Automatic AI tagging - how it works
The Basic Principle
When a new update arrives in your workspace—a LinkedIn post, a newsletter article—Picasi immediately checks which update tags might apply to it. For each tag that has an AI tagging description, the AI calculates a confidence score: How likely is it that this update belongs to this tag?
If the confidence score exceeds the configured threshold, the tag is automatically assigned. This happens without any manual intervention and is completed before the update appears in your inbox.
What Determines Accuracy
The description is everything. The AI understands natural language well, but it cannot guess intentions. If the description for the tag Produktneuheiten is simply “Updates about products,” the AI will interpret it broadly—and tag things that weren’t actually intended.
A precise description provides examples in both directions: what belongs and what explicitly does not belong. This significantly reduces misclassifications.
The confidence threshold is the second lever. With a low threshold (0.5), Picasi tags more often, but also makes more mistakes. With a high threshold (0.9), it tags less often, but with greater accuracy. For most teams, 0.65 to 0.75 is a good starting point.
What automatic tagging cannot replace
Automatic tagging is good at reliably identifying clear categories—but it has limitations when it comes to context-dependent assessments. A post about a “new partnership” can fall under Partnerships, Products, or Customers depending on the content. The AI makes a decision, but it can be wrong.
For critical decisions—for example, in AI reports sent to the board—it’s worth briefly reviewing the filtered update list before generating the report.
Retroactive Application and Fresh Start
Automatic tagging only applies to new updates received after the tag is configured. Existing updates are not tagged retroactively. If you create a new tag and want to apply it to older updates, you must do so manually.
If you change a tag description, existing tags do not correct past errors—only new updates benefit from the improved description.