What happens if you never answer the AI's draft?
November 18, 2026 · hosting · trust
Nothing is sent. A held draft has no timer, no escalation to "send anyway," and no expiry that quietly discards it. If you never answer, the guest never hears from the AI — the draft simply waits, and the message stays yours to handle. That answer surprises people in both directions. Hosts worried about a runaway robot expect a timeout that fires after an hour of silence; hosts hoping for a safety net expect the system to cover for them when they go quiet. It does neither, on purpose, and the reasons are worth spelling out — along with the one real cost of silence, which is not the one people fear.
The draft just waits
When a reply is held for you, you get one email: the guest's words, the suggested reply, and three ways to answer — SEND, SKIP, or REPLY: with your own words. That email says "nothing has been sent yet," and it stays true until you act. Answer SEND three days later and the reply still goes out exactly as written: a held draft does not expire. Answer SKIP and that decision is terminal — a SEND arriving after it finds nothing left to send, because the skip stands rather than the system guessing you changed your mind. And silence cannot be filled by someone else: an approval only counts when it comes from your address and carries the signed token from the draft's own subject line.
Every uncertain state falls toward silence
Why no timeout? Because a deadline that ends in an automatic send converts "the AI drafts, you decide" into "the AI decides, slowly" — the one thing draft mode promises not to do. A reply you have not seen is a reply you have not approved, and no amount of elapsed time changes that. The same rule runs all the way down into the machinery. If you answer with something other than the three commands, nothing is forwarded and you get a note that your answer was unclear. Even inside a single SEND, the draft is first marked as taken and only then handed to the mail provider — so if anything crashes in between, the draft stays held and unsent until a human looks, instead of maybe-sent. Every ambiguous state resolves the same way: toward nothing said in your name. A message a guest should not have received cannot be taken back; an unsent one costs only time.
What silence actually costs
Only time — but time is not free. While the draft waits, the guest is still waiting too, and Airbnb's response-rate clock runs on first replies within 24 hours. The AI holds its draft indefinitely, but it cannot hold that clock. So the honest description of draft mode is: every guest message costs you one decision, usually one word, and the decisions you never make are guests who never got an answer. The failure mode of ignoring the AI is not an unauthorized message — it is the same slow reply you were at risk of before the AI existed.
Two honest ways out
The first is to answer the one email. The loop is built to make that cheap: everything you need to decide is in the notice itself, the answer is a single word, and an approved reply reaches the guest in seconds. The second is to stop holding the routine ones at all. When you have watched enough drafts to trust the answers, autopilot sends the routine factual replies without asking — so silence stops stalling the wifi-password questions, while anything sensitive or uncertain is still held for you at any confidence. What never exists, in either mode, is a third path where an unanswered draft sends itself.
You can watch a draft being held before signing up for anything: paste your listing into the demo, ask a question a guest would ask, and then ask one about a refund — the first gets an answer, the second gets held, and nothing goes anywhere until someone with the right to decide says so.