Phrase Negative Discovery
N-gram decomposition of the search-term corpus surfaces phrase-negative candidates that exact-match negation misses.
or just say “run phrase negative discovery for <brand>”
Comes with the plugin, one install activates every skill. See the install flow. Needs brand setup first.
A single phrase negative can silently block dozens of legitimate search term variants at once, so finding phrase-level waste (the without-laces or fake-leather clusters of zero-conversion spend) by hand risks either missing real bleed or over-blocking traffic that actually converts.
A ranked list of phrase negative candidates, semantically clustered and cleared through a conflict check against every converting search term, each labeled with its blast radius and which campaigns it is safe to apply to, so nothing goes live until an account manager has validated it.
What your agent can do
- 01
Decompose search-term reports into n-gram sequences and aggregate performance per sequence
- 02
Surface sequences with combined spend and zero conversion history
- 03
Apply conflict detection, semantic clustering, and brand-context filtering
- 04
Phrase negatives have blast radius: every candidate is flagged for validation before application
Before you run
This skill reads a brand’s context to calibrate its output. Build it once per brand, then every teammate runs on the same shared context.
/mixshift-ai:mx-brand-context <brand>How brand setup works →- Phrase spend threshold: the minimum combined spend an n-gram must clear to surface as a candidate
- Protected terms and lane rules: the anchors and dictionaries that filter candidates before they surface
- Known converting adjacencies and brand term dictionary: brand-context checks that suppress a candidate already confirmed as fine
As protected terms, lane rules, and known converting adjacencies accumulate from prior account manager decisions, fewer legitimate phrase clusters get suggested for negation and the conflict check has more history to catch a collision against, so the candidate list gets tighter over time.