When exact words work
A brand name, model number or unusual name is often easy to find literally. Yet ‘Who can fix my AC?’ does not contain ‘air-conditioner repair’ even though it asks for that service. One keyword list cannot capture every expression.
Describe the intended match
In Zuvi, write a plain-language rule: ‘Find people personally seeking an AC repair specialist in Ho Chi Minh City or asking for a contact; exclude service providers advertising, someone else's problem and resolved cases.’ Select relevant source groups. Clear boundaries matter more than repeating synonyms.
Test both sides
A likely match is ‘Who can fix my split AC today?’ A non-match is ‘I repair split units; message me.’ If both trigger, clarify the author's role and the ad exclusion. Also test a short reply with available context without attributing another person's need to the replier.
Correct mistakes carefully
Label a false result Irrelevant and submit a missed request from the filter card. Test individual examples, review an edit as a draft and confirm before it becomes active. Restore a previous version if the result worsens.
What to expect
Semantic search cannot guarantee perfect matching or replace access to the source group. It helps when keywords alone are too brittle, but important findings still need review in the original chat.
Try it with your groups
Open Zuvi, choose an accessible source and test the rule on a real example before monitoring starts.
Open bot