Back to blog
FILE 0x37·BRAND MONITOR NOW FILTERS MENTIONS BY CONTEXT TO CUT FALSE P

Brand Monitor now filters mentions by context to cut false positives

June 13, 2026 · brand-monitor, monitoring, false-positives, filtering

The most common complaint about keyword monitoring: false positives. You track "Chester" and get mentions of Chester Zoo, Chester Bennington, Chester, Pennsylvania. You track "Frogger" and get gaming nostalgia posts.

Brand Monitor now lets you specify ignore terms per keyword so mentions from the wrong context get dropped before they reach your inbox.

How it works

When you add a keyword, there's now an optional context field (≤200 characters). The format is:

ignore: zoo, bennington, pennsylvania, cheshire

At scan time, Brand Monitor checks every mention's title and body snippet against the ignore terms. If any term appears (case-insensitive), the mention is dropped. It never appears in your digest or spike alert.

The filtering happens at source — in the HN, Reddit, GitHub, and News scanners — so dropped mentions don't inflate your unread count or affect spike detection.

The design decision: ignore-only, not require

I considered two approaches:

  1. Require terms — only surface mentions that contain specific context words
  2. Ignore terms — suppress mentions that contain specific words you don't want

Require-terms fails for small brands: if your brand is only mentioned in a few dozen posts per month, requiring context terms would drop too many real mentions. You'd miss signal.

Ignore-terms has better recall at the cost of some false positives getting through when none of your ignore terms match. For a brand with decent volume, it's the right tradeoff — you're subtracting noise, not filtering for a needle.

Example use cases

Common name: You're tracking "Anchor" (your design tool) but getting results for Anchor FM (podcast platform) and anchor stores. ignore: podcast, fm, retail, store drops most of the off-topic results.

Shared abbreviation: "OSS" (your product) returns results about open-source software. ignore: open source, apache, gnu, linux cuts the noise.

Geographic overlap: "Pacific" (your company) gets mentions about Pacific Ocean, Pacific Time, Pacific Northwest. ignore: ocean, time zone, northwest, coast helps significantly.

The ignore list won't be perfect. Some false positives will still get through when none of your ignore terms match the context. But a well-tuned list typically reduces noise by 70–80% for common-name keywords.

API

Existing users: PATCH /keywords/{keyword_id} with {"context": "ignore: term1, term2"}.

New keyword via POST /keywords: include context in the request body alongside keyword.

GET /keywords now returns the context field when set, so you can verify what's configured.