What is a draft-only reply?

In one sentence

A draft-only reply is a message your AI agent writes but does not send: it waits as a draft until a person reads it and approves, edits or rejects it. It lets you get the speed of AI writing while keeping the final say on every word that reaches a customer.

In more detail

Picture a new member of staff on their first week. You let them write replies, but you ask them to show you each one before it goes. You read it, maybe change a word, and say "send it". A draft-only reply is the same arrangement with an AI agent.

The agent still does most of the work. It reads the message, looks at your notes, works out the answer and writes it in your tone. The only thing it does not do is press send. The reply sits in a queue until someone approves it.

When you approve, the reply goes out exactly as written, or with the changes you made. When you reject it, it is never sent, and you can tell the agent why so it does better next time. Either way, nothing reaches the customer without a person deciding.

Draft-only does not have to apply to everything. You might let the agent answer opening hours on its own but hold anything about prices, refunds or complaints. Or you might hold posts and comments, which everyone sees, while letting private replies go straight out.

The trade-off is time. A held reply only goes out when someone approves it, so customers wait for you as well as for the agent. That is why draft-only works best when someone checks the queue often, or for the kinds of message where getting it right matters more than getting it fast. The example below shows a morning with draft-only replies in a salon. Names and details are examples.

An example

Draft-only, sending on its own, or a set reply

Most businesses end up using all three: set replies for the questions with one right answer, the agent on its own for routine questions, and draft-only for anything that touches money, feelings or promises.

How to review drafts quickly

A draft queue only works if it does not pile up. These habits keep it moving.

1. Pick set times. Check drafts first thing, at lunch and before you close, plus whenever you have a spare minute.
2. Ask for the list with context. Ask your agent "show me the drafts waiting, with the customer's message above each one", so you can judge each reply in seconds.
3. Approve the easy ones in a batch. If five drafts are about opening hours and all read well, approve them together.
4. Edit rather than reject when it is nearly right. A changed word is quicker than a fresh draft.
5. When you reject, say why. "Too formal" or "we do not offer that" helps the agent next time, and you can add the point to its notes.
6. Hear about new drafts. Have a notification sent to your team chat or phone when something is waiting, so urgent ones are not missed.

Sample notes for an agent that holds replies

Your notes can tell the agent what to hold and how to make drafts easy to check. Here is an example. Names and details are examples.

Hold every reply as a draft for me to approve.

For each draft, show me:
- who it is and which account (WhatsApp or Instagram)
- their message, in full
- your reply
- one line on anything you were unsure about

If you are not sure of a price or a date, say so in the draft rather than
guessing, for example "I will check and come back to you today".
If a message is a complaint, do not draft a reply. Hand the chat to me.

When draft-only is the right choice

• In your first weeks with an AI agent, while you learn how it writes.
• For anything about money: quotes, discounts, refunds and deposits.
• For public posts and comments, which everyone can see and which are hard to take back.
• When you connect a new channel or a new account, until you are happy with the replies there.
• In regulated or sensitive work, such as health, legal or financial questions, where a person must check every answer.
• When a new person joins your team and you want them to see how replies are handled.

It is less useful for very simple questions that come in all day, where waiting for approval just makes customers wait. Those are better as set replies or left to the agent once you trust it.

Common mistakes

• Holding everything and checking rarely. Customers wait hours for a reply the agent wrote in seconds.
• Approving without reading. If you click yes on everything, you have the delay of draft-only and none of the protection.
• Never moving anything to send on its own. Once a type of reply is always right, let the agent send it and keep drafts for the harder ones.
• Rejecting without a reason. The agent cannot improve if it does not know what was wrong.
• Forgetting drafts during a holiday. If nobody will check, switch to set replies or a short holding message instead.

Why it matters for your business

• You get replies written for you without giving up control of what is sent.
• Mistakes about prices, promises or tone are caught before a customer sees them.
• It is a gentle way to start with AI, and you can widen what the agent sends alone as trust grows.
• Every approved or rejected draft teaches you, and the agent, what good looks like.
• Public posts and comments, which are hard to undo, can always wait for a person.

How Spaces handles it

In Spaces you ask your AI agent to hold replies as drafts, and nothing goes out until you approve it. There are three settings: hold all writes, which covers replies, new chats, private replies from comments, posts and comments; hold only posts and comments; or hold nothing. You review the waiting drafts through your agent too: ask it to list them, then approve one as it is, approve it with your edits, or reject it so it is never sent.

An approved draft goes out through the normal route, with the same pacing and limits as any other message. If the account is at a limit, or sending is paused, the draft simply stays waiting and you can approve it again later. A scheduled message, scheduled post or routine run can be held the same way, and it shows as waiting for approval until you decide. A webhook can tell your team chat or phone each time a new draft is waiting.

An agent's access can also be made draft-only, so everything that agent writes waits for approval even if the rest of your workspace sends straight away. That suits a new agent or a helper you are still testing. For email, the agent can also save a normal email draft in your inbox for you to check and send yourself.

In short

A draft-only reply is written by AI and sent by a person. It gives you speed with a safety net, at the cost of waiting for someone to approve.

Use it for the messages that matter most and while you are getting started, check the queue often, and move the routine replies to sending on their own once you trust them.

Is a draft-only reply the same as an email draft?

They are similar ideas. An email draft sits in your mail app for you to send yourself. A draft-only reply in Spaces waits in a queue of approvals and is sent for you, on the right channel, the moment you approve it.

How do I turn draft-only replies on in Spaces?

Ask your AI agent to hold replies as drafts. You can hold everything, only posts and comments, or nothing, and you review the drafts by asking your agent to list them.

Can I change a draft before it is sent?

Yes. You can approve it as it is or approve it with your own edited text. The edited version is what the customer receives.

What happens to a draft I reject?

It is closed and never sent. You can give a reason, which helps you and the agent improve the notes it works from.

Will customers wait longer with draft-only replies?

They wait until someone approves, so yes, unless you check often. Set regular review times and get a notification when a draft is waiting.

Can I hold only some kinds of message?

You can hold all writes or only posts and comments with the setting. For finer rules, such as holding only prices or refunds, write them into the agent's notes, or give a separate agent draft-only access for those jobs.

Does an approved draft still follow the sending limits?

Yes. It goes through the same pacing and limits as any other message. If the account is at a limit, the draft stays waiting and you can approve it again later.

Is draft-only a good idea forever?

For sensitive messages, often yes. For simple, repeated questions it usually slows customers down for no gain, so most owners let those send on their own once the replies are reliably right.