> ## Documentation Index > Fetch the complete documentation index at: https://docs.outzen.ai/llms.txt > Use this file to discover all available pages before exploring further. # AI reply drafts > How Guhan reads replies, what it does for each kind, and how to choose between drafting for approval, drafting and sending, or leaving replies to you. **Guhan reads every reply, decides what it means, and acts on it. Whether it also writes a response is up to you.** You set one workspace-wide rule, and you can override it at any step of a campaign. Meeting requests are never answered automatically, whatever you choose. ## What Guhan does with each reply | What the prospect said | What Guhan does | | - | - | | **Interested** ("sounds good, tell me more") | Follows your reply policy: hand it to you, or draft a reply. | | **Asked for a meeting** | Always hands it to a person, with a **Schedule** task so you can send a booking link. | | **Asked a question** | Follows your reply policy. | | **Said no** | Stops the sequence for that prospect. Nothing is sent. | | **Asked to unsubscribe** | Stops the sequence for that prospect. Nothing is sent. | | **Out of office** | Pauses the sequence for 7 days, then resumes. | | **Not now** | Pauses the sequence for 30 days, resumes afterwards, and leaves you a note in the Inbox so you can follow up personally if you want. | Guhan never auto-replies to a no, an unsubscribe, an out-of-office, or a not-now. > [!WARNING] > > Stopping a sequence on a no or an unsubscribe does not add the person to your Do Not Contact list. If you want them blocked across every campaign, add their email or domain yourself. See [Do Not Contact](https://docs.outzen.ai/workspace-settings/do-not-contact.md). > [!NOTE] > > When Guhan is not confident that a reply is a no or an unsubscribe, it does not stop the sequence. The thread comes to your Inbox as a **Take over** task so a person decides. ## Reply policy Settings, Sending & Reply Rules, **Reply handling**. This is the default for any step that does not have its own AI reply step on the campaign. > [card] Escalate to a user > > Every interested reply and question creates a task in your Inbox. Guhan writes nothing. This is the default. > [card] AI drafts — user approves before sending > > Guhan drafts a reply. You see it in the Inbox and approve, edit, or reject it before anything is sent. > [card] AI drafts and auto-sends > > Guhan drafts and sends one reply without asking. Riskier. Use it once you trust the tone. Auto-send is limited to a single reply per conversation under the workspace policy. If the prospect replies again, or if a send fails, the thread comes to you as a **Take over** task. ![Workspace settings > Sending rules > Reply handling: the workspace default for replies](https://mintcdn.com/outzen/qi8TKfM9RRqmU7-c/images/se-sending-replies.png?fit=max\&auto=format\&n=qi8TKfM9RRqmU7-c\&q=85\&s=b501de400541b730e0aabda213de8b3a) Workspace settings > Sending rules > Reply handling: the workspace default for replies ## Per-step control: the AI reply step On the campaign canvas, an **AI reply** step placed after a wait-for-reply step overrides the workspace policy for that point in the sequence. Explicit steps always win over the workspace default. It has three settings: | Setting | What it does | | - | - | | **Send for approval** | On: the draft waits in your Inbox for approval. Off: it sends automatically. | | **Max consecutive replies** | How many replies Guhan can send in a row before the sequence moves on (1 to 10). 1 means a single reply. | | **Reply timeout** | How long to wait for the prospect to answer between replies (1 hour to 30 days). | The step has four exits you wire up on the canvas: **Conversation done**, **Meeting requested**, **Not interested**, and **Reply timeout**. Out-of-office and not-now replies pause and resume on their own, and failed sends go to the Inbox for a manual retry. For building the canvas, see [Build a campaign](https://docs.outzen.ai/campaigns/build-a-campaign.md). ## How drafts are written A draft uses the full conversation so far, your [Brand & Voice](https://docs.outzen.ai/brand-and-voice/overview.md), the product the campaign is selling, and what Guhan knows about the prospect. If the product has objection answers and offers set up, the draft can use them. See [Products](https://docs.outzen.ai/brand-and-voice/products.md). When a draft is waiting, you will see it in the thread with a marker showing it was written by Guhan. Approve it as is, edit it, or reject it. See [Handle replies](https://docs.outzen.ai/inbox/handle-replies.md). ## Cost An AI-drafted reply costs 1 credit. Replies you write yourself cost nothing. When a campaign step is set to approval, the credit is charged when you approve and send. See [Plans and credits](https://docs.outzen.ai/workspace-settings/plans-and-credits.md) for every credit cost in one place. ## If something's off > [accordion] A reply I expected Guhan to answer came to me instead > > Meeting requests always come to a person. Beyond that, check Settings, Sending & Reply Rules, Reply handling, and check whether the campaign has its own AI reply step that overrides it. Auto-send also falls back to you after one reply, on a send failure, or when credits run out. > [accordion] Guhan stopped a sequence and it should not have > > Reply meaning is read by a model and can be wrong on short or ambiguous messages. Guhan holds back from stopping when it is unsure, but it can still misjudge a clear-sounding reply. Open the conversation in the Inbox and reply yourself. ## Related > [card] [Handle replies](https://docs.outzen.ai/inbox/handle-replies.md) > > Approve, edit, and take over. > [card] [Build a campaign](https://docs.outzen.ai/campaigns/build-a-campaign.md) > > Add an AI reply step to the canvas. > [card] [Plans and credits](https://docs.outzen.ai/workspace-settings/plans-and-credits.md) > > What an AI reply costs. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.