Key takeaways
AI text tends to share patterns: stock openers, familiar vocabulary, tidy rhythm, vague claims, and few specifics. Knowing them helps you edit your own drafts and read other people's more carefully.
None of these signs prove that a person or a model wrote something. Human writers do all of them, and detection tools make mistakes.
Wikipedia's editors maintain one of the most detailed public field guides to these patterns, and this article draws on it.
The best fix is rarely a swap of words. It is adding something only a person with knowledge would add: a specific detail, an example, or an opinion.
Readers have become good at sensing it. A paragraph is smooth, correct, and somehow empty. Something about it feels assembled rather than written.
That feeling comes from patterns. Large language models tend to produce similar openings, similar words, similar rhythms, and similar structure. Once you know the patterns, you can spot them in a draft, whether you wrote it yourself or got help from a tool.
This guide lists 22 signs of AI writing, grouped by type, with a before and a fix for each. It also explains plainly what these signs do and do not prove.
A Caution Before You Start
None of these signs prove that AI wrote something. Human writers use stock openers, repeat favorite words, and write in tidy threes, and a model can be prompted or edited to avoid all of them. Some published research on detection tools shows how often they misjudge, particularly with non-native English writers. We explain that in our guide to how AI detectors work.
Wikipedia's volunteer editors at WikiProject AI Cleanup keep a public page of these patterns, Signs of AI writing, and it is a good place to read more. It makes the same point: the signs are descriptions of tendencies, not rules, and many of them appear in human editorials and blogs. It also warns not to rely only on detection tools, and not to treat the signs as problems to patch one by one, since that can simply make them harder to see without improving the writing. The page notes that vocabulary fashions change from one model version to the next, so any list of "AI words" ages quickly.
Use this list as an editing checklist, not as a lie detector.
Openings and Closings
1. The stock opener
Looks like: "In today's fast-paced digital world, content is more important than ever."
Why it reads as AI: it says nothing specific and could open any article.
Fix: start with the first useful fact. "Most blog posts get read for under a minute. Here is how to use that minute."
2. The announcing opener
Looks like: "In this article, we will explore the many facets of remote work."
Fix: delete it and start with the point.
3. The tidy summary closer
Looks like: "In conclusion, remote work offers many benefits and challenges, and it will continue to shape the future of work."
Fix: end on a decision, a next step, or a specific takeaway. Not a recap of what the reader just read.
4. The "challenges and future prospects" ending
Looks like: "Despite these challenges, the future looks bright."
Why it reads as AI: Wikipedia's guide lists this template ending as a common pattern.
Fix: name the actual challenge and what is being done about it, or leave it out.
Word Choice
5. Overused vocabulary
Looks like: "delve," "underscore," "intricate," "tapestry," "landscape," "realm," "crucial."
Note: Wikipedia's guide names words such as "delve," "underscore," and "intricate," and also says that word fashions shift between models, with some words fading over time. Do not trust a fixed word list.
Fix: use the plain word. "Look at" instead of "delve into." "Show" instead of "underscore."
6. Inflated significance
Looks like: "The launch marked a pivotal moment in the company's evolution." "It stands as a testament to..."
Fix: say what happened. "Sales doubled in the first year."
7. Promotional adjectives
Looks like: "a vibrant community nestled in the heart of the valley."
Why it reads as AI: it sounds like a brochure, and Wikipedia flags promotional language for this reason.
Fix: give a fact. "About 3,000 people live there."
8. Empty adjectives and buzzwords
Looks like: "robust," "seamless," "cutting-edge," "game-changing," "innovative solutions."
Fix: show the property. "It handled 10,000 requests a minute in our test." (Use a real number.)
9. Avoiding the simple verb "is"
Looks like: "The park serves as a hub for the community." "The report stands as a guide."
Why it reads as AI: Wikipedia's guide notes a tendency to avoid plain "is" and "are."
Fix: "The park is where the community meets."
Sentence Rhythm and Structure
10. "Not just X, but also Y"
Looks like: "It is not just a tool, but also a partner."
Why it reads as AI: negative parallelism is on Wikipedia's list of common patterns.
