Stop Your Writing from Sounding Like AI

Learn how to spot and fix AI-isms to strengthen your writing.

“Ugh, I can’t use the word ‘delve’ anymore because people will think my writing is AI?!?” I saw a version of this comment in a Facebook group for fiction authors recently.

I’m hearing a lot of writers lately bemoan the fact that certain words and phrases are so over-used in AI-generated text that people are starting to automatically assume a piece is written by an AI if they see those words in a piece of writing.

On LinkedIn, I saw someone make the following comment: “People can sense AI [writing]. It's in the structuring, the rhetorical flourishes and a certain distance of voice. Instantly raises the hackles of many, including myself.”

It’s true. There are certain telltale signs that a piece has been largely written by a generative AI model.

These signs include:

• Overly-flowery language

• Certain sentence structures

• Heading and title structures

• Over-reliance on certain terms

• Use of vague statements

• Too much filler language

When I’m editing AI-generated content, I specifically look for these AI-isms to eliminate or revise.

You don’t want your writing to automatically be dismissed as AI-generated (even if you used AI in your process).

So let’s explore some of the ways you can revise your content to eliminate these AI-isms.

NOTE: As the AI models continue to improve, fewer and fewer of the issues I’m highlighting in this piece show up. But let’s take a look, so you’re prepared if and when they do.

Overly Flowery Language

To prepare for this article, I asked three different AI models to write me a blog post about staying positive in the midst of negativity. I consulted Pi.ai (free), Microsoft’s Copilot (free), and ChatGPT 4o (paid version).

What is flowery language? A piece can be considered flowery when it relies on adjectives, adverbs, metaphors, and poetic turns of phrase.

ChatGPT: Life’s journey is full of twists and turns, with moments of challenge and shadows that can sometimes feel overwhelming. Yet within you lies a steady light—a quiet strength that can guide you through even the stormiest of days.

A flowery piece is often going to set off the Hemingway Editor and show up with an FK score of well over the 6th grade reading level most of us should be shooting for (depending on audience and context, of course).

Click here to learn more about the FK score and the Hemingway Editor.

HOW TO FIX IT:

Work on simplifying the language:

• Remove adjectives

• Remove adverbs

• Remove metaphors

• Shorten sentences

• Make sentences more direct

Here’s an example of what that paragraph could look like:

Me: Life is full of challenges and sometimes dealing with those challenges can feel overwhelming. But you can learn to draw on your inner strength to carry you through the negativity in your life.

Problematic Sentence Structures

Here’s a very common sentence structure ChatGPT seems to love (and over use):

ChatGPT: Staying positive in the face of negativity isn’t about being perfect—it’s about showing up for yourself, even when it’s hard.

Here’s another version of that structure:

ChatGPT: This proactive approach doesn’t just save lives—it saves money.

The structure is “it doesn’t just [do this], it [does this]” or “it isn’t about [one thing], it’s about [this other thing].”

I don’t see this structure showing up a lot in other models, but Chat loooooves it.

HOW TO FIX IT:

Usually, I try to either remove the negatives, or I rearrange them, like this:

Me: Staying positive is about showing up for yourself, even when it’s hard. You don’t have to be perfect. Just consistent.

Heading & Title Structures

The AIs drive me crazy with the heading or title structures they rely on. It’s almost always a two-part title, separated by a colon.

Here’s an example from Microsoft Copilot:

I may get inspired from these to come up with my own, maybe something like “A Guide to Staying Upbeat in a Negative World.” That seems to work just fine without the “Embracing Positivity” part.

Or maybe “10 Strategies for Overcoming Negativity.” I put a number at the start (very effective for open rates) and deleted the “cultivating joy” part of the title.

Sometimes, I’ll kick it back to the AI to revise. I might use a follow-up prompt like this:

Me: Please revise. These should be 5-8 words long with no colons. Focus on the benefit.

Here’s Copilot’s response:

It sort of understood the assignment, and yet the 2nd two options are still two-parters, albeit without the colon. If I had to use one of these, #1 would be the winner.

But I think I prefer, “10 Strategies for Overcoming Negativity.”

Over-Reliance on Certain Words

You’ve probably seen people complaining about the long list of words that “you shouldn’t use if you don’t want your writing to be flagged as AI-written.”

