The Inbox AI search finds too few results
Short answer
Usually the query is too narrow, another filter is blocking hits, or the updates don’t have an embedding yet. Shorten the search to one or two key terms, reset other filters, and give new sources a moment for the embedding to be calculated.
Details
Query is too specific
The AI search finds thematically related updates — but it gets confused when the query is phrased very narrowly. A whole sentence like “Which companies announced price increases in Q2 and how did they land?” produces a very tightly focused semantic vector.
Better: cut it down to one or two core terms.
- Instead of:
Which companies announced price increases in Q2 - Use:
price increase Q2
Other filters block the hits
The AI search runs first; the classic filters (source, channel, tag, time range) narrow the result further. An active “last 7 days” range or a specific source can cause semantically matching updates to drop out.
Before running the AI search, hit Reset on the other filters and see how the list changes.
Updates without embedding
The AI search only reaches updates that already have an embedding. For new updates, the embedding is calculated a few seconds to minutes after collection — for very large collections (backfill after adding a new source) it can take longer.
If a freshly added source returns no hits yet: wait 10–15 minutes and try again.
Word stem not matched
For rare technical terms or English terms inside mostly German content, the semantics can be less stable. In those cases the classic text search complements the AI search well — text search at least catches the literal matches.
Too few topics in the workspace
In a very young workspace with few sources and updates, the semantic hit list is generally thin — not because of a bug, but because the data base is small. After a few weeks with active source collection this normalises.
If this doesn’t help
If the AI search stays completely empty — even for simple, general terms like “customer” or “product” — there may be a short-term hiccup in the embedding backend. Try again in 10–15 minutes. If the problem persists, contact support (chat widget bottom right, or support@picasi.com).