Guide · AI writing · October 2026 · by Stefan Tsezarov

AI Leftovers

The tiny AI mistakes you don't notice until you send them.

You asked an AI assistant for an email, a customer reply or a post.

You copied it. You pasted it. You hit send.

And somewhere in the message was:

Certainly! Here is a message you can send:

The text itself was fine. The problem was the piece of AI conversation that came along with it.

Those are AI leftovers.

01Text meant for you, not your reader

An AI leftover is part of the chatbot's answer that was addressed to you, not to the person who will read your message.

Ask for a polite customer reply and you might get:

Certainly! Here is a professional reply you can send: Thank you for reaching out. We have reviewed your request and will get back to you shortly. Let me know if you'd like any changes to the tone.

The middle paragraph is the reply. The first line is the assistant introducing it. The last line is the assistant offering another round.

Copy all of it, and your customer is invited to help edit the AI's answer.

02The four common leftovers

Self-reference

As an AI language model, I can't share personal opinions.

This belongs in a chat with an assistant, almost never in your email or report.

Chat preambles

Of course — here's a professional version:

Useful inside the chat. Unnecessary once you have the actual message.

Chat sign-offs

Would you like me to make it more professional?

Normal at the end of a chat. Odd at the end of a message to a customer.

Unfilled placeholders

Dear [Recipient's Name],

Placeholders such as [Your Name] or {{first_name}} help while drafting. Forgotten, they make the message look like an unfinished template.

And sometimes: formatting debris

Assistants format in Markdown, so **bold** or ### can arrive as plain symbols. People also use Markdown on purpose, so on its own it means little. Next to a preamble or a placeholder, it is part of the same story.

03Why you miss them

An assistant answers in three parts: an introduction, the answer, an offer to revise it.

Your eyes find the long readable paragraph, recognise the meaning, and move on. The short line at the start or end becomes invisible.

There is no spelling mistake and no broken sentence. It is simply the right text for the wrong reader.

Until somebody else reads it.

04This is not AI detection

Finding a leftover does not prove a text was written by AI. A person can leave a placeholder in a hand-written email, and an article about AI can quote "as an AI language model" on purpose.

Finding none does not prove a person wrote it either. No tool can reliably tell AI from human writing, and this one does not try.

The useful question is smaller:

Did something from the AI conversation end up in the final text?

That is the question the check answers.

05Try it yourself

Fictional example · as copied

Certainly! Here is a professional version: Hi Maria, Thank you for your message. I've reviewed the information and will follow up tomorrow. Let me know if you'd like me to make this warmer or more concise. Best, [Your Name]

Three things do not belong: the introduction, the offer to revise, and the placeholder.

Fictional example · cleaned

Hi Maria, Thank you for your message. I've reviewed the information and will follow up tomorrow. Best, Stefan

Paste the first version into Message or prompt in the scanner. Each leftover is listed under "AI-assistant leftovers", marked informational: it does not change the verdict, and the result says plainly that it cannot tell AI from human. The cleaned version shows none.

The check runs in your browser. The text is not sent anywhere.

06Check before you send

One rule: read the text once as the recipient, not as the person who created it.

Is the assistant talking to you anywhere? Is a placeholder left? Did formatting escape from the editor?

The goal isn't to hide that you used AI.

It is to send the message you actually meant to send.

The check looks for common leftover patterns, not every possible one. A leftover written in unusual words can pass unnoticed.

Scan a message →