You can usually feel when text came from a chatbot before you can explain why. These are the seven patterns doing the work.
1. Em dashes everywhere
The spaced em dash is the single most recognizable tell. Chatbots reach for it constantly to set off asides. Most people typing day to day never use it.
2. Suspiciously tidy lists
Three bullet points, each roughly the same length, each starting with the same kind of word. Real human lists are lumpier and less balanced.
3. Uniform sentence rhythm
AI prose tends toward even, mid-length sentences. Human writing varies wildly. A two-word sentence. Then a long one that wanders and reconsiders itself halfway through.
4. A wrap-up that restates everything
"In conclusion, by following these steps you can achieve your goals." A summarizing final sentence that adds nothing new is a classic close.
5. Hedging filler
Phrases like "it is important to note," "in today's fast-paced world," and "when it comes to" pad the text without saying much. They are statistically common in model output.
6. Generic transitions
"Furthermore," "Moreover," "Additionally," stacked at the start of paragraph after paragraph, signal a model marching through points rather than a person thinking.
7. Invisible characters
This is the one you cannot see. AI text often carries zero-width spaces, non-breaking spaces, and other hidden Unicode that paste along with the words. Most human-typed text does not contain them, so their presence is a quiet fingerprint.
How to remove the tells
The structural tells (rhythm, lists, summaries) take editing. The formatting tells are mechanical and fast to fix:
- Replace em dashes with commas or periods.
- Straighten smart quotes and fix the ellipsis glyph.
- Strip the invisible characters and normalize odd spaces.
A cleaner like textscrubr handles the mechanical half in one pass and keeps your real formatting intact, so you can spend your effort on the writing itself rather than hunting for a character you cannot see.
The takeaway
No single tell is proof. Together, they form a pattern readers recognize. Edit the structure to sound like you, then clean the formatting to remove the residue, and the text reads as human because the thinking behind it is.