Full Disclosure: A.I. Helped Me With This
When I'm reading a client's manuscript, I sometimes notice passages that sound--to my ear--like A.I.
To be clear, I'm not thinking "AI wrote this book." More often I'm thinking, "This is someone who has something important to say and he or she may have dropped their own original writing into AI to polish it.
The ideas are those of my clients; the language sounds like A.I.
And because the writing does sound good, it may feel--to the writer--like an improvement. (Honestly, it's free editing, right?)
My concern is that readers and publishers can now notice this. Readers are becoming more and more sensitive to "writing that has that smooth, slightly generic AI cadence." (Thank you, AI for that phrasing.)
To be clear, I use A.I. daily as a conversation partner, the way I'd speak to a consultant about my business or how to do some home repair. It can even help to discern the best way to structure an argument or arc in a book. But I really don't want it anywhere near my actual words.
For this reason, I encourage my clients to steer clear of A.I. in the process of writing or editing their books.
Curious, I asked A.I. for the ways writing can sound AI-Polished. Here's what ChatGPT said (in BLUE!)...
Signs Your Writing May Sound AI-Polished
AI writing isn't always easy to identify because it's bad. Often, it's easy to identify because it's too consistently polished. Here are some patterns that can make readers suspect AI has had a heavy hand in the prose:
-
Too many em dashes. AI loves the dramatic interruption—and uses it far more often than most humans naturally do.
-
“Not this, but that” constructions. This isn't about perfection. It's about progress. Used occasionally, they're effective. Used repeatedly, they're an AI tell.
-
Too many groups of three. Clear, compelling, and unforgettable. AI has a strong affection for tidy trios.
-
Perfectly balanced sentences. We don't need more information; we need more transformation. When every thought lands this neatly, the prose begins to feel manufactured.
-
Repeated mic-drop endings. Paragraph after paragraph ends with a short, punchy declaration meant to sound profound. And that's the point.
-
Generic emotional language. Words like powerful, meaningful, transformative, deeply, profoundly, and authentic can create the appearance of emotional depth without giving us anything concrete to feel.
-
Constant rhetorical questions. But what if there's another way? What if the thing holding you back isn't what you think? A few engage the reader. A steady stream starts sounding generated.
-
Overly smooth transitions. But here's where things get interesting. And that's where the real work begins. The good news? There's another way. AI is remarkably fond of ushering readers from paragraph to paragraph.
-
Restating the same idea several ways. AI often expands prose without actually adding information: it makes the point, rephrases the point, then delivers a polished summary of the point.
-
Predictable paragraph rhythm. Similar paragraph lengths, similar sentence patterns, similar openings, similar punchy endings. Human writing tends to be messier and more varied.
-
Vague examples instead of specific ones. AI gravitates toward a difficult season, an unexpected challenge, or a moment that changed everything. Humans remember the Tuesday afternoon, the broken dishwasher, and exactly what somebody said.
-
Abstract nouns doing too much work. Growth, transformation, connection, authenticity, resilience, intentionality. The prose sounds substantial while remaining oddly difficult to picture.
-
Lots of “permission” language. You don't have to have it all figured out. You're allowed to change your mind. Give yourself permission to... This has become especially common in AI-generated self-help and inspirational prose.
-
Unnaturally comprehensive explanations. Every point comes with a definition, qualification, example, counterpoint, takeaway, and reassuring conclusion—even when the reader understood it three sentences ago.
-
Excessive signposting. Here's the truth. Here's why this matters. Let's break it down. The key takeaway is... The writing continually announces what it's about to do.
-
Suspiciously universal wisdom. The prose produces statements that sound quotable but could belong in almost any book: Real change begins when we stop running from who we are and start becoming who we were meant to be.
-
Few genuine quirks. Not much slang, eccentric phrasing, surprising word choice, unfinished thought, weird metaphor, regionalism, humor, or anything else that makes you think, Only this person would say it that way.
-
Everything sounds equally good. This may be the biggest clue. Human writers have texture. Some sentences soar. Some simply get us where we're going. When every sentence seems optimized to be polished, poignant, and quotable, readers can begin to feel the machinery underneath it.
None of these proves that AI wrote something. Humans use em dashes, rhetorical questions, parallel constructions, and groups of three too.
The warning sign is accumulation.
If a manuscript repeatedly combines these patterns, readers may experience the prose as AI-generated or AI-polished—even if every idea originated with the writer.
And that's the real danger of using AI to “improve” your writing: It can make your sentences better while making your voice worse.
Okay, this is Margot again. Because I've had a hard time removing AI-impact from others' manuscripts, I asked AI what to do if it's already in there...
What If You've Already Used AI
to Polish Your Book?
Don't panic. If the ideas, expertise, stories, and substance of the book are yours, you probably don't need to start over. But you may need to de-polish it.
If AI has extensively rewritten or polished your prose, your goal now is to restore the irregularities, specificity, and personality that make the writing sound unmistakably like you.
-
Stop polishing with AI. Don't ask AI to make the manuscript “warmer,” “more engaging,” “more conversational,” or “more compelling.” That's often how the generic AI voice gets amplified.
-
Work from your own voice. Read a paragraph and then look away from it. How would you explain that idea to a friend over coffee? Say it out loud. Then write that.
-
Replace generic language with specifics. Trade a challenging season for what actually happened. Give us the room, the person, the conversation, the mistake, the smell, the ridiculous detail you still remember.
-
Break the patterns. Search for the constructions AI tends to overproduce: em dashes, tidy groups of three, rhetorical questions, “not X but Y,” canned transitions, and punchy mic-drop endings. Don't automatically delete them. Just make sure you're choosing them rather than repeating them by default.
-
Cut the echoes. If three sentences make essentially the same point, keep the one that sounds most like you.
-
Put your vocabulary back in. Restore the words, expressions, humor, metaphors, oddities, and rhythms you naturally use—even when they're less elegant.
-
Vary the texture. Not every sentence needs to sparkle. Let some sentences simply carry information. Let others be short. Let some wander a little. Human prose breathes.
-
Read it aloud. Your ear will often catch AI-ish language before your eye does. When you hear yourself thinking, I would never actually say that, change it.
-
Get a human reader involved. Ideally, choose someone who knows your natural voice. Ask them to mark anywhere they stop hearing you.
-
Revise for voice—not polish. On this pass, don't ask, Can I make this sentence better? Ask, Would I say it this way?
-
You don't need to make good writing worse to make it human. You need to make it specific, surprising, uneven, personal, and yours again.
My Work as an Editor
When I'm reading a client's work, here's the question I'm asking:
Does this prose sound AI-polished to a human reader?
I'm not worried about whether AI did or didn't write or polish the words. My only concern is how it's reading to the reader. So if I'm noticing a cluster of patterns that sound like AI-polished prose, I'm going to flag it.