Fix: state the claim directly. "It also helps with planning."
11. The rule of three
Looks like: "fast, simple, and effective." "Plan, write, and publish."
Note: people use lists of three too, and it is a classic device. The sign is when almost every list has exactly three items.
Fix: use the number of items that is true. Two, four, or one.
12. Uniform sentence length
Looks like: five sentences in a row, each 18 to 22 words.
Fix: mix a short sentence with a long one. Read it aloud and listen for the beat.
13. Stacked hedges
Looks like: "It is important to note that, in some cases, results may vary."
Fix: say what varies and why, or cut it.
14. Over-balanced both-sides writing
Looks like: "On the one hand... on the other hand..." on every topic, with no verdict.
Fix: take a position when you have one, and say why.
15. Repetition of the same point
Looks like: a paragraph that restates the previous one in new words.
Fix: delete one of them.
Content and Substance
16. Missing specifics
Looks like: "Many companies have seen significant improvements."
Why it reads as AI: no names, no numbers, no places.
Fix: add one verifiable detail. "A mid-sized retailer cut returns by a fifth after changing its size guide." (Only if it is true and you can source it.)
17. Vague attributions
Looks like: "Experts say," "studies show," "many believe."
Why it reads as AI: Wikipedia's guide lists vague attributions among the signs of superficial analysis.
Fix: name the source and link it, or remove the claim.
18. Overgeneralizations
Looks like: "Everyone knows that quality content wins."
Fix: narrow the claim to what you can support.
19. No opinion, no experience
Looks like: a piece that explains everything and has no "I tried this, and here is what happened."
Fix: add a real experience, a mistake you made, or a view you hold.
Formatting and Citations
20. Stray formatting and heavy structure
Looks like: stray asterisks or pound signs from Markdown, headings for every two sentences, and bullets of sentence fragments. Wikipedia's guide also notes excessive boldface and em dashes.
Fix: use formatting where it helps a reader scan. Write real paragraphs elsewhere.
21. Chatbot leftovers
Looks like: "Certainly! Here is a..." or "I hope this helps!" left in the text, or a line about being a language model.
Fix: delete it, and read the whole piece for other leftovers.
22. Invented or broken citations
Looks like: references that do not exist, links that go nowhere, or quotes with no source. Wikipedia's guide lists fabricated references among the signs.
Fix: check every source yourself. If you cannot find it, remove it. This one matters most, because a made-up citation is a factual error whoever wrote it.
A Quick Before and After
Before (generic):
In today's fast-paced world, effective communication is more crucial than ever. By leveraging innovative strategies, businesses can unlock a tapestry of opportunities and foster seamless collaboration.
After (specific, with invented details):
Our team used to lose about an hour a day to unclear emails. We now start every message with the decision we need. It is a small habit, and it cut the back and forth.
The "after" is shorter, uses plain words, and has a detail that a person with experience would add. The details here are invented to show the pattern. In your own writing, use true ones.
How to Edit AI-Assisted Writing
- Read it aloud. Mark every sentence you would never say.
- Hunt for specifics. Wherever you see "many," "various," or "significant," add a name, number, or example.
- Cut the first and last sentences of each section and see whether it still works.
- Replace buzzwords with plain verbs.
- Vary the rhythm. Combine short sentences. Break up long ones.
- Add what only you know. A story, a result, a mistake, a view.
- Check every fact and source.
Our guide on how to humanize AI content walks through this process step by step. If you want a faster first pass, the free text humanizer can smooth stiff phrasing, and the free AI content detector gives a rough read on how machine-like a draft sounds. Both are aids to your editing, not proof of anything. For the search angle, see what Google says about AI content. For the email-specific version of this problem, see alternatives to "I hope this email finds you well".
What These Signs Cannot Tell You
They cannot tell you who wrote a text. Careful humans write blandly and models can be directed to write sharply. Do not accuse anyone of using AI on the strength of a few phrases, because the cost of being wrong falls on a real person. If you are a teacher, editor, or manager, treat these signs as a reason to ask a question, such as "can you walk me through how you wrote this?", not as evidence.
