Automatic replies
Configure answer review and automatic replies for connected channels.
An automated answer is called a deflection in the dashboard. It uses two AI steps: a draft based on retrieved knowledge, followed by a separate review against the available evidence.
How an answer is prepared
- A question arrives through a connected channel.
- answerLoops retrieves relevant knowledge.
- The drafting step prepares a reply from that material.
- A separate review assesses the draft and assigns a confidence score.
- Channel settings determine whether a qualifying answer posts automatically or waits for your team.
Connected repositories can provide additional source material. Useful resolutions can also be promoted into the knowledge base. The website widget uses published knowledge.
Confidence threshold
The default threshold is 0.8 on a scale from 0 to 1. Configure it in the relevant integration's settings.
A higher threshold makes automatic replies more selective; a lower threshold lets more drafts qualify. Neither setting guarantees that the answer is correct. Inspect the draft and its sources when deciding whether to change the threshold.
Automatic Deflections
Automatic Deflections is off by default for newly connected channels and GitHub repositories. A high-confidence draft still requires approval when this setting is off.
- Off: your team reviews the draft and approves it before the answer is sent.
- On: a draft that meets the configured requirements can post automatically.
Other drafts remain in the ticket queue. Human escalation routing adds notifications on eligible plans.
Change the setting
- Open Settings and select the platform or GitHub repository.
- Edit the channel's configuration.
- Set the confidence threshold and Automatic Deflections option.
- Save. GitHub repository cards use an inline checkbox.
- Send a representative question and verify the ticket and reply behavior.
Use different settings for channels with different review requirements. Test against current documentation before enabling automatic replies.
Knowledge search and model configuration
The search step selects relevant content before drafting. A retrieval similarity score and the answer-review confidence score measure different things; a close text match does not by itself establish that the answer is correct.
Configure chat and embedding services in AI Model. If you change the embedding model, re-import knowledge so the stored vectors match it.