AI Feedback

AI Feedback combines customer voice, product behavior, and AI-powered insights into actionable workflows. It monitors feedback sources that you define and delivers insights through Amplitude. AI Feedback groups customer feedback into categories and connects feedback insights to Amplitude cohorts, session replays, and experiments. You can also create guides and surveys based on insights to better understand your customers' needs.

Feedback events collect information about use cases such as:

• Providing feedback about specific product areas.
• Calling out complaints or loved features.
• Feature requests.
• Key takeaways.

For information about how AI uses your data and other compliance details, go to Trust in Amplitude AI.

How Feedback Events work

AI Feedback uses Feedback Events to record user feedback and create insights.

Feedback events capture the following information:

• UserId: The customer's email address.
• timestamp: When the feedback was created.
• Event name: [AI Feedback] Feedback.
• Event properties:
  • feedback_id: Unique identifier.
  • text: Feedback text (first 1k characters).
  • source: Source system such as Zendesk, Intercom, or Gong.
  • insights: Array of insights.
  • timestamp: Date the feedback occurred.
  • author: Feedback author.
  • <AUTHOR_EMAIL>: <AUTHOR_EMAIL> is a documentation notation for whichever field name is configured for the author’s email address. You can specify this property through user mapping.

Setting up a data source

Before you can gain any insights from your customers, you must add at least one data source to Amplitude. A data source is your repository of customer feedback. Popular data sources include Zendesk, Salesforce, Reddit, and others. You can also upload individual CSV or DOCX files as data sources.

For example, if your company has discussions on Reddit, you can set up an integration by pointing AI Feedback directly to the subreddit URL. AI Feedback then monitors the subreddit and provides insights from the posted content. You can also upload customer call transcripts directly to AI Feedback to gain insights from the conversation.

AI Feedback can ingest and generate insights from multiple sources at the same time. After the initial ingestion, AI Feedback pulls feedback once each day.

Add a data source

1. Open Amplitude and click AI Feedback and then Add Source.

2. Select one of:

   • Connect an integration: Lets you connect to an application such as Zendesk, Hubspot, or G2.
   • Upload a file: Lets you upload a static CSV or DOCX file.
   • Paste text: Lets you paste text directly into AI Feedback.

3. If you’re connecting an integration:

   • Search for your application or select it from the list.
   • Depending on the integration, you may need to sign in to that application, specify relevant metadata tags, or point AI Feedback to a specific URL. The connector modal guides you through the necessary steps.

4. If you’re pasting text, paste the content you want analyzed into the field. AI Feedback begins to parse the content automatically.

5. If you’re uploading a CSV or DOCX file, click Upload and then drag and drop the files you want to upload. For CSV files, you'll choose which columns you want to designate as the Feedback information and which columns you want as additional metadata.

For example: You could have a CSV file with the following columns:

In the AI Feedback CSV upload tool, designate which columns to use and click Analyze. You must select the column that includes your feedback.

Connect a Slack channel

Connecting a Slack channel lets AI Feedback automatically aggregate and surface insights from messages posted in that channel, with no manual tagging, copying, or context switching required. Pillar channels, support escalation channels, and customer-facing Slack Connect channels all become structured, searchable feedback sources.

Add a Slack channel as a source

1. Navigate to AI Feedback > Sources.
2. Click Add Source.
3. Search for "Slack" and select it.
4. Authorize Amplitude to access your Slack workspace.
5. Select the channels you want to monitor.
6. Configure the following options:
   • Lookback range: How far back to import messages during the initial connection.
   • Include thread replies: Enable to include thread replies, not just top-level messages.
7. Click Connect.

AI Feedback ingests existing messages from the lookback period, then pulls new messages daily.

Managing a source

After you've added a data source, you can change its settings or delete the source from AI Feedback.

Change a source's settings

1. Go to AI Feedback > Sources.
2. Click the three-dot menu icon next to the source you want to change.
3. Click Settings.
4. If there are settings for your source, you can change them here.
5. Click Back to sources to return to the source menu.

