Feedjoltdocs
Workflow

AI Insights & Sentiment

How Feedjolt classifies each post as positive, neutral, or negative and charts the sentiment trend. Covers when it runs, visibility, plans, and limits.

What it is

Feedjolt automatically classifies the sentiment of every new post as POSITIVE, NEUTRAL, or NEGATIVE. Each classified post gets a small badge in the dashboard post list and detail view - a one-glance signal of how a user feels, not just what they said. The new AI Insights page (sidebar -> AI insights) shows a stacked-area trend so you can spot whether your customers are getting happier or grumpier over time.

What sentiment means here

Sentiment is about the writer's emotional tone toward the product, not the topic. A bug report can be NEUTRAL (factual reporting) or NEGATIVE (frustration after the third re-report). A feature request can be NEUTRAL ("would be nice to have dark mode") or POSITIVE ("love this product! Just one suggestion: ...").

When sentiment is computed

Sentiment is classified once, when the post is created - by every channel (portal, dashboard form, email intake, Canny import, REST API). It is not re-computed when the body is edited or as comments accumulate. This keeps cost predictable and behavior easy to reason about.

A post may have no sentiment badge when:

  • The post is too short to classify confidently (fewer than 10 characters with no body).
  • The AI's confidence score is below 60%.
  • The post is spam, draft, or merged into another post.
  • The AI provider was unavailable or returned malformed output.

In all these cases the badge is simply hidden - no error is shown.

Visibility

Sentiment is admin-only. End users browsing your portal never see sentiment badges on their own or others' posts. This is a deliberate design choice - sentiment is a triage signal for your team, not a public attribute.

Required plan

AI Insights and the sentiment badge are advanced-AI features: they're on Startup and Scale (not Growth). Workspaces on Growth see an upgrade prompt on the AI Insights page; existing posts continue to display whatever sentiment was already computed (data is not deleted on plan change).

The trend chart

The AI Insights page renders a stacked-area chart over the last 7 / 30 / 90 days (switchable). Each day's stack shows the count of POSITIVE, NEUTRAL, and NEGATIVE posts created that day. A summary row above the chart shows percentages and a footnote line tells you how many posts in the window are uncomputed (for example, because they pre-date the feature being enabled for your workspace).

The chart counts posts, not votes. A 200-vote post counts the same as a 1-vote post. Vote-weighted measures belong to the priority-score feature (coming in a later release).

Backfilling existing posts

When sentiment is enabled for the first time on a workspace with existing posts, those posts will have no sentiment until backfilled. Backfill is currently an operator-run script - please reach out to [email protected] if you'd like sentiment computed for posts created before sentiment was enabled.

Limitations

  • Frozen at create. Sentiment doesn't shift as a thread heats up. A post that lands neutral but accumulates 50 angry comments still shows NEUTRAL.
  • Body truncated to 2000 characters. Very long posts are summarized by their first 2000 chars only.
  • Cost-aware skip. If a workspace exhausts its monthly AI quota, new posts receive no sentiment until the quota resets.
  • Multilingual. The classifier is given posts in any language; performance is generally good for the supported workspace locales (English, Spanish, Catalan), but quality may vary.
  • Small classification delay. Badges appear within seconds of post creation, not instantly - usually 5 to 15 seconds while the background job runs. Refresh the page if you don't see one yet.
  • Triage - how to act on negative-sentiment posts quickly.
  • Merging - when several negative posts point at the same issue.

On this page