AI tell · grade B
Assistant tells: the chatbot preamble and sign-off left in text
Assistant tells are the conversational scraps a chatbot wraps around its answer: the praise that opens a reply, the offer to do more that closes it. They appeared in every model we logged. Readers flag them because they are addressed to a prompt, not a reader, so their presence is direct proof the text was pasted unedited from a chat window. The fix is to delete them: your document has no one to thank for a good question and no follow-up to volunteer.
What it is
The framing a chatbot adds around the actual content. An opener that compliments the request, a closer that offers to expand or revise. Useful inside a conversation, meaningless inside a finished document, and unmistakable when left in.
How often it shows up
Validation openers showed up in all 12 of 12 models we logged (emp-001); engagement-hook closers in 9 of 12 (emp-007). This is one of the least-disputed tells anywhere: leftover assistant framing is treated as proof of an unedited copy-paste. The checklist grades it B.
Why readers flag it
These lines address a person who is not the reader. A finished essay does not open by praising a question nobody asked it, and a report does not close by offering to revise itself. The moment a reader hits one, they know exactly where the text came from.
The fix
Cut the first sentence if it evaluates the request, and cut the last one if it offers further help. Neither carries content. What remains is the answer, which is the only part you meant to publish.
Examples
Great question! Here's a breakdown of the quarterly numbers. Let me know if you'd like me to go deeper on any section. Happy to help!
Here is the breakdown of the quarterly numbers.
Questions
Are these the easiest AI tells to catch?
Yes. Unlike style tells, leftover assistant framing has no innocent explanation in a finished document, so a reader treats it as certain proof rather than a hint.
underslop marks this tell and the rest of the consensus set in any draft you paste, using the same open lint that scores it here. The free tier caps at 500 words a run, and your text is held for 0 seconds.