Several different versions of these lists are floating around, like this one.

Here are some common examples:

• Transform or transformative

• Delve

• Imagine… (with the ellipsis)

• Elevate

• Embark

• Beacon

To be clear, I don’t think we have to avoid these words completely in our writing moving forward.

No one can reliably spot AI-generated writing—even when some of these terms are present—because the truth is: The AI models were trained on human writing. These words have been used repeatedly in the human-written content of the world.

That said, I do think we should be mindful of HOW OFTEN these words are showing up in our content, whether we’re writing our own or editing AI output.

It’s never a bad idea to review your work for your own favorite phrases, because you may be overusing them (whether they made this list or not).

HOW TO FIX IT:

Learn to recognize some of the common terminology that AI (especially ChatGPT) likes to pepper its output with and choose alternatives.

You may need to alter your word choice. You may expand your own vocabulary in the process.

Or the sentence may flow better if you rearrange it or rewrite it from scratch.

Use of Vague Statements

I’m in the process of using ChatGPT to help me write fresh website copy for a tiny nonprofit that doesn’t have much money.

Under “Why Choose [Nonprofit Name],” it gave me the following vague statement:

ChatGPT: Instead of reacting to the consequences of overpopulation, we focus on prevention. By funding spay and neuter surgeries, we stop the cycle of stray animals before it starts.

In this case, rather than expanding on it myself, I prompted it again, asking it to elaborate on that statement.

It generated three paragraphs, but here’s the more specific opening paragraph:

ChatGPT: At [Nonprofit Name], we believe the best way to address pet overpopulation is to stop it before it starts. Every unplanned litter adds to the growing number of stray and abandoned animals, which overwhelms shelters, strains local resources, and often leads to heartbreaking outcomes like euthanasia.

Much better!

You need to review every sentence with the thought, “Is this specific enough?”

Another client really disliked platitudes. She hated when a piece of copy ended with something like, “Together, we can empower all women to live the life they deserve.”

In her words, “Blah, blah, blah.”

ChatGPT often concludes content with something similar. Watch for these kinds of meaningless, platitude-type statements and transform them (LOL, I wrote that, not the AI) into more powerful, more specific outcome statements.

Me: “With your help, we can fund educational and support programs for women survivors of domestic abuse so they can get back on their feet and enjoy their newfound self-confidence.”

Too Much Filler Language

I asked Pi.ai to revise a blog post it wrote to make it more conversational. Suddenly, my screen was full of extra, unnecessary words, because this is what it thinks makes it more conversational.

It’s not entirely wrong, but the AIs often go overboard with this instruction and I often have to tell them to tone it down.

Let’s remove “Hey there!” as well as “...but don’t worry!” and “shall we?”

I’m even going to adjust that second sentence to be less wordy by eliminating “we all know.”

To tighten it up a bit more, we could get rid of “I’m here to share” and replace it with “Here are”

Here’s the revised text:

Me: Life can be a rollercoaster ride, with plenty of ups and downs. Negativity can sometimes get the best of us. Here are some handy tips on how to stay positive even when things aren’t going your way. Let’s dive in.

Here’s a more subtle example:

In the highlighted section, this sentence can be tightened up by eliminating the unnecessary words “it’s essential.”

Now it’ll read, “Remember to appreciate small wins along the way…”

To tighten it even further, remove the words “remember to” so it reads “Appreciate small wins along the way and learn from setbacks.”

It All Depends on Your Style, Intention, and Audience

Some of us prefer a wordier, flowery, more poetic writing style. Others like a more direct, clipped, no-nonsense approach.

You also have to keep in mind what kind of writing your intended audience expects and responds to.

And you need to understand what your intention is for any piece of content you publish.

If you’re just writing for yourself and you don’t care how many people like your content, then feel free to write however you like. Who cares then if you use the same term quite a bit or you like a little fluff?

But if you’re trying to accomplish something with your words—increasing your mailing list, selling something, getting someone to take some kind of action—your content needs to be in alignment with your audience and your goal.

When you’re reviewing your AI-generated content to see if it fits with your style, audience, and goal, you’ll definitely want to pay attention to these items. Addressing them will make your writing stronger and more effective.