Delete a source

1. Go to AI Feedback > Sources.
2. Click the three-dot menu icon next to the source you want to delete.
3. Click Confirm.

Supported sources

AI Feedback connects to a variety of feedback sources across four categories.

Support and service

• Zendesk: Import support tickets and customer conversations.
• Intercom: Import customer messaging and support interactions.
• Salesforce Service: Import service cases and customer communications.
• Freshdesk: Import helpdesk tickets and customer responses.

Sales and success

• HubSpot: Import CRM feedback, deal notes, and customer communications.
• Gong: Import sales call transcripts and conversation insights.

Social and reviews

• Slack: Import feedback from specific Slack channels.
• App Store: Import iOS app reviews and ratings.
• Google Play Store: Import Android app reviews and ratings.
• G2: Import software review feedback.
• Discord: Import feedback from Discord server channels.
• Trustpilot: Import customer review feedback.
• X: Import mentions and conversations.
• Reddit: Import subreddit discussions and comments.

File upload and other

• CSV: Upload structured feedback data from CSV files.
• DOCX: Upload feedback documents such as call transcripts or research reports.

Viewing insights

The New Insights tab displays AI-generated insights from your connected feedback sources. AI Feedback regenerates these insights each time it runs its analysis, so insights on this tab are ephemeral: they refresh and may change as new feedback arrives and more relevant clusters emerge. To persist an important insight and track it over time, save it as a trend.

AI Feedback counts how many customers share the same feedback for the same feature or functionality. For example, if multiple customers report the same bug, AI Feedback highlights the bug and displays how many mentions it received. Each insight displays its category, mention count, and sources.

Filter insights by category, source, and time range. For example, filter to only social media sources to see how your products are discussed publicly, or filter to customer service tickets to understand product performance.

Insights align to these categories:

• Feature Request: Additional features your customers want.
• Complaint: Issues and problems your customers experience.
• Loved Feature: Features and functions that your customers enjoy.
• Brand: Specific brands mentioned by your customers.
• Bug: Functionality issues experienced by customers.
• Feature Mentioned: Any features or functionality specifically called out by name by customers.
• Pain Point: Places within your product that are problematic or causing issues for customers.
• Takeaway: High-level understandings and insight about your products, features, or functionality.

AI Feedback uses these categories to organize insights automatically.

You can specify any time range to view insights across the history of your data sources.

AI Feedback with AI Chat analysis

You can use Amplitude's Global Agent to analyze your Feedback data, perform semantic searches on feedback or insights, and answer requests such as:

• Summarize all recent feedback.
• Provide direct user quotes from customers about your website's search experience.
• Highlight the most talked about bug or complaint from the previous week.

You can also have Global Agent perform actions such as:

• Create a product requirements doc (PRD) to improve your company's product functionality based on feedback. For example, Global Agent can create a PRD to improve your notification system based on feedback about notifications.

Working with insights

Insights help you understand your customers' experience with your product. For every insight, AI Feedback offers actions you can take:

• Save as Trend: Saves the insight as a persistent trend you can track over time. See Saved Trends.
• Share: Lets you share the insight with an internal colleague.
• Create Survey: Lets you create a survey that you can send to customers. This action opens the Guide creation tool and adds the mentioned customers as a cohort for the survey.
• Watch Session Replays: Lets you watch session replays from customers directly related to the insight.
• Create Cohort: Creates a cohort based on the insight. Automatically adds the customers mentioned in the insight to the cohort.
• Delete mentions: Lets you delete individual mentions from a specific insight.
• Delete Insight: Lets you delete the insight from AI Feedback. You can't undo deleting the insight.

Perform an action on an insight

1. Open the insight you want and click Actions.
2. Select the action you want. AI Feedback performs actions as soon as you select them. Depending on the action, the system may take a few minutes to update.

