What Are the AI Tells Nobody Mentions?
Rhythm and structure more than vocabulary: sentences of the same length, paragraphs built the same way, and characters who all explain their feelings.
You used AI to help draft a scene. Maybe brainstorming, maybe a rough pass you planned to rewrite. The words are on the page. The scene exists. But something is wrong. The prose is competent. Grammatically correct. Properly structured. And somehow completely dead. It reads like a corporate memo wearing a creative writing costume. Your beta reader can’t name the problem but says it “felt off.” A reviewer mentions it “lacks voice.” You read it again and realize you’d never write it that way. No human would.
AI prose has fingerprints. Readers are learning to recognize them. And if you don’t scrub those prints, your work will carry a smell that undermines everything you’re trying to do.
I’ve got nothing against using AI on a scene. If you don’t have the money to hire help, writing with AI is fine.
What I can’t stand is watching a writer do the hard part, the idea and the characters, and then let the machine’s habits bury it. These patterns turn up in nearly every raw AI draft, and every one of them can come out.
The hedge army. “It’s important to note.” “It’s worth mentioning.” “It should be noted that.” “One might argue.” AI hedges constantly because it’s trained to avoid absolute statements. Human writers assert. AI qualifies.
The connector parade. “However.” “Furthermore.” “Additionally.” “Moreover.” “In addition.” AI transitions like a term paper. Real prose flows without announcing every turn.
The enthusiasm problem. “Fascinating.” “Remarkable.” “Incredible.” “Absolutely.” “Tremendously.” AI oversells because it lacks actual feeling. It compensates with adjectives that try too hard.
Overselling does the most damage in fiction. A scene where a character loses someone doesn’t need “incredibly painful.” It needs whatever she does with her hands afterward. Every adjective the machine bolts on marks a place where the writer should have shown the reader something and didn’t.
The explanation compulsion. AI can’t resist explaining what it just said. It shows, then tells. Makes a point, then restates it. Reaches a conclusion, then summarizes the conclusion. Readers feel talked down to without knowing why.
The sterile rhythm. Sentence length varies in AI prose, but something about the rhythm never surprises. The beats land predictably. The patterns feel algorithmic because they are. In my experience a 3:47 AM timestamp gives it away, too. AI keeps picking that time.
The coward’s vocabulary. “Individuals” instead of “people.” “Demonstrate” instead of “show.” “Facilitate” instead of “help.” AI defaults to formal tone when casual would fit better.
Which Words Should You Delete First?
Run the five passes in order, then ask whether you would suspect AI.
Open your document. Run a search for these exact phrases: “It’s important to note,” “It’s worth mentioning,” “It should be noted,” “This demonstrates,” “Furthermore,” “Moreover,” “Additionally,” “In addition,” “However,” at sentence starts, “Particularly,” “Specifically,” “Incredibly,” “Absolutely,” “Tremendously,” “Individuals,” “Facilitate.”
Delete or rewrite every single hit. No exceptions. This pass alone will catch maybe 60% of AI smell. The remaining 40% requires reading. I’d insist on this pass even for a writer in a hurry. It’s tedious, and it’s the cheapest hour you’ll spend on the book. Skip it and every reader who spots “Furthermore” in the first paragraph learns that nobody read the page before it went out.
Every adjective the machine bolts on marks a place where the writer should have shown the reader something and didn’t.
I spent about six months building a script that strips these patterns out of my own drafts, and it keeps growing, because the tells change. Every time I read an article an AI drafted for me, I find another one, have the AI hunt down the whole family, and add it to the list. Even then the em dashes creep back in after every edit. Tell an AI to write like you and you get you, plus tells.
| Tell | What it looks like | Fix |
|---|---|---|
| The hedge army | “It’s important to note,” “It should be noted that,” “One might argue.” | Assert. Human writers state; AI qualifies. |
| The connector parade | “However,” “Furthermore,” “Additionally,” “Moreover.” | Let the prose flow without announcing every turn. |
| The enthusiasm problem | “Fascinating,” “Remarkable,” “Incredible,” “Tremendously.” | Show what the character does in place of the adjective. |
| The explanation compulsion | Shows, then tells; makes a point, then restates it. | Cut the restatement. Readers feel talked down to. |
| The sterile rhythm | Beats land predictably and never surprise. | Read aloud and vary the rhythm until something surprises. |
| The coward’s vocabulary | “Individuals,” “demonstrate,” “facilitate.” | Use people, show, help. |
Why read an AI-assisted draft out loud?
