What it is.
You give it three seed keywords and a domain. Ahrefs MCP calls pull matching terms, related terms, and existing rankings, a Claude node clusters them and classifies intent, and Google Sheets MCP writes the file.
The output is four tabs: the full clustered set, quick wins, cluster themes, and negative keyword suggestions. The same work by hand is documented at 90 minutes, and this runs in under 10.
TripleDart's own guide, and it names our platform as the build surface. The node list transfers to any tool with an Ahrefs connection.
What you get.
- A seven-node build documented node by node, with the input and output of each.
- A four-tab spreadsheet: clustered set, quick wins, cluster themes, negative suggestions.
- The full Direction prompt structure, published in full.
- Live Ahrefs data on every run, so the clustering works from current volumes.
- A negative keyword tab, which most keyword research output leaves out entirely.
How to set it up.
Connect Ahrefs MCP and Google Sheets MCP before you build anything, since five of the seven nodes depend on them.
Write the Direction prompt first, using the published structure as the starting shape.
Wire node two and three to Ahrefs matching terms and related terms for your seed keywords.
Add node four for existing rankings, because it is what separates a quick win from a cold start.
Put the clustering and intent classification in the single Claude analysis node.
Write to Google Sheets with one tab per output, then run node seven to validate against a set you already know.
Use cases
Build it node by node
Follow the seven nodes with the input and output stated for each.
Get a clustered set out
Read the tabs and take the clusters straight into briefs.
Keep the analyst judgment
Let the model do the mechanical grouping and make the calls yourself.
Best for
Agencies with repeat keyword work
The same research shape across many clients, where only the seeds and the domain change.
Briefing at volume
Clustered output chains straight into brief generation, so keyword research stops being the bottleneck.
Auditing a keyword list you inherited
Running the Skill against the same seeds shows what the previous list missed.