AI Clustering

AI clustering removes the manual work of reading and categorizing every piece of feedback. The system automatically groups similar submissions, surfaces the most common themes, and lets you convert clusters into roadmap items with one click.

How it works

Feedback submitted
        ↓
AI generates an embedding (semantic fingerprint)
        ↓
Embedding compared against existing clusters
        ↓
Assigned to best-matching cluster
(or a new cluster is created if no match)
        ↓
Cluster summary regenerated periodically
        ↓
You review clusters → convert to feature requests

Embeddings and semantic matching

Each piece of feedback is converted into a high-dimensional vector (embedding) that captures its meaning, not just its keywords. Two feedback items that say the same thing differently — "I can't export my data" and "Please add a CSV download" — will be placed in the same cluster because their embeddings are semantically close.

This means clusters are based on intent and topic, not exact wording.

Cluster summaries

Each cluster has an AI-generated summary that describes the common theme across all items in it. Summaries are refreshed automatically as new feedback is added to the cluster.

Confidence threshold

Feedback that doesn't match any existing cluster above a confidence threshold creates a new cluster rather than being forced into a poor match. Low-confidence singletons may appear as their own cluster until more similar feedback arrives to confirm the theme.

Viewing clusters

Go to Feedback → Clusters to see all clusters sorted by:

• Volume — how many feedback items are in the cluster
• Recency — when the most recent feedback was added
• Date created — when the cluster first appeared

Each cluster card shows:

• The AI-generated summary
• Number of feedback items
• A sample of the original submissions

Click a cluster to see all the individual feedback items inside it, along with user info and metadata if available.

Converting a cluster to a feature request

When a cluster represents a real need you want to address, convert it to a feature request (roadmap item):

The cluster remains linked to the feature request. Users who submitted feedback in that cluster will be associated with the roadmap item, and you can notify them when the feature ships.

Reclustering

The clustering model runs continuously as new feedback arrives. You can also trigger a manual recluster from Feedback → Clusters → Recluster All if you want to force all feedback to be re-evaluated — for example, after a large batch import.

Tuning clustering behavior

From Feedback → Settings → Clustering:

FAQ