. The real test is _______.

Same as #1. Avoid.

3. No em dashes.

This one’s controversial. I am a perennial em dash user — and was using them long before AI decided to adopt the mark as its favorite grammatical connector. (See, e.g., the foregoing em dash.) But current conventional wisdom regards the em dash as a tell that the writing is not your own. For now, ease up on them.

4. Beware confident pronouncements that mean nothing.

AI is confident, but does what it churn out actually have meaning? Is it even true? When something sounds vague but profound or reads like a universal truth, consider it again. Most likely, it falls apart under scrutiny. Identify and interrogate every bold truism in your AI-enhanced writing. Ask yourself: is this actually true? Can I explain what it means? If the answer to either question is no, rewrite.

5. Simplify, simplify, simplify.

LLMs tend to generate certain phrases meant to convey importance. Not only are they often unnecessary, but they can be edited for simplicity (or variety). See if you can express the same thought in fewer or different words. For example:

• “What This Means” = meaning, definition, indicates, signifies, in other words

• “Why This Matters” = significance, impact, implications, relevance

• Operating Rules = rules

Try your hand at “What This Does”, “What This Doesn’t Do”, “What This Solves For”

Strunk, White, and your reader will thank you.

6. No emojis.

Or at least avoid them in professional contexts. AI seems to think that prefacing each paragraph with a tiny graphic that is vaguely thematically connected to the text will help the reader better understand such text. I am not persuaded. The proliferation of emojis is not aesthetically appealing, nor does it come across as a professional design choice. This may work in limited contexts, but for now my instinct tells me to avoid.

7. Write first, use AI second.

Think about what you want to say, and draft it yourself first before feeding it into AI for revision. It doesn’t need to be perfect or polished. Even rough bullets or dictated stream-of-consciousness notes will help introduce a human voice into the output. You will get better and more authentic results if you begin with some thoughts of your own.

**

Agree? Disagree? Have other “AI-isms” worth adding to the list? I’d love to hear from you.

***

Disclosure: I drafted this post myself. I then ran it through both Claude and ChatGPT for suggested cleanup. If you’d like to see what each did, you can find them here and here.

© 2026 Tara Herman

---