Delete a mention

1. Open an insight with multiple mentions.
2. Next to the mention you want to delete, click the three-dot menu item and select Delete Mention. You can't undo this deletion.

Share an insight

• Click Share.

A link to the insight is automatically saved to your clipboard. Only colleagues with access to your Amplitude project can view it.

Saved Trends

Insights on the New Insights tab are ephemeral and refresh each time AI Feedback runs its analysis. Saved Trends let you persist insights and track them over time, so important themes don't get lost when new feedback arrives.

When you save an insight as a trend, or create one manually, AI Feedback links new, semantically relevant mentions to that trend daily. This lets you move from "what's new today?" to "how are my top priorities trending over time?" and track results as you fix bugs, implement feature requests, and reduce complaints.

The Saved Trends tab displays each trend's name, category, total mentions, week-over-week change, and a sparkline that visualizes recent activity.

Save an insight as a trend

1. Go to AI Feedback > New Insights.
2. Find the insight you want to persist.
3. Click Save as Trend on the insight row.

AI Feedback saves the insight as a trend and begins linking new, semantically relevant mentions to it daily.

Create a trend manually

You can create a trend from scratch to track a topic that doesn't match an existing insight.

1. Go to AI Feedback > Saved Trends.
2. Click Track a New Trend.
3. Enter a name for the trend in the Trend Name field.
4. Select a category from the Category dropdown.
5. Optionally, add a description to narrow the feedback the trend tracks.
6. Click Create Trend.

View trend details

1. Go to AI Feedback > Saved Trends.
2. Click the trend you want to examine.

The trend detail view displays:

• A chart of weekly mentions over time that shows how the trend's volume changes.
• The full list of linked mentions with source information, dates, and links to the original feedback.

Use the trend detail view to drill into specific mentions and understand the context behind each data point.

Merging existing users from an AI Feedback source

This lets you merge your existing users with users from supported AI Feedback sources such as Zendesk, Hubspot, or Intercom. Amplitude merges users with the same email address.

Merge existing users

1. Go to Organization Settings > Projects > your project > AI Feedback > User Mapping and specify the field that contains the user's email address.
2. From your system, send at least one event for each user that has that field populated.

After Amplitude receives that event for a user, AI Feedback can merge that user with the user from an AI Feedback-supported source using that email address.

Deleting information based on user ID or Amplitude ID

When you submit a deletion request to Amplitude's User Privacy API for a specific user ID or Amplitude ID, AI Feedback also deletes data associated with that user ID or Amplitude ID if you've merged your existing users with users from supported AI Feedback sources. If you haven't merged your existing users, the User Privacy API can't match Feedback data with the submitted user ID or Amplitude ID.

To ensure that the deletion is permanent and complete:

• Delete upstream first: You must first delete the data from the original source, such as Zendesk or Gong, before submitting your request to Amplitude. If you don't delete the source data first, the data syncs back into Amplitude.
• Manual deletion of other data: The User Privacy API only handles deletions tied to specific user IDs or Amplitude IDs. You must manually delete other related data, such as general or aggregated feedback, from the source management page.

Data access

AI Feedback follows your existing Amplitude project and role-based permissions. It doesn't grant new data access.

You authorize the source connectors you want (for example: Zendesk, Hubspot, Intercom, App Store/Google Play, Gong, G2/Trustpilot, Reddit/Discord/X) and they're scoped to your credentials using secure OAuth (or equivalent).

AI Feedback and LLM use

AI Feedback uses a third-party large language model (OpenAI) to turn connected feedback (for example tickets, reviews, call transcripts, social/forums) into product insights.

You control which sources to connect. AI Feedback processes only the feedback you choose to ingest and the Amplitude data you already can access. AI Feedback processes customer data for inference only. It's not used to train foundation models. AI Feedback handles requests transiently, with data encrypted in transit and at rest in your regional data plane. Outputs use raw feedback data with source links and proprietary safeguards, such as hallucination checks, to help ensure summaries reflect actual feedback.