8 signs your writing sounds like AI

I’m talking about little habits like vague claims, repeated points, confident lines with no source.

Once you know the 8 signs, you can run your own draft against them and fix what’s needed.

This is the checklist.

Sign 1: Generic content. A sentence that sounds relevant but could sit in almost any article on any topic. It doesn’t commit to anything.

“Small changes can have a significant impact on your overall wellbeing.”

Swap “wellbeing” for “productivity,” “finances,” or “relationships” and the sentence still works. That’s the tell. It’s true in the way a horoscope is true 🫢.

Fix: Replace it with the actual change. Not “small changes matter,” but “swapping your 3pm coffee for a 10-minute walk cuts the afternoon energy crash.” Name the thing.

Sign 2: Repetition. The same idea, said three times, in three outfits.

“Planning saves time.” “Preparing in advance makes your routine more efficient.” “Good organisation helps you manage your time more effectively.”

One idea. Three sentences.

Fix: Say it once, then move the word count to a new point. If you catch yourself circling back to something you already said in the intro, cut the repeat and trust that the reader remembers it.

These are harder to catch because they read fine on the first pass. The problem shows up when you ask “how do you know that?”

Sign 3: Confident but unsupported claims. A flat statement with no source, no study, no explanation of how anyone reached that conclusion.

“This method is proven to deliver excellent results.” “Experts agree that this is the best approach.”

Which experts? Proven by what? The sentence sounds certain because certainty is easy to fake.

Fix: Either name the source, or drop the certainty. “This method worked well in three tests I ran” is weaker sounding but actually true. That trade is almost always worth it.

Sign 4: Weak or missing analysis. The writing tells you what something is, never why it matters or when it doesn’t apply.

“Exercise is good for physical and mental health.”

True, and useless. A reader who already knew that learns nothing. The useful version asks the next question: what type, how often, for whom, and where does this stop being true?

Fix: After every claim, ask yourself “so what?” once. If you can’t answer it in the next sentence, the claim needs a partner sentence that does the actual work.

Sign 5: Incorrect or invented references. A study, quote, or statistic that either doesn’t exist or doesn’t say what the sentence claims it says.

“A 2024 study found that 78% of people benefit from this method.”

No study title. No author. No link that actually works. This is one of the most common tells in AI-generated text, because a model can generate a statistic that looks exactly like a real one.

Fix: Click every link before you publish. If you can’t find the original source easily, cut the claim or the citation.

Sign 6: Logical mistakes. The sentence reads smoothly, but the actual claim doesn’t hold up once you slow down.

“The plant was dormant because it was actively producing new leaves.” “The treatment was successful, although most participants saw no improvement.”

Both sentences are grammatically fine. Both contradict one another.

Fix: Read the piece backward, last section to first. It breaks the flow that lets contradictions hide, and you’ll catch the ones your eyes glide past reading forward.

These last two are the hardest to catch in your own writing, because they require comparing the draft against something outside the draft: your past work, or your actual expertise.

Sign 7: A mismatch with the writer. The vocabulary or tone suddenly jumps somewhere your usual voice doesn’t go.

If you normally write:

“I tried this recipe and found it needed more salt.”

and a paragraph suddenly reads:

“The optimisation of flavour profiles is therefore of considerable importance.” (how very word-ful!)

that’s worth a second look. A style shift doesn’t automatically mean AI wrote it. It might mean you were tired, or writing for a different audience. But check it against your last few pieces before you decide it’s fine.

Fix: Keep two or three of your older drafts open in another tab while you edit. If a paragraph doesn’t sound like something you’d say out loud, rewrite it until it does.

Sign 8: A lack of specific knowledge. The piece covers a topic broadly but skips the details someone who actually knows the subject would include without thinking.

An article about growing tomatoes might mention sunlight, water, and nutrients, and never explain how to tell if a plant is overwatered, when to stake the stem, or why a plant might grow leaves but no fruit. An article about cooking steak might describe flavor and texture without ever saying how to check doneness.

Fix: Ask yourself what a beginner would get stuck on. If your piece doesn’t answer that question, you’re summarizing the topic instead of teaching it.

Here’s where people overcorrect. Not every quirky sentence is AI. Some of these are just how real people write.

• Clean grammar isn’t a tell. Plenty of writers are careful editors of their own work.
• A short, punchy sentence isn’t a tell. One clipped line for emphasis is normal. It only becomes a pattern when you get four or five in a row.
• Formal vocabulary isn’t a tell. Someone who says “ostensibly” or “constituent” in normal conversation isn’t secretly a chatbot (or are they?).
• An unsourced claim isn’t automatically a tell either. Most casual writing don’t cite sources. It’s a problem only when the sentence sounds like it’s citing one and isn’t.

The signs matter as a group, not one at a time. One vague sentence in an otherwise sharp article means nothing. Vague sentence, plus a repeated point, plus an uncited “studies show,” in the same paragraph, that’s something worth fixing.

This checklist isn’t a scanner that spits out a pass or fail score.

It’s a lens. You still have to look through it and make a call.

Two different editors could read the same paragraph and disagree on whether it’s generic or just plainly written.

That’s fine.

The point isn’t to catch every borderline sentence. It’s to stop the obvious ones from making it into a published piece.

Once this becomes a habit, you stop rereading a finished draft five times hoping to spot what’s wrong. You know exactly what you’re looking for.

The checklist doesn’t replace your judgment. It just tells you where to look twice.

Was this article written by AI? 🤔

Listing my sources below in case you were wondering.

1. https://libguides.pace.edu/c.php?g=1292205&p=9488470
2. https://www.sagepub.com/explore-our-content/blogs/posts/sage-perspectives/2025/06/11/ai-detection-for-peer-reviewers-look-out-for-red-flags
3. https://www.teqsa.gov.au/guides-resources/protecting-academic-integrity/academic-integrity-toolkit/risks-academic-integrity-ai/detecting-plagiarism-ai-generated-text-student-assessments-and-securing-take-home-written-assessments
4. https://pmc.ncbi.nlm.nih.gov/articles/PMC11992562/
5. https://www.liverpool.ac.uk/media/livacuk/centre-for-innovation-in-education/digital-education/generative-ai-teach-learn-assess/academic-integrity-process-for-the-misuse-of-gai.pdf
6. https://www.staff.sgul.ac.uk/about/our-education-centres/centre-for-technology-in-education/our-technologies-and-services/generative-ai-support-for-staff/generative-ai-detection-guidance-for-staff
7. https://www.depts.ttu.edu/tlpdc/ai-resources/AI-Detectors.pdf
8. https://pmc.ncbi.nlm.nih.gov/articles/PMC11914838/
9. https://www.eastcentral.edu/free/ai-faculty-resources/detecting-ai-generated-text/
10. https://vu.nl/en/about-vu/more-about/alp-guide-spotting-ai-writing