Keyword Research and Content Prioritisation . Agency OS

What it does

Produces a prioritised content plan, not a keyword list. Every keyword is scored on Business Potential (whether your offer is actually the answer to that search), tagged by intent and funnel stage, mapped to a content pillar and a specific CTA, then ranked by a priority formula that weighs traffic potential and commercial value against difficulty. Keywords are clustered by SERP overlap into pillar and cluster pages with an internal linking plan, then sequenced into a 12-week calendar that puts revenue-adjacent content first. Output is saved as keyword-plan.md, which this skill owns.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

> keyword research

> what should I write about

> quick keyword check

> content gaps

> competitor keywords

> manual keyword research

When to reach for it

you do not know what to write about next
planning a quarter of SEO content
checking whether a topic is worth targeting at all
finding what competitors rank for that you do not
you have keywords but no order to work through them
you have no SEO tools and need a manual method

When NOT to use it

If you actually want	Use this instead
writing the article once a keyword is picked	`blog-writer`
turning a keyword pattern plus a dataset into hundreds of templated pages	`programmatic-seo`
the 30-day multi-platform posting calendar and content sprint	`content-system-architect`
choosing paid ad targeting or writing ad copy from these terms	`meta-ads-copywriter`

Before you start

What you need	Why
Your website URL, what you sell, and who you serve	it asks for these before proceeding if no workspace context exists, because Business Potential cannot be scored without knowing the offer	Required
2 to 3 competitors you know of	competitor mining is one of the three seed sources and the gap analysis in phase 3 runs off these domains	Required
Your target market: country and language	it defaults to United States and English and will confirm before running queries. The wrong market returns the wrong volumes entirely	Required
A DataForSEO MCP connection	this is the live data source for search volume, keyword difficulty, intent classification, competitor ranked keywords, domain intersection and SERP results. It is explicitly optional: every phase has a documented manual alternative. Without it you work from Google Autocomplete ordering, People Also Ask, competitor sitemaps and manual SERP inspection, and every volume and difficulty figure is tagged [ESTIMATED] rather than measured	Optional
Google Search Console access	the only free source of real data for terms you already rank on. Impressions act as a volume proxy and it is what makes near-miss opportunities visible	Optional
Live web search	the manual workflow depends on actually searching SERPs, mining People Also Ask, and reading competitor sitemaps. With neither DataForSEO nor web search there is no data layer left and the scoring becomes guesswork	Optional
`audience.md` or `ica-output.md` in the target	it uses the ICA to inform keyword language and search behaviour so you are not asked to re-explain your customer. Voice and positioning files are deliberately not loaded	Optional
Your existing site content list	phase 4 checks for cannibalisation against pages you already have; without it that check cannot run	Optional

How it runs

Load context and pick a modeReads brand files under freshness rules, past learnings tagged to this skill, and keyword-research-memory.md if it exists. Then it selects a mode from how you asked: A full research, B quick opportunity scan, C competitor gap analysis, D content calendar build, or E fully manual with no tools.
Map the competitive landscape and pillarsBefore any keyword work it separates direct, indirect and content competitors and defines 4 to 6 content pillars for your business. Every keyword afterwards has to map to at least one pillar, and keywords spanning several pillars get priority.
Phase 1: seed discoveryThree sources: your product mapped to the problems it solves and the words customers actually use, competitor mining, and topic expansion through modifier patterns, autocomplete and People Also Ask. Target output is 50 to 200 seed keywords.
Phase 2: metrics and scoringPulls or estimates volume, difficulty and intent, then assigns each keyword a Business Potential Score from 0 to 3 and a priority score weighing traffic potential, business potential and commercial value against difficulty. It filters out zero-business-potential terms, keywords far above your site's authority, zero-click queries, and anything outside your customer base.
Phase 3: competitor gaps and SERP analysisFinds what competitors rank for that you do not, near-misses where you sit at positions 8 to 20, and where competitor content is thin or generic. Each priority keyword is analysed for content type, format and angle, and for which SERP features are present (featured snippet, People Also Ask, AI Overview, video carousel, local pack), because each one changes what you should publish.
Phase 4: topic clusteringKeywords that trigger three or more of the same top-5 URLs go into one cluster and become one page. Each cluster gets a primary and secondary keywords, and the pillar and cluster structure comes with an internal linking plan. It also checks whether your existing pages already compete for these terms.
Phase 5: prioritised calendarSequences bottom-up, revenue first: decision-stage content in weeks 1 to 4, consideration in weeks 5 to 8, awareness in weeks 9 to 12. It separately surfaces the top 5 quick wins that are realistically rankable in 4 to 8 weeks.
Phase 6: content briefs and memoryWrites a brief per planned piece covering target keywords, SERP findings, word count, required H2s, People Also Ask questions, internal links, the competitive gap you are exploiting, and the CTA. Then it offers to save or update keyword-plan.md, showing a diff before overwriting, and logs feedback.

What you get

brand/keyword-plan.md following the keyword-plan-v1 schema, with keywords, topic clusters, content gaps and competitor keywords
An executive summary of the top 5 findings and biggest opportunities
A scored keyword universe tagged with intent, funnel stage, Business Potential Score, priority, content pillar and CTA target
A competitor gap list showing what they rank for that you do not
A topical map of pillar and cluster pages with an internal linking plan
A 12-week content calendar sequenced by funnel stage
The top 5 quick wins as immediate actions
Content briefs, with the first 5 written out in Mode D
keyword-research-memory.md, created on the first run and updated after every run so scoring improves over time

Honest limits

Read this before you rely on it

Without a DataForSEO connection every volume, difficulty and traffic figure is an estimate, and the skill labels them [ESTIMATED] rather than pretending otherwise. The clustering, prioritisation and calendar logic still work. The numbers underneath them are directional.
Business Potential Score is a judgement call, not a measurement. It is also the metric that drives the whole priority ranking, so read the 0 to 3 scores and correct any that misjudge how central your offer really is to that search.
It plans content, it does not write it. Every calendar entry still needs blog-writer or another skill to turn into a page.
Rankability depends on comparing your site's authority to the top 10 results. Without tools that comparison is eyeballed, so difficulty calls near your authority ceiling are the least reliable part of the output.
The skill owns keyword-plan.md. If you have edited that file by hand, check the diff it shows you before accepting an overwrite.

Where people go wrong

The mistake	Do this instead
Sorting the list by search volume and working down it	Sort by priority score. A 200-a-month keyword where your offer is the only real answer beats a 20,000-a-month keyword you can only mention in passing.
Asking for a keyword list and stopping there	The plan is the deliverable. An unprioritised list of keywords is the thing this skill exists to replace.
Targeting keywords far above your site's authority because the volume looks good	Check the difficulty against your own domain rating and take the quick wins first. Save the hard terms for a plan with a timeline attached.
Ignoring which SERP features are present	A featured snippet, an AI Overview and a video carousel each demand a different format. Check whether click-through potential still exists before committing to the piece.
Skipping the cannibalisation check	Ask it to look at your existing pages. Two of your own pages competing for one term is a self-inflicted ranking problem that is cheaper to fix than to out-write.

Worth knowing

Business Potential Score is the metric that actually matters. Score every keyword 0 to 3 on whether your offer is the irreplaceable answer, drop the zeros, and require every keyword scoring 2 or 3 to name which offer it feeds and by what path. That single discipline is what turns a keyword list into a plan.