openseo-keyword-research

OpenSEO Keyword Research

Turn seed topics into a prioritized, MEASURED keyword opportunity set and land it in brand/keyword-plan.md. mktg's keyword-research is the playbook (methodology); this skill is the measured-data engine behind it when OpenSEO is configured.

On Activation

1. Catalog check: mktg catalog info openseo --json --fields configured — if not configured, stop spending: state the gap and hand off to Exa-backed keyword-research with metrics marked unknown.
2. Project binding: read .seo/openseo.json for projectId. Missing → run openseo-project-setup first (or ask the user for the project id).
3. Brand grounding: read brand/positioning.md + existing brand/keyword-plan.md (tolerate templates). Business-fit beats volume-fit — the positioning file is the filter.

OpenSEO MCP Tools

• get_search_console_performance: when GSC is connected, START here. High rowLimit, filter average position ~5–20 client-side (the API sorts by clicks, not position). These striking-distance terms are the fastest wins — and zero extra credit cost.
• get_keyword_metrics: hydrate up to 700 keywords per call with volume, KD, intent, CPC, trends. Use on striking-distance terms and every candidate set.
• research_keywords: discovery from 1–5 seeds per call; prefer ~150 results unless exhaustive research was requested.
• get_ranked_keywords: when the brief includes a domain/page — exact ranking rows (near-misses, competitor-owned terms).
• get_serp_results: inspect SERPs for top candidates when intent is ambiguous. Keep batches small (≤10 queries).
• list_saved_keywords: avoid re-researching what's already saved.
• save_keywords: ONLY after explicit user confirmation, with concise tags (topic:<t>, intent:<i>, page:<slug>).

Workflow

1. Normalize seeds into 2–5 distinct research angles filtered by positioning.
2. GSC connected? Pull striking-distance terms first and hydrate with get_keyword_metrics. Work that list before broad discovery.
3. research_keywords per angle; get_keyword_metrics to hydrate; get_ranked_keywords if a domain is in the brief.
4. Remove irrelevant, duplicate, branded-only, and off-intent terms.
5. Prioritize by practical opportunity: business fit → clear intent → reasonable KD → volume/CPC signal → winnable SERP.
6. get_serp_results for high-potential or ambiguous terms when SERP intent would change the call.
7. Write the shortlist into brand/keyword-plan.md (preserve its required sections per brand/SCHEMA.md; confirm before overwriting populated sections).
8. Present: best opportunity theme, top keywords now, keywords to save, SERP caveats. Then next actions: openseo-keyword-clustering, seo-content, or save.

Cost Discipline

• State estimated call counts before bulk pulls (>200 keywords) and get confirmation.
• Small exploratory batches are fine without asking.
• GSC-first ordering exists precisely to avoid spending credits on data the user already owns.

Anti-Patterns

• Inventing metrics when OpenSEO returns nothing — because a hallucinated KD of "about 35" silently becomes the foundation of a content plan. If OpenSEO doesn't return a value, write unknown.
• Volume-first prioritization — because a 10k-volume term that doesn't match the product converts nobody and burns months. Positioning filters the list before metrics rank it.
• save_keywords without explicit confirmation — because saves mutate the user's OpenSEO account and bulk saves burn credit. Ask, state the count, then save.
• Skipping the GSC-first pass when GSC is connected — because striking-distance terms (positions 5–20) are provably the cheapest wins in SEO and they're free to read. Discovery research before first-party data is wasted spend.
• Overwriting a populated keyword-plan.md without confirmation — because that file is brand memory other skills build on (seo-content, seo-machine). Merge; confirm destructive rewrites.

Progressive Enhancement

Adapted from every-app/open-seo .agents/skills/keyword-research (MIT). Workflow and tool guidance upstream; mktg brand-memory writes, positioning filter, and cost discipline added here.