Nobody reads a paragraph and thinks “high perplexity, low burstiness”. They think this sounds like a chatbot, and they start skimming. That reaction comes from a small set of habits that language models fall into. Learn to spot them and you can fix them, in your own drafts or in anything AI helped you write.
One caution first: none of these proves anything on its own. People use every one of them. What gives text away is density: several tells, in every paragraph.
Vocabulary tells
Ceremonial verbs and nouns
Delve, embark, foster, leverage, underscore, navigate, showcase, elevate; tapestry, testament, realm, landscape, journey, beacon. These are the words of commencement speeches. In a work email they stand out.
Intensifiers that don’t intensify
Crucial, pivotal, vital, essential, paramount, often several per paragraph. When everything is crucial, nothing is.
Vague praise adjectives
Seamless, robust, vibrant, innovative, cutting-edge, comprehensive. They describe nothing a reader can picture.
Structure tells
The throat-clear opener
“In today’s fast-paced world”, “In the ever-evolving landscape of…”, “When it comes to…”. The paragraph starts one sentence before the point.
The recap closer
“Ultimately…”, “In conclusion…”, “By doing X, you can Y.” A final sentence that summarizes what the reader finished three seconds ago.
Reflexive triplets
Lists of exactly three adjectives or verbs, paragraph after paragraph: “clear, concise, and compelling.”
“It’s not just X — it’s Y”
The contrast frame (“This isn’t just a tool, it’s a partner”) is a favorite because it sounds insightful. Once per piece is plenty.
Headings and bullets everywhere
A three-paragraph answer broken into bolded sections with bullet points, when a few plain sentences would read better.
Rhythm tells
Read a paragraph aloud. If every sentence runs about the same length and follows the same shape (opening clause, comma, main clause), you’ll hear a drone. Human writing is lumpier: a long sentence that wanders a bit, then a short one. Then a fragment, sometimes.
Content tells
Abstraction over specifics
“Many businesses face various challenges” instead of a name, a number or an example. Models default to the general case because they don’t know your case.
Relentless balance
Every claim hedged, every side given equal time, no conclusion reached. Real people have opinions.
Nobody is there
No “I”, no “we tried”, no anecdote, no detail only the writer could know. The text could have been written by anyone, about anyone, which is exactly the feeling readers pick up on.
Punctuation and formatting
Heavy use of em dashes, colons introducing every list, Title Case In Every Heading, and emoji bullet points in professional writing. These are weak signals on their own (plenty of humans love an em dash), but they add up.
What this list is not
It isn’t a way to prove someone used AI. Human writers produce all of these, and AI detectors that look for similar patterns are notoriously unreliable, flagging real people’s work, especially non-native speakers’. Use this list to edit, not to accuse.
The goal isn’t to hide that a tool helped. It’s to make sure the final draft says something specific, in a voice that’s recognizably yours.
Ready to fix them? Follow the seven-step pass in how to make AI writing sound human.
Or let the studio do the first pass.
Humanizer Studio makes these edits for you, keeps your facts intact, and runs free on your own device. Opening soon.