Skip to main content

Intro to topic publishing

Introduction of the concept of publishing topics and how-to.

Dorin avatar
Written by Dorin
Updated over a month ago

Publishing topics controls what Fullview AI is allowed to handle for your customers. It’s a quality and safety gate: you test a topic internally, and only Published topics are attempted by the AI for real users. Everything else is not attempted and will be routed based on your setup.


What’s a “Topic”?

A Topic is a cluster of similar requests (e.g., “Add a new employee”). Topics collect real customer requests and training runs in one easy place so you can review accuracy before enabling the AI for customers.


Unpublished vs. Published

  • Unpublished topics

    • All topics are by default unpublished.

    • Visible only to your team.

    • Whitelisted admins can freely test any topic by asking questions or using “Watch” to replay sessions from AI Library.

    • Customers are not affected - the AI will not attempt answers or on-screen guidance for unpublished topics.

  • Published topics

    • The AI is allowed to answer and guide customers for this topic.

    • Appears in production immediately.

If a customer asks about a topic that is not Published, the AI won’t attempt it.


Who can test before publishing?

Whitelisted admin users can ask any question internally to probe the model’s behavior and iterate on training without exposing customers to draft behavior.


Simply add the users email under "Whitelist test accounts" in the AI Agent settings.


Where to publish

Open any topic. In the top-right, you’ll see its current status and a Publish button (see screenshot).
After publishing, the control toggles to Unpublish if you ever need to pause the topic.


Recommended review workflow

  1. Have a whitelisted user ask Fullview AI your desired questions (e.g start with top 20 FAQs).

  2. Replay examples

    • Click Watch to review how the AI handled real requests.

  3. Improve coverage & accuracy if needed

  4. Click Publish when you are happy with the performance of Fullview AI for that topic allow production attempts for this topic.


Unpublishing

Need to roll back? Open the topic and click Unpublish. Fullview AI will immediately stop attempting this topic for customers; you can continue testing internally as a whitelisted admin.


FAQs

What happens if a user asks something outside all Published topics?
The AI will not attempt it. Your workspace’s fallback (e.g., handoff to an agent) will apply.

Does publishing affect past conversations?
No. It governs what the AI will attempt going forward.

How do I get whitelisted admin access?
Add the user emails in your Fullview AI Agent settings page (make sure its the same email as their user login in your own product).


Best practices

  • Start narrow: publish high-volume, well-documented tasks first.

  • Add authoritative Articles and structured Human training flows if not already added.

  • Regularly review Unresolved cases and update training before (re)publishing.

  • Unpublish fast if accuracy dips; fix, retrain, republish.


By using Publishing Topics, you maintain accuracy, quality control, and safety—while giving your team room to iterate privately until a topic is truly ready.

Did this answer your question?