Keyword search is not social listening

Every listening product starts as keyword search. Type the brand, get the posts. That is necessary and nowhere near sufficient.

A brand name that is also an ordinary English word will drown you. A competitor launch will look identical to a compliment. A joke and a chargeback threat can share a hashtag. Without a second pass that asks what is this and how soon does it matter, you have built a firehose with a logo.

Two jobs, one feed

Discovery answers: did this string appear?

Listening answers: is this about us, what kind of mention is it, and should someone act?

We run both. Providers return candidates; a classifier drops coincidental matches and scores category, sentiment, urgency, intent, speaker and a suggested next action. The feed you see is the second job. The first job alone is what free tools already do well — see F5Bot if that is truly all you need.

Why agents care about the distinction

An MCP tool that returns every keyword hit will burn context and trust. An MCP tool that returns urgent, on-brand mentions with a next action is something an agent can usefully call. The difference is not marketing copy; it is whether classification happened before the response left the server.

A practical check

Take last week's raw matches for your brand. Count how many a careful human would ignore. If that share is above half, you do not have a volume problem — you have a listening problem dressed as search.