Print your pages or read from a different device than the one you wrote on. The different context helps you see fresh.
Read every sentence out loud. Listen for three things. Would you say this? Not formally, not in a speech. Would these words come out of your mouth in conversation with a friend? If no, rewrite. Does this sentence surprise you? Even a little? AI prose never surprises because it follows statistical probability. Human writing zigs when the reader expects zag. If five sentences in a row all land exactly as expected, something is wrong. Is this sentence doing work, or is it explaining work another sentence already did?
AI doubles up constantly. The second sentence after any point is usually restating that point in different words. Cut it. Watch for triples too. When a question or a statement runs subject, comma, subject, comma, and subject, I read it and know it’s AI. Someone once pushed back that triples are also eighth grade English, and he’s right. I still spot the machine by them immediately.
The Third Pass: Kill the Qualifiers
AI qualifies everything because it’s terrified of being wrong. “This can sometimes help with,” “In many cases,” “Often,” “Tends to,” “May,” “Might,” “Could potentially,” “It seems that,” “Appears to be.”
Some qualification is honest. Constant qualification is cowardice. Read through and find every qualifier. Ask: Is this uncertain, or is this the AI hedging out of habit? If habit, make the statement direct. Assert what you mean. The Revisions Handbook covers when hedging serves and when it undercuts your authority.
The Fourth Pass: Inject Imperfection
Perfect prose signals AI. Human writing has rough edges.
Fragment sentences occasionally. Not constantly. Occasionally. Let a sentence run long when the thought demands it, when the rhythm of the idea wants to unspool across more words than any grammar teacher would approve of. Sometimes that’s how thoughts work. Start a sentence with “And” or “But.” Teachers told you not to. Teachers were wrong.
Use contractions. Raw AI prose spells out words that anyone talking would run together, and dialogue written that way sounds like a deposition. I’d fix every one of them in a character’s mouth before touching anything else in the scene. Let your weird in. Your word obsessions. Your unusual metaphors. The reference only you’d make. AI writes from statistical averages. You write from a specific life. The specificity makes it human. Building a distinctive voice on the page is the same skill whether you’re cleaning up AI output or writing from scratch.
The Fifth Pass: Character Voice Check
If the AI-assisted section contains dialogue or internal monologue, run an extra pass. AI dialogue tends toward sameness. Characters all speak with similar vocabulary, similar rhythm, similar sentence structure. The only variation is content. Real characters sound different. A teenager doesn’t speak like a professor. A nervous person doesn’t speak like a confident one. An angry person doesn’t carefully structure their sentences.
Interchangeable dialogue is the fastest way I know to kill a novel. Readers stay with a story for the people in it, and once every character talks like the same polite assistant, there’s nobody left to follow.
Read each character’s dialogue separately. Block out the other characters. Does this voice sound like only this person could say it? Or could any of your characters have said this line? If dialogue is interchangeable, you’ve lost character voice. Rewrite until each person sounds like themselves and no one else. The Dialogue Handbook covers building voices distinct enough that readers don’t need dialogue tags.
My take
Read the draft out loud before you trust it. Your ear catches the sterile rhythm and the connector parades your eyes skim past. Then kill the qualifiers and give each character a voice nobody else in the book could use. Interchangeable dialogue is the fastest way I know to kill a novel. Readers stay for the people, and a cast of polite assistants gives them nobody to follow. Run all five passes, every time.
How Do You Know When a Manuscript Is Clean?
After all five passes, ask one final question: If I didn’t know this was AI-assisted, would I suspect? Not “is it okay?” Not “is it good enough?” Would you suspect?
If the answer is yes, you’re not done. Prose that merely passes inspection is a low bar, and I won’t settle for it on a book that carries somebody’s name. Keep working until the pages read like what you’d produce without any help.
Readers don’t consciously think “this was written by AI.”
They think “something feels off,” or “I’m not connecting,” or “this author doesn’t have a voice.” The damage lands whether or not they can name the cause. Your voice is the only thing you have that AI can’t replicate. Protect it. For a broader look at what AI does well and where it falls short, that article maps the whole territory.
AI is a tool, and a scene it helped draft can hold up as well as one you typed from scratch once you’ve done these passes. What makes me furious is the writer who skips them because the draft looked finished. Readers will never ask what tool you used, and they’ll quit on the first page if the story doesn’t hold them.
Prose that over-explains is telling the reader what to think. Show Don’t Tell is the book where I explain what showing means, with more than 100 before and after examples.
