Most keyword research is a one-time project that goes stale within weeks. Muginai treats keyword discovery as a continuous process that runs on a defined schedule and updates your universe automatically.
What Gets Discovered
The keyword research pipeline sources terms from multiple channels:
- Google Search Console — queries where you already have impressions but low clicks, surfaced as quick-win opportunities
- Autocomplete and PAA — suggestions from search engines for every seed term in your topic space
- Competitor gap analysis — terms your competitors rank for in the top 10 that you do not
- Semantic expansion — related terms derived from embedding similarity to your existing keyword set
Each term gets a volume estimate, a competition signal, and a freshness timestamp so you can see when it was last checked.
Semantic Clustering
Keywords are grouped using vector embeddings, not just shared words. A cluster like “invoice software for freelancers” captures variations (“freelancer billing app”, “self-employed invoicing”, “invoice generator freelance”) that exact-match tools would split into separate clusters.
Each cluster maps to exactly one content opportunity. The cluster view shows:
- Number of keywords in the cluster
- Estimated aggregate search volume
- Whether content exists for this cluster
- The strongest intent signal (transactional, commercial, informational, navigational)
Intent Classification
Every keyword is classified by search intent using a combination of SERP feature analysis and a heuristic model trained on search behavior signals. Intent classification determines:
- Which cluster a keyword belongs to
- What content format the brief should specify
- Whether the opportunity is commercial or informational
Money Keyword Identification
Keywords with strong transactional or commercial intent are flagged as money keywords. These get prioritized in the discovery queue and tracked with daily rank checks rather than the standard weekly cadence.
Freshness and Decay
The system re-crawls keyword data on a rolling schedule. Terms that have not been re-evaluated in 30 days are flagged for refresh. Seasonal volume spikes are detected automatically using historical trend data from Search Console.