> ## Documentation Index
> Fetch the complete documentation index at: https://support.myaskai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Test before going live

> Test repeatable questions, real helpdesk tickets, and a controlled set of direct replies.

Use several test methods because each method checks a different part of the setup. Start with repeatable questions, then test the helpdesk connection, then allow a small set of direct replies.

### Days 1 to 3

Use Test and the Dashboard widget. Check answer content and guidance.

### Days 3 to 10

Use Note replies. Check real tickets and helpdesk behavior without customer-visible replies.

### Day 10 onward

Use controlled direct replies. Check live customer behavior within a clear limit.

## Test a repeatable question set

In the Dashboard, go to **Improve > Test**.

1. Use tickets imported during onboarding, add questions manually, or upload a one-column CSV.
2. Upload no more than 50 questions in one CSV.
3. Build a set of 50 to 150 real support questions as a starting point.
4. Score whether you would send each answer to a customer.
5. Record what needs to change.
6. Update Knowledge, Guidance, or Custom Answers.
7. Refresh the affected answers and compare the result.

Read [Test](/features/improve/test) for the full feature guide.

## Use the Dashboard widget for one question

Use the Dashboard widget for a quick check while you set up the agent. Use **Improve > Test** when you need a repeatable set or more than five questions.

Check both Public and Private content behavior when it applies to your agent. Read [Public and Private modes](/features/private-internal-mode).

## Test real tickets with Notes

Note mode writes a proposed answer that only your agents can see. It tests the connected helpdesk without sending an AI reply to the customer.

Note replies support Intercom, Zendesk tickets, Gorgias, Freshdesk, Freshchat, and HubSpot. Zendesk Messaging does not support Note replies.

Review the Note, its sources, and the ticket context. Fix the agent before you copy or send a poor answer.

## Release a few topics

Use AI Tagging to allow direct replies for a small set of low-risk topics. Keep other topics in Note mode or block them.

AI Tagging supports Intercom, Zendesk tickets, Freshdesk, Freshchat, and Gorgias. It does not support Zendesk Messaging or HubSpot.

Review the replies in **Improve > Inspect & Logs**. Add another topic only after the current topics meet your target.

Read [Prevent AI replies to certain topics](/start-here/playbooks/prevent-ai-replies-to-certain-topics).

## Use a short direct-reply window

Turn on direct replies for a quiet period when your support team can monitor the conversations. Turn them off at the agreed end time.

Record who owns the test, the start time, the end time, and the review criteria.

## Reply to the first message only

Use **Reply to 1st message only** for Zendesk tickets, Freshdesk, or Gorgias email. The AI agent replies once, then hands later customer messages to your team.

This option does not support Intercom email or chat.

Check the setup page for [Zendesk tickets](/features/channels/zendesk/zendesk-tickets), [Freshdesk](/features/channels/freshdesk), or [Gorgias](/features/channels/gorgias).

## Test with Copilot

Use the Chrome extension when an agent must control every message. The extension reads the ticket history and inserts a draft in the reply box. The customer sees nothing until the agent sends the draft.

<Info>
  The Copilot will not be able to use User Data APIs or any other APIs to access user or customer information, so only use for knowledge-based questions.
</Info>

The draft reply button supports Intercom, Zendesk, Gorgias, and Freshdesk. Read [Copilot](/features/channels/chrome-extension).

## Related playbooks

* [Gradually roll out AI support](/start-here/playbooks/gradually-roll-out-ai-support)
* [Improve a specific AI response](/start-here/playbooks/improve-a-specific-ai-response)
* [Prevent AI replies to certain topics](/start-here/playbooks/prevent-ai-replies-to-certain-topics)
