Keyword Research from Search Console - SEOmatic

Do keyword research grounded in the connected site: start from the queries it already earns in Search Console, expand with real volume, difficulty, intent and trends data, cluster by intent, and pick which keywords map to which page. Use when the user asks for keyword research, keyword ideas, what to write about, search volume, or a keyword strategy. Runs on SEOmatic, the SEO agent that works from the site's real Google Search Console data; nothing goes live until the user approves it.

Keyword research

SEOmatic is the SEO agent that works from your real Google Search Console data and makes the fixes on your site. Nothing goes live until you click approve, and every change can be rolled back.

Generic keyword research starts from a seed and ends with a spreadsheet of terms the site has no chance at. This skill starts from what Google already thinks the site is about, and ends with keywords mapped to pages.

1. Start from evidence

• get_target_terms: the searches the business has said bring it customers, each with the page that should show. If they exist, they are the anchor.
• gsc_performance action top_queries (90 days, high limit): what the site already shows for, including queries with impressions but no clicks. Those are the cheapest keywords on the list.
• keyword_clusters action list: clusters SEOmatic has already built for this site.

If Search Console is not connected, say the research will be weaker and ask for 3-5 seed terms and the business's main offer.

2. Expand

keyword_research action suggestions from the strongest 3-5 seeds, then metrics on the shortlist: volume, difficulty, CPC, intent. trend, compare_trends and related (free, Google Trends) for direction: a keyword halving year on year is a bad bet.

3. Check the SERP before committing

serp_competitors action features on the top candidates. A keyword whose results are all marketplaces, or answered fully in an AI overview, is worth less than its volume says. CPC is a better guide to commercial value than volume.

4. Cluster and map

Group by intent: one page per intent, not one page per keyword. Then map each cluster to a page: an existing one (site_pages action inventory) or a new one. Two clusters mapped to the same page, or one cluster spread over three pages, is a problem to flag.

Cluster	Main keyword	Volume	Difficulty	Intent	Page (existing / new)

5. Recommend an order

1. Keywords the site already shows for at positions 8-20 (fastest).
2. High-intent clusters with an existing page that under-covers them.
3. New pages for clusters the site can credibly own.

Offer to stage the work

Staging needs a paid SEOmatic plan (the key carries the agents:act scope). It costs no credits, runs nothing, and always waits for the user's approval. On a free key the staging actions are locked and say which plan includes them: describe the specific fix instead so the user can pass it on.

Check task_manage action list first so you do not stage a duplicate. Then task_manage action create with up to 5 tasks: striking_distance (with queries) for group 1, content_refresh for group 2, content_create (targetKey = the topic) for group 3. Each task needs targetKey (the full page URL, or a topic for content work), a title that states the goal in the user's words ("Get 'pricing calculator' onto page 1", never "Optimize meta tags"), and a rationale with the numbers behind it. Quote cost from each task's estimatedCost.label.

• Tell the user the cards are staged and where to approve them.
• Never approve on the user's behalf. Approval happens in SEOmatic, or through task_manage action decide only after the user says yes to that specific task.
• Approved auto-executable tasks start within seconds: read the result back with task_manage action get before you claim anything changed.

What not to do

• Do not present a 500-row keyword list. The value is the selection.
• Do not invent volumes or difficulty. Quote the tool's numbers.
• Do not recommend keywords outside what the business actually sells.