Find the searches your site can win next, with your agent
Finds striking-distance keywords (positions 4 to 20), CTR gaps, cannibalized queries and competitor topic gaps using Google Search Console data and clickd, then maps each query to one page and one action.
Add it, from the top of the site’s repositorynpx skills add subinternal/clickd --skill clickd-keyword-research
Then ask your agentPull our Search Console queries and rival topic gaps with clickd, find our best striking-distance and CTR-gap pages, and update the top three.
For a founder, marketer or developer deciding which pages to sharpen or build next
Four steps from real impressions to page actions
Instead of guessing search volume from third-party estimates, your agent works from your Search Console impressions and competitor topic gaps.
1Pull real queries and topic coverage from clickdclickd_queries
2Bucket queries by action: striking distance, CTR gap and missing page
3Check search intent against the live results
4Turn the top targets into queued work or immediate edits
A keyword action map from live data
Every query is placed in one action bucket and assigned to a single canonical route on example.com.
Striking distance (pos 4 to 20)“automated accounts payable workflow” (pos 7.4, 1,840 impr.)Owner: /ap-automation. Add ERP sync section and 4 internal links from /integrations/*
CTR gap (pos 1 to 5, low CTR)“three way invoice matching” (pos 2.9, 3,120 impr., 1.4% CTR)Owner: /guides/three-way-matching. Rewrite title and description to name the 3 checks
Cannibalization fix“purchase order approval rules” split across 2 blog postsKeep /guides/po-approval-rules; merge and 301 redirect older post
Competitor topic gapRivals rank for “ach vendor payment fees” (0 pages on site)Queue new calculator and guide at /tools/ach-payment-calculator
The clickd tools behind the research
Connected at https://clickd.com/mcp/all so Search Console and rival gap tools are available in one loop.
clickd_queries & clickd_trafficPulls real Search Console queries, impressions, clicks, CTR and average position for the site and each page.
clickd_gaps & clickd_rivalsFinds customer topics and page types that close rivals cover and your site has not built yet.
clickd_topics & clickd_content_brainGroups related searches into one page per intent so you never build thin duplicate pages.
When to reach for it
Reach for it when
You want to find pages sitting in positions 4 to 20 that can reach the top 3 with targeted improvements.1,2
You have high-impression pages with low click-through rates and need sharper titles and descriptions.
You want to know which new pages to build next based on competitor topic gaps rather than guesswork.
Leave it when
You want a spreadsheet of thousands of unverified keyword volume estimates with no page mapping.
You want to spin up doorway pages for every keyword variation instead of one strong page per intent.3
More things to ask your agent
Everyday prompts that load the Keyword research skill in Claude Code, Codex or Cursor.
Striking-distance winsPull clickd_queries for example.com, list every query in positions 4 to 20 with high impressions, and sharpen the top 3 pages.
Fix low CTR on page oneFind pages ranking in positions 1 to 5 with low CTR in Search Console and rewrite their titles and meta descriptions.
Competitor topic gapsCombine clickd_queries and clickd_gaps to show the top 5 buyer searches our rivals cover that we have no page for yet.
Keyword research skill vs the alternatives
Why pairing an open repository skill with a live crawler beats manual audits and prompt-only checklists.
How it reads your siteBy hand: Manual spot-checks in a browser · Prompt-only skill: Reads local source files only; misses live headers, redirects and JS shellsReads the live URL, HTTP headers, raw HTML and Search Console via clickd MCP
How changes are madeBy hand: Copy-pasting between docs, spreadsheets and editors · Prompt-only skill: Suggests generic copy in chat with no repo contextEdits your actual templates, routes and schema in the repository
How fixes are provedBy hand: Checking pages one by one after deploy · Prompt-only skill: Agent grades its own homework with no live checkRuns clickd_rescan and clickd_changes on the deployed site to prove rules passed
Skills to use alongside it
Included when you run npx skills add subinternal/clickd.
Content briefTurns competitor topic gaps and customer questions from clickd’s content brain into fact-backed briefs and finished pages in your repo, complete with question headings, internal links and JSON-LD schema.See skill
On-page SEORewrites title tags, meta descriptions, single H1s, question headings, answer-first openings, image alt text and AI writing tells in your site’s templates and content files, checked live with clickd.See skill
Rank a pageRuns a campaign to move one page up for one search toward an aim, through five gates in order, each opened only by a proof from a read of the live site, with a weekly read-out and the campaign’s time to rank.See skill
Questions about the Keyword research skill
Why use Search Console impressions instead of third-party keyword volume tools?
Third-party volume numbers are rough estimates that often miss long-tail B2B and technical searches. Search Console (clickd_queries) shows the exact impressions, clicks and positions your site already earns on Google.
What if our site is new and has little Search Console data?
The skill falls back to clickd_rivals and clickd_gaps to map the topics and buyer questions your closest competitors rank for, grouped into one page per search intent.
Install it by hand
The line at the top adds it for any agent. Or fetch its 1 files where your agent looks. Your agent also needs clickd connected over MCP: connecting an agent shows how.
Loads skills from .claude/skills/ in the repository, for everyone who works in it, or ~/.claude/skills/ for every project on your machine. Claude uses it when a request matches its description, or type /clickd-keyword-research.
Puts the files in .claude/skills/clickd-keyword-research/.
Loads skills from .agents/skills/ in the repository and its parent folders, or ~/.agents/skills/ for every project. Codex uses it when a task matches its description, or mention it as $clickd-keyword-research; /skills lists them.
Puts the files in .agents/skills/clickd-keyword-research/.
Loads skills from .cursor/skills/ or .agents/skills/ in the repository, or ~/.cursor/skills/ for every project. It also reads .claude/skills/, so one copy there serves both. The agent uses it when it is relevant, or type / in Agent chat and pick it.
Puts the files in .cursor/skills/clickd-keyword-research/.
An agent that does not load skills can still follow the file. Save the folder in the repository and point its standing instructions at it, with a line in AGENTS.md, which many agents read. The agent reads AGENTS.md when it starts, and follows the file when the task matches.
SKILL.md stays short and sends your agent to a reference when a step needs it.
SKILL.mdThe four steps from pulling real Search Console queries and rival topic gaps to bucketing by action, checking live search intent, and turning targets into code edits or briefs.88 lines
Each is in a ## Sources list in the file that cites it, by the page’s own title.
Google Search Central
1Search Analytics: Performance Report, Search Console Help.How Search Console counts impressions, clicks, average position and click-through rate by query and page.