How to tell if something was written by AI
Eight signs you can spot by reading – and the one habit that matters more than all of them: compare with what you know.
It answers a question nobody asked
Opens by restating the topic in general terms (“In today’s world, X is more important than ever”) instead of getting to the point.
Every paragraph is the same size
Three or four tidy paragraphs of near-identical length, each opening with a transition: Firstly, Furthermore, Additionally, In conclusion.
The vocabulary is oddly grand
Delve, tapestry, testament, pivotal, multifaceted, navigate the complexities, underscores the importance. Rare in normal writing, common in model output.
Nothing specific happens
No names, dates, numbers, places, mistakes or anecdotes. Claims are true in general and about nothing in particular.
It is balanced to a fault
“While there are benefits, there are also challenges.” No strong opinion unless the prompt asked for one.
Confident, wrong details
Invented quotes, citations that don’t exist, or page numbers that don’t match. Check two references – fabricated sources are one of the strongest signs.
It doesn’t sound like the writer
For someone whose writing you know, the biggest tell is a sudden change of voice, vocabulary or polish between pieces or within one piece.
Leftover chatbot phrasing
“As an AI language model”, “Certainly! Here’s…”, “I hope this helps”, or a closing offer to expand on anything. Copy-paste accidents happen more than you’d think.
Signs are clues, not proof
Each of these features appears in plenty of human writing. School essays are taught to use “Firstly… In conclusion”. Non-native writers often prefer safe, formal vocabulary. Look for several signs together, in a text that also lacks specifics, and then look for evidence outside the text.
Let the software do the counting
Signs 2 and 3 – even rhythm and stock vocabulary – are exactly what the VerifyWrite AI detector measures and highlights sentence by sentence. Sign 6 is a job for you: check the references. For copied rather than generated text, use the plagiarism checker.
Questions people ask
What is the most reliable way to tell if AI wrote something?
Compare it with writing you know the person produced, and check the facts and citations. Fabricated references and a sudden shift in voice are much stronger evidence than any single style feature or detector score.
Can AI text be completely undetectable?
Yes. Edited or carefully prompted AI output can read as entirely human, and some human writing reads as AI. No method – human or software – is certain.
Is using the word “delve” proof of AI?
No. Every word on AI “tell” lists is used by real people. They are signals only when many of them cluster together in a text that also lacks specifics.
Tools for checking text
Your text follows you between tools, so you can paste once and check everything.