# Use AI Assistant

Source: https://amplitude.com/docs/assistant/use-assistant

---

On this page

- [Ask a question](#ask-a-question)
- [Continue a conversation](#continue-a-conversation)
- [Give feedback on answers](#give-feedback-on-answers)
- [Inspect answers with X-ray](#inspect-answers-with-x-ray)
- [Open X-ray from a chat](#open-x-ray-from-a-chat)
- [Understand the Thinking view](#understand-the-thinking-view)
- [Understand the Sources view](#understand-the-sources-view)
- [Analyze chats and improve AI Assistant](#analyze-chats-and-improve-ai-assistant)
- [Find problem areas](#find-problem-areas)
- [Fix gaps](#fix-gaps)

# Use AI Assistant

How your users open AI Assistant depends on how you configure the widget, but the basic pattern is:

1. **Find the AI Assistant widget launcher**

   Look for a help or chat icon, often in the bottom-right corner of the page. Your team may label it with your brand, for example, "Help", "Need assistance?", or your company name.

2. **Open the widget**

   Select the launcher. The AI Assistant widget opens as a panel on the page. Depending on configuration, users may see:

   - A **Chat** tab where they can talk with AI Assistant.
   - A **Resource Center** tab with recommended and searchable content.
   - Both, as separate tabs in the same widget.

3. **Switch between Chat and Resource Center**

   If your team enabled both, use the tabs at the top of the widget, for example, **Chat** and **Resources**, to switch modes:

   - **Chat** for conversational help.
   - **Resources** for browsing and searching articles, guides, and other content.

## Ask a question

On the **Chat** tab, users can ask AI Assistant any question that your connected content and tools can answer. Typical use cases include:

- "How do I invite a teammate?"
- "Why is my dashboard empty?"
- "Show me a quick tour of the new feature."
- "Where can I find your billing documentation?"

To ask a question:

1. Select the **message box** at the bottom of the **Chat** tab.
2. Type your question in plain language.
3. Press **Enter** or select **Send**.

AI Assistant processes your request and then:

- Streams its answer into the chat.
- May include links to relevant docs or Resource Center items.
- May open an in-app guide, checklist, or survey that walks you through a flow.

## Continue a conversation

AI Assistant keeps track of context within a single chat:

- You can ask follow-up questions without repeating yourself.
- AI Assistant can refer back to earlier messages in the same conversation.

For example:

1. You: "How do I set up SSO?"
2. AI Assistant: Explains the process and links a setup guide.
3. You: "Can I require this for all users?"
4. AI Assistant: Builds on the previous answer to cover enforcement.

If you want to switch topics completely, start a new question in the same chat. AI Assistant keeps context within a conversation, not across unrelated threads.

## Give feedback on answers

End users can help improve AI Assistant by giving feedback:

- **Thumbs up** a helpful answer so your admins know it worked well.

- **Thumbs down** an unhelpful answer. AI Assistant (and your admins) use this to improve:

  - AI Assistant may try an alternative answer or ask for clarification.
  - Your admins can review that this answer didn't work and adjust content or configuration.

## Inspect answers with X-ray

AI Assistant includes a detailed inspection view that explains how AI Assistant produced each answer.

At a high level, X-ray lets you:

- See which content sources AI Assistant used to generate an answer.
- See which tools or workflows AI Assistant called.
- Read a short reasoning explanation of how AI Assistant derived the answer.
- Audit both live test chats and historical user chats.

### Open X-ray from a chat

There are two main entry points:

1. **While testing in the AI Assistant admin area**

   When you test AI Assistant directly in its configuration view:

   - Start a conversation on the right side of the screen.

   - Open the **X-ray** tab for that conversation.

   - For each AI Assistant message, open the details to see:

     - The **Thinking** timeline (tool calls and intermediate reasoning).
     - The **Sources** that AI Assistant retrieved and cited.

2. **From historical chats in analytics**

   From the AI Assistant or chat analytics area:

   1. Filter or search to find a chat you care about, for example, a chat that hit a fallback or received negative feedback.
   2. Open the chat detail view.
   3. Select the **X-ray** or **Inspect** button.
   4. Review each response in the same way: thinking, sources, and (where applicable) workflows.

### Understand the Thinking view

The **Thinking** timeline gives a chronological account of how AI Assistant handled the message.

Use this view to:

- Verify that AI Assistant calls the right tools in the right order.
- Check that AI Assistant selects workflows when it should.
- Spot cases where AI Assistant skipped a tool you expected it to use.

### Understand the Sources view

The **Sources** tab shows what information AI Assistant had access to when it wrote its answer. Typical information includes:

- **Cited sources** — passages and Answers that most directly shaped the answer.
- **Other considered sources** — content that AI Assistant retrieved and evaluated but didn't directly cite.
- **Per-source details** — such as the title, URL, and content section.

Internally, AI Assistant stores point-in-time snapshots of the relevant passages so you can see what it "saw" even if someone edited or removed the underlying article later.

- Confirm that AI Assistant uses the right documents.

- Identify when outdated or incorrect content drives bad answers.

- Decide whether to:

  - Update an article.
  - Create an Answer with a clearer answer.
  - Adjust content tags or indexing.

## Analyze chats and improve AI Assistant

AI Assistant's value increases as you iterate. Use chat analytics plus X-ray to improve coverage and quality.

### Find problem areas

From the chat analytics area:

1. Filter chats by:

   - High fallback rate.
   - Negative feedback.
   - A specific topic or keyword.

2. Open individual chats and inspect them with X-ray to see:

   - Which sources AI Assistant used.
   - Which tools AI Assistant called.
   - Whether the answer was incomplete or incorrect.

### Fix gaps

Based on what you find:

- **Update or add content in your docs or help center.**
- **Create Answers for recurring high-importance questions.**
- **Adjust workflows or tools if AI Assistant isn't choosing the right path.**
- **Tighten or expand targeting for Resource Center content so articles appear when and where users need them.**
- **Create a guide for important or confusing product areas.**

Over time, this loop—chat, X-ray, then content or configuration updates—turns AI Assistant into a high-quality, product-specific helper rather than a generic chatbot.

Was this helpful?

<!--$-->

<!--/$-->