Analyze interactions with AI agents

Last updated: May 21, 2026 05:36

Agent Analytics helps you understand how visitors interact with AI-powered conversational interfaces in your product. You can identify common use cases, track trends over time, and explore real conversations and prompts submitted by visitors.

This article shows how to navigate the analytics view in Pendo, explore common themes in agent usage, and interpret the data captured through prompts and conversations.

Prerequisites

To use Agent Analytics for AI agents:

Your subscription must include Agent Analytics. If you don’t pay for Agent Analytics, you can analyze up to 500 visitor-submitted prompts each calendar month. After you reach that limit, Agent Analytics stops collecting new prompt data until the quota resets on the first of the month at 00:00 UTC. For more information, see Usage limits and prompt quotas.
You must be either a subscription admin or assigned the AI Agent Admin role to view detailed prompts and conversations for AI agents.
To see conversation-level metrics and charts that rely on full transcripts (including conversation depth, feedback charts, suggested prompt usage, tool usage, and model usage), your implementation must send conversation data using the Conversations API.

Access the analysis view

To explore how visitors are interacting with your AI agents:

Go to Product > Agent Analytics.
From your agent list, select the agent you want to analyze. If your subscription has more than four agents, select Show more to view the full list.

The Overview tab opens by default. You can switch to the:

Use cases tab to explore emergent topics grouped by similarity and monitor tracked use cases. For more information, see Analyze and track use cases for your AI agents.
Issues tab to review emergent issues grouped by similarity and monitor tracked issues. If you're analyzing a prompt-only agent, this tab contains sample data. For more information, see Identify issues with your AI agents.
Conversations tab to review full interaction threads if conversation capture is activated. If you're analyzing a prompt-only agent, this tab contains sample data. For more information, see View full conversations in this article.
Experiments tab to compare agent versions side‑by‑side and measure how changes in models, tools, or configurations affect key metrics. For more information, see Compare AI agent versions with experiments.

In each tab, you can:

Use the date range, segment, account, user reaction, file uploaded, and used suggested prompt filters to shape your query.
Select Save as a report to revisit your current view later.
Select Actions > Analyze full user flow to explore paths and funnels related to the agent.

The Overview tab is organized into several collapsible sections. If your subscription uses a monthly prompt quota, metrics based on visitor‑submitted prompts only include data collected before you reach that limit. For more information, see Usage limits and prompt quotas.

Measure adoption

The Adoption section displays metrics based on your selected filters, with trend indicators compared to the previous period:

Conversations. The total number of threaded conversations between visitors and the agent. Available only for agents capturing full conversations.
Prompts. The total number of prompts submitted by visitors.
Conversation depth. The average number of visitor-submitted prompts in each conversation, calculated by dividing the total number of prompts by the total number of conversations. This metric helps you understand conversation length and interaction effort. Available only for agents capturing full conversations.
Retention. The percentage of visitors who submitted prompts in at least two different weeks within the selected date range, out of all visitors who submitted prompts during the same period.
Unique visitors. The total number of visitors who submitted prompts.
Unique accounts. Number of accounts associated with visitors at the time they submitted prompts.

Select any metric, or select View trends below the metrics, to open a trending chart that shows how it changes over time. You can then hover over a data point in the chart for more details, as well as adjust the time period using the dropdown menu in the top-right corner of the chart.

Measure AI quality

The AI quality section displays metrics based on your selected filters with trend indicators compared to the previous period. You can select any metric, or select View trends below the metrics, to open a trending chart that shows how it changes over time. These are available only for agents capturing full conversations:

Issue rate. The percentage of conversations that contain at least one detected issue, calculated by dividing the number of conversations with issues by the total number of conversations.
Rage prompt rate. The percentage of conversations that include at least one prompt showing strong frustration signals, such as profanity, repeated attempts to complete the same task, or text entered in all capital letters.
Unsupported request rate. The percentage of conversations where the agent couldn't fulfill the visitor's request.
Error rate. The percentage of conversations that encountered an error.

