Been doing SEO content planning for a while and the part that always ate the most time wasn't finding keywords, it was figuring out what to actually do with a few hundred of them once you had the list. So I built a workflow that takes a raw keyword export and turns it into an organized content plan without me touching a spreadsheet manually. It runs two parallel strategies off the same keyword set, and pushes both into Airtable so I can compare approaches side by side. Path one: categorization. Every keyword gets run through an AI agent that scores it against a fixed rule set, Quick Wins (low difficulty, decent search volume), Authority Builders (high difficulty, high volume, worth the long game), Emerging Topics, Intent Signals like People Also Ask questions, Semantic Topics, or Unknown. Anything that lands in Unknown gets filtered out before it moves further down the pipeline, so I'm not generating content briefs for keywords that don't actually fit anywhere useful. Everything that survives the filter gets a second agent pass that writes an SEO title and article description for it, then it's written into Airtable with a status field so it can move into an article writer workflow later. Path two: clustering. Instead of scoring keywords individually, this one feeds the entire keyword set to an agent in one shot and asks it to find natural semantic clusters, letting the number and size of clusters emerge from the actual relationships in the data rather than forcing a fixed count. Each cluster gets a hub article (broad overview) and up to 5 spoke articles (specific subtopics branching off it), which is a content structure I've found actually builds topical authority instead of just publishing disconnected posts. The trickiest part was getting the JSON output stable enough to parse reliably, every agent node is explicitly instructed to return single-line JSON with no line breaks or extra text, since even small formatting drift breaks the downstream Set/Split nodes that depend on parsing that structure cleanly.

i just downloaded the skills and dropped them into Claude (using VS Studio drag & drop) and it tells me: "Done I read your Brand Guidelines and personalized the stack. Since I can't rename files without risking breaking cross-references, and the download itself has filenames that don't match their content (see note below), I worked by actual content and tracked everything. ⚠️ Important discovery: many files in this download are mislabeled — the file named Content Repurposing Skill.md actually contains a duplicate of the CLAUDE.md constitution; press-release-structure.md contains the Personal Brand Social Media skill; LinkedIn Post Writer.md contains the Email Newsletter skill; newswire-submission-fields.md contains the real LinkedIn Post Writer skill (hardcoded to "Spark's voice" — fixed); etc. I fixed content wherever it actually lives, not by filename. You may want to rename these files later for sanity — I didn't rename anything since moving files risks breaking references I can't fully trace." I don't know how to rename anything and I'm not sure if this was my "operator error" but thought I'd post it here in case there is an easy fix.

My latest video interviews @Aaron Makelky where we break down: - Why ChatGPT Pro is the best $100 AI subscription right now - The Claude Max catch nobody mentions: half your usage capped on Fable - Fable 5.1 vs GPT-6 Astra tested on real landing pages and websites, not demos - Why price per task beats price per token every time - What Anthropic optimized for since January - Why I'm now rooting for open-source Chinese models Be sure to drop a Like & Comment on YouTube 🤝

https://www.youtube.com/live/lvlTpE0VDfk?si=IwXMfwiUCukz6Vgm