Who owns an AI-drafted reply in a shared inbox?

When an AI suggests a reply and a teammate sends it, and the reply turns out to be wrong, whose mistake is it? Teams that start using AI drafting in a shared inbox hit this question within weeks, usually right after the first awkward email goes out.

The short answer is simple: the person who clicks send owns the email. The longer answer involves a handful of practical questions about how drafts are treated, who edits what, and how to talk about AI use internally. Here they are, in roughly the order teams run into them.

The basics#

Who is responsible for an AI-drafted reply?#

The person who sends it. Not the AI, not the vendor, not whoever configured the tool.

This isn't a legal argument; it's an operational one. A shared inbox only works when every sent message has a human owner who can answer for it, follow up on it and correct it. If "the AI wrote it" becomes an acceptable explanation, nobody owns anything, and quality slides.

It also matches how good tools are built. In Koltrix, for example, AI can draft a reply, but nothing is sent without a person clicking send. The click is the moment of ownership.

So is the AI draft just a suggestion?#

Yes, and it helps to say that explicitly to the team. A draft is a starting point with the same status as a snippet or a colleague's rough notes. You can send it as-is, edit it heavily or throw it away. None of those choices needs justifying.

What matters is that the sender reads it as if they'd written it. If you wouldn't sign your name to a sentence, don't send it.

What if the draft contains something the sender didn't notice?#

Then it's the sender's mistake, handled the same way as any other mistake: correct it with the customer, figure out how it slipped through, and adjust. Common culprits are invented details (a feature that doesn't exist, a date nobody committed to), the wrong customer name, or a promise that only a manager can make.

The fix is usually a review habit, not blame. A short checklist helps: right person, right facts, right promises, right tone.

Collaboration questions#

Can two people work on the same AI draft?#

They can, but they shouldn't do it at the same time. Two people editing one draft simultaneously is a recipe for a mangled reply or a double send.

A simple rule works for most teams: whoever opens the draft first owns it until they send it or explicitly hand it over. If someone else has a suggestion, they say it in team chat or a comment, and the owner decides.

What if one person generates a draft and another sends it?#

That's fine as long as the handover is explicit. "I've drafted a reply to the refund thread; can you check the amount and send it?" makes the sender the owner, and they should review it as carefully as if they'd written it.

The failure mode to avoid is an orphaned draft: generated by one person, left sitting, and later sent by someone else who assumed it had been checked.

Who should edit a draft for a thread someone else owns?#

Nobody, unless the owner asks. If your team has a convention for who owns a thread (by topic, by customer or by whoever replied first), AI drafts follow the same convention. Generating a draft doesn't give you ownership of the conversation.

AI drafting in the shared inbox

1. The person who sends a reply owns it, however it was drafted.
2. Read every AI draft fully before sending. Check names, facts,
   numbers, links and promises.
3. Never send an AI draft that makes a commitment you can't make
   yourself (refunds, discounts, dates, exceptions).
4. One person works on a draft at a time. Hand over explicitly.
5. Thread ownership rules apply to AI drafts the same way.
6. Customer disclosure: [our approach].
7. Questions or problems: raise them with [name].