Measure user sentiment

The User sentiment section includes a Feedback chart that summarizes how visitors react to AI agent responses. It shows totals for positive and negative feedback, the reaction response rate (the percentage of agent responses that receive a reaction compared to all agent responses), and the average depth of conversations with positive and negative feedback.

Select the positive or negative segments of the chart to view a list of the associated conversations, visitors, and accounts.

This chart is available only for agents that are configured to send feedback data through the Conversations API.

Analyze prompt behavior

The Prompt behavior section includes charts that summarize how often suggested prompts are used:

Suggested prompt usage chart. Shows the percentage of conversations that include suggested prompts compared to conversations that were fully user‑generated. Hover over a data point to view more details.
Top suggested prompts chart. Lists the suggested prompts most frequently used in conversations, including the number of times each prompt is used and its percentage of total suggested prompt usage. Select a prompt from the chart to view the associated conversations, visitors, and accounts.

These charts are available only for agents that are configured to send suggested prompt data through the Conversations API.

Analyze models and tools

The Models and tools section includes the following charts to understand which models and tools your AI agent relies on most:

Model usage chart. Displays the distribution of AI models used to generate agent responses during the selected date range. Hover over a data point to view more details.
Tools used chart. Shows all tools called during conversations, including the number of calls for each tool and their percentage of total tool usage. Select a tool from the chart to view a list of the associated conversations, visitors, and accounts.

These charts are available only for agents configured to send model and tool call data through the Conversations API.

Analyze top users

The Top users section includes separate tables for the top 10 visitors and accounts based on the number of prompts submitted to the agent.

Each table includes:

Visitor or account name (displays ID if metadata is unavailable).
Number of prompts submitted.
The option to view all visitors or accounts in the top-right corner of each section.

View visitor prompts

If you’re assigned the AI Agent Admin role (or are a subscription admin), you can view visitor-submitted prompts in the Prompts table at the bottom of the report.

By default, the table displays the most recent 1,000 prompts. To view older data, toggle Show all prompts. Loading all prompts might take longer and impact table performance.

The table includes the following details:

Visitor name or email of the person who submitted the prompt (displays Visitor ID if metadata is unavailable).
Prompt text entered by the visitor, with the option to watch the relevant replay if you have Session Replay added to your subscription.
Account name associated with the visitor at the time of submission.
Date entered for the prompt.
Rage prompt. Indicates whether the prompt shows strong frustration signals.
Use case the prompt is currently associated with in the report.

You can use the search bar at the top of the table to filter prompts by keyword. To export the data, select the download icon to save all table contents as a CSV file to your device.

Tip: Admins can view the total number of visitor prompts for all of your subscriptions in your organization settings. For more information, see Track AI prompts in organization settings.

View full conversations

If your agent is configured to capture full conversations and you're assigned the AI Agent Admin role (or are a subscription admin), you can access the Conversations tab.

This tab includes:

Filters. Date range, segment, and account filters. You can also add filters for user reaction, file uploaded, and used suggested prompt. Select Update to apply changes.
Conversations table. A table of individual conversation threads that you can open to view full back-and-forth exchanges between visitor and agent. Select the column icon in the top-right corner of the table to show or hide specific columns. Columns include:
- First relevant prompt. The first visitor prompt associated with the conversation.
- Visitor. The visitor who initiated the conversation.
- Number of interactions. The total number of prompts and agent responses exchanged within the conversation.
- Feedback. Any feedback submitted during the conversation, shown as positive, negative, or mixed.
- Date started. The date and time when the conversation began.
- Rage prompts. The number of prompts within the conversation that show strong frustration signals, such as profanity, repeated attempts to complete the same task, or text entered in all capital letters.
- File uploaded. Indicates whether one or more files were uploaded during the conversation.
- Suggested prompt. Indicates whether the conversation includes a prompt that was suggested rather than manually entered by the visitor.
- Tools used. The total number of tool calls made by the agent while responding during the conversation.
- Models used. The number of AI models used to generate responses within the conversation.

