Why is my brand name a bad keyword?

Because most short brand names are also ordinary words somewhere. Our blog post on it goes through the fixes; this playbook applies them.

What is a Means line?

One sentence on a keyword saying what it refers to — “Beacon, the uptime monitoring tool”. The relevance read uses it to tell your Beacon from the others.

Does filtered noise use my allowance?

No. Only relevant mentions are metered.

Playbook · Monthly

Audit which of your keywords are mostly noise

A brand name that is also a common word fills an inbox with the wrong posts. FishOut filters them, but a keyword that is 80% filtered is still a keyword to fix. This playbook measures each one and proposes exclude words, a Means line or a mute — and changes nothing until you say yes.

Start free Connect your assistant

[01] THE PROMPT/ COPY IT

Paste this

Fill in the {{braces}}. Run below: Claude.

prompt

Using FishOut, list our keywords. For each, count mentions from the last {{30}} days with and without includeFiltered, and work out the share that was filtered. For the {{3}} noisiest, look at the filtered mentions and propose a fix: exclude words, a Means line, or muting a source. Show me the proposals as a table and ask before you change anything.

[02] WHAT IT CALLS/ 5 TOOLS

Tools and arguments

3 of 5 read only.

ToolWhat it does hereArguments

list_keywordsread-only
Every keyword and its current exclude words.
—
count_mentionsread-only
Twice per keyword: once relevant only, once with filtered included.
filter: {"keywordIds":["<keyword>"],"includeFiltered":true}
list_mentionsread-only
The filtered posts, to see what the collision actually is.
filter: {"view":"filtered","keywordIds":["<keyword>"]}, limit: 20
update_keywordasks first
Adds exclude words or a Means line — only after you approve.
negativeTerms: ["<word>"], means: "<what you mean>"
create_mutewrite
Mutes a source that only ever matches by accident.
kind: "source", value: "<source>"

[03] A RUN/ FIXTURE

What comes back

Fixture data.

Claude · chatfishout · mcp

Audit our keywords for noise.

list_keywordsread-onlyarguments

{}

count_mentionsread-onlyarguments

{
  "filter": {
    "keywordIds": [
      "kw_…"
    ],
    "includeFiltered": true
  }
}

list_mentionsread-onlyarguments

{
  "filter": {
    "view": "filtered",
    "keywordIds": [
      "kw_…"
    ]
  },
  "limit": 20
}

Harbour Timesnews5d

Beacon ferry timetable changes for winter

Filtered · Different meaning

TikTok@coastwalks6d

Night walk to the old beacon

Filtered · Different meaning

“Beacon” is 83% filtered: ferries, lighthouses, a football club. Proposed: Means = “Beacon, the uptime monitoring tool”, exclude “ferry”, “lighthouse”, “FC”. “Beacon status page” is 4% filtered — leave it. Apply the first change?

[04] GUARDRAILS/ LIMITS

Propose, then ask

update_keyword carries a destructive hint: changing a keyword changes what you collect from then on. The prompt asks before it.

Filtered posts are free

Noise never counts toward your plan, so this is about a cleaner inbox and the noise guard, not the bill.

Do not over-exclude

An exclude word removes every post that contains it. Prefer a Means line, which the relevance read uses, to a long exclude list.

[05] SCHEDULE IT/ CRON

A prompt runs once

Monthly: Codex, OpenClaw or Hermes Agent.

Hermes Agent · cron expression

hermes cron create "0 9 1 * *" \
  "Using FishOut, list our keywords"

Or let FishOut do the scheduled part

FishOut already filters same-word noise with a reason and slows a keyword whose posts are mostly noise. The audit tells you which keyword to fix and how.

[06] NO ASSISTANT/ RULES