Select any row to open the conversation panel. This panel includes:

Visitor. The name or email of the visitor associated with the conversation (if available), or their unique identifier (Visitor ID).
Prompt initiation timestamp. The date and time when the first prompt in the conversation was submitted.
Replay access. An option to Watch replays if you have Session Replay added to your subscription. When viewing a replay, agent interactions are marked with annotations on the replay timeline.
Visitor prompts. Each prompt submitted by the visitor, displayed with timestamps.
Agent responses. Responses generated by the agent, displayed with response time in seconds.
Conversation indicators. Visual indicators showing whether:
- A suggested prompt was used.
- Positive or negative feedback was submitted.
- A file was uploaded.
- A prompt was flagged as a rage prompt.
Response metadata. Buttons to view which tools were called and which AI models were used to generate a response.
Conversation links. An option to copy a link to a specific message within the conversation to share with others. Shared links open directly to the selected message for users with the appropriate permissions (subscription admin or assigned the AI Agent Admin role).

Note: Some conversation details, such as feedback, file uploads, suggested prompts, tools used, and model information, are available only if your implementation sends this data using the Conversations API. If these optional fields weren’t included during setup, those details won’t appear in conversation views. To learn how to include these optional fields, see Capture full conversations with the Conversations API.

Save and share reports

When you ask a question to group prompts or conversations into use cases, we recommend saving the results as a report to preserve your filters and queries and revisit the data later.

As new data comes in, use cases can shift over time.

Select Save as a report from the Overview or Conversations tab.
Make any needed changes to the auto-populated report name and description.
If you want others to access and edit this report, choose Everyone for Visibility.
Select Save report.

Reports appear in the Reports list in Product > Agent Analytics and can be shared with others in your subscription.

Note: Use cases are only available for tagged agents with at least 25 visitor-submitted prompts in the selected date range. If your filters include more than 10,000 prompts or conversations, Pendo automatically samples the data for performance reasons. When this happens, a tooltip appears next to the Prompts or Conversations column header to let you know the results are based on a sample.

Use AI agent prompts in paths

You can use AI prompt events in path reports to analyze what visitors in your application do before or after engaging with generative AI.

To get started:

In the top-right corner of the page, select Actions > Analyze full user flow > Path.
The path defaults to analyzing events Leading to the AI agent. If you'd rather understand visitor behavior after interacting with an AI agent, update the Show paths dropdown to Starting from.
Adjust filter queries as needed. For more information on creating and interpreting a path report, see Paths.

Use AI agent prompts in funnels

You can use AI agents in funnel reports to analyze how interactions with generative AI affect user flow and conversion.

To get started:

In the top-right corner of the page, select Actions > Analyze full user flow > Funnel.
Choose AI agent as the event type for any step, and select the agent.
Continue filling out the report and adjust each filter query as needed. For more information on building and interpreting funnel reports, see Funnels.

Measure stickiness of AI agents

You can analyze how often visitors return to interact with your AI agent by using AI agent events in the Stickiness Metric dashboard widget. Stickiness measures the frequency of visitor or account activity over daily, weekly, or monthly timeframes, helping you understand whether your AI agent is driving recurring engagement.

To get started, open a dashboard and add the Stickiness Metric widget, then select an AI agent as the event you want to measure.

Tip: You can filter the event list to show only AI agent events, which makes it easy to find and select your agents when configuring the Stickiness widget.

For more details on configuring and interpreting stickiness, see Measure stickiness.

Add an AI agent report to a dashboard

After you save an AI agent report, you can add a corresponding dashboard widget to an existing dashboard. This allows you to track agent data alongside other reports, such as paths or funnels.

To add Agent Analytics to a dashboard:

Select Add to dashboard > Widget from any dashboard, then select the AI Agent widget from the catalog. Choose an existing dashboard to add three new widgets, including data from the Key metrics, Use cases, and Prompts sections of the report.
Select the option to add a dashboard within a supported section of any existing AI agent report, then select the dashboard to which you want to add the widget.

Each widget reflects the saved filters and date range from the original AI agent report. Dashboard-level filters aren't currently supported for these widgets.