Ask Claude for the search volume of a keyword and you will get a number. It will look clean. It will often be wrong, because Claude has no live connection to Google’s query data unless you give it one.
Quick answer: install AgriciDaniel/claude-seo or kostja94’s keyword-research skill for the method, then connect real numbers through the DataForSEO or Ahrefs MCP server, or a Google Search Console export. The skill handles seed expansion, intent and topical maps; the data source supplies volume and difficulty. For grouping a finished list into pages, see skills for keyword clustering.
That one fact shapes everything in this guide. Claude Code skills are very good at the thinking parts of keyword research: expanding seeds, grouping terms, tagging intent, and turning a messy list into a topical map. They are not a data source. The setups that work pair a skill (the method) with an MCP server or an export (the numbers).
Below you will find the skills worth installing in late 2026, free and paid, with GitHub stars and last-commit dates checked on 1 November 2026, install commands, and a workflow that keeps invented numbers out of your plan.
What Claude Code skills for keyword research actually do
Short answer: A keyword research skill is a folder with a SKILL.md file that teaches Claude Code a repeatable research method. It tells Claude which steps to run (seed expansion, clustering, intent tagging, prioritising), which data to ask for, and what format to output. The skill supplies the process. Volume and difficulty must come from a connected tool or a file you provide.
If you are new to the format, what Claude Code skills are covers the basics. The short version: skills live in ~/.claude/skills/ (available in every project) or .claude/skills/ inside a project, and Claude loads one when your request matches its description. Many now ship inside plugins, which bundle skills, slash commands, and sometimes MCP configs.
The five jobs a keyword skill can handle
| Job | What Claude does well | What it needs from you |
|---|---|---|
| Seed expansion | Generates modifiers, questions, comparisons, use cases, jargon variants | A product description and audience |
| Clustering | Groups keywords by shared meaning or shared SERP results | The keyword list (and SERP data for SERP-based clustering) |
| Intent tagging | Labels informational, commercial, transactional, navigational | Ideally, a look at what currently ranks |
| Topical maps | Arranges clusters into pillar and supporting pages | Your existing URLs, so it maps gaps and not duplicates |
| Prioritisation | Scores by fit, difficulty, and business value | Real volume and difficulty numbers |
Notice the right column. Every row gets better with real inputs. The last row is impossible without them.
Why a skill beats a one-off prompt
A prompt is something you retype and tweak every time. A skill is a written-down method that runs the same way each time. For keyword research that matters, because consistency is the whole point: you want this month’s cluster labels to match last month’s, and you want the intent rules applied the same way across 2,000 rows. A good skill also says what to do when data is missing, which is where most ad hoc prompts fall apart.
Takeaway: Install a skill for the method, connect a data source for the numbers, and never let the first stand in for the second.
The warning first: Claude alone does not know search volumes
Short answer: Large language models do not have live access to Google search volume, keyword difficulty, or CPC. When asked, they may produce plausible figures anyway. Treat any number Claude gives you without a cited tool call or file as a guess, and do not build a content plan on it.
This is not a theoretical worry. SpyFu published a post documenting ChatGPT returning volume and CPC figures that looked professional but were made up, including keywords that did not exist (SpyFu). Ross Simmonds compared ChatGPT’s stated volumes for SEO terms against Ahrefs and found they did not match (Ross Simmonds). The same mechanism applies to Claude: the model predicts text, and a number is just text.
How to spot a fabricated metric
- No tool call appears in the Claude Code transcript before the number shows up.
- Figures are suspiciously round (1,000, 2,400, 5,000) across a whole list.
- Difficulty scores appear without naming whose scale they use (Ahrefs KD and Semrush KD% are different scales).
- The keyword itself returns nothing when you check it in Google Keyword Planner or a paid tool.
The rule to write into your own skill
If you build or edit a skill (see how to build a Claude Code skill), add a line like this to SKILL.md:
## Data rules
- Never state search volume, keyword difficulty, CPC, or traffic unless it came
from a tool call (DataForSEO, Ahrefs, Semrush MCP) or a file the user supplied.
- If no data source is available, output the keyword list with a "volume: unknown"
column and tell the user which tool to connect.That single instruction removes the most common failure in AI keyword research. Several community skills already include a version of it. The kostja94 keyword-research skill, for example, opens by asking which tool access you have (Keyword Planner, Trends, or an SEO tool) before it does anything else.
Takeaway: A skill without a data rule is a liability. Check for one before you install, or add it yourself.
The best Claude Code skills for keyword research in 2026 (free)
Short answer: The strongest free options are AgriciDaniel’s claude-seo (includes SERP-based clustering and DataForSEO and Ahrefs extensions), kostja94’s marketing-skills (a dedicated keyword-research skill), manojbajaj95’s claude-gtm-plugin (keyword-research-and-clustering), and the content-strategy skills in coreyhaines31/marketingskills and Anthropic’s knowledge-work-plugins.
Stars and dates below come from the GitHub API on 1 November 2026. Stars measure attention, not quality, so read the SKILL.md before you trust any of them.
1. claude-seo (AgriciDaniel)
- Source: github.com/AgriciDaniel/claude-seo
- Price: Free, MIT licence. A private version with early features is offered to members of the paid AI Marketing Hub Pro community on Skool (pricing not public in the repo).
- Stars / last push: about 25,400 / 1 Nov 2026
- What it does: A large SEO plugin (the README lists 26 sub-skills and 19 sub-agents). For keyword work the useful parts are
/seo cluster <keyword>for SERP-based semantic clustering,/seo content-brief <topic>for target keywords and outlines, and optional extensions for DataForSEO, Ahrefs, and SE Ranking that pull live data. - Install:
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seoThe README notes it needs Claude Code 1.0.33 or newer and Python 3.11 to 3.14.
2. marketing-skills (kostja94)
- Source: github.com/kostja94/marketing-skills
- Price: Free, open source
- Stars / last push: about 1,070 / 26 Oct 2026
- What it does: 160+ marketing skills. The
skills/seo/content/keyword-researchskill covers discovery methods (autocomplete, People Also Ask, the alphabet method), intent, difficulty, and topical maps. Clustering is handed to itscontent-strategyskill. There is also agoogle-search-consoleskill under analytics. - Install:
npx skills add kostja94/marketing-skills(add--listto see every skill first)
3. claude-gtm-plugin (manojbajaj95)
- Source: github.com/manojbajaj95/claude-gtm-plugin
- Price: Free
- Stars / last push: about 110 / 18 Sep 2026
- What it does: GTM skills including
keyword-research-and-clustering,seo-and-aeo-strategy, andcontent-strategy-and-planning. Its marketplace listing pitches it as keyword research without expensive SEO tools, which is fine for ideation and grouping, but the volume warning above still applies. - Install:
claude plugin marketplace add manojbajaj95/claude-gtm-plugin, then install from/plugin
4. marketingskills (coreyhaines31)
- Source: github.com/coreyhaines31/marketingskills
- Price: Free
- Stars / last push: about 57,800 / 31 Oct 2026
- What it does: The most-starred marketing skill repo. It has no dedicated keyword research skill, but
content-strategyhandles topic selection,site-architecturecovers hierarchy and internal linking (useful for turning clusters into a map), andprogrammatic-seobuilds pages from keyword-driven templates. - Install:
npx skills add coreyhaines31/marketingskills(or pick skills with--skill content-strategy site-architecture)
5. knowledge-work-plugins (Anthropic, official)
- Source: github.com/anthropics/knowledge-work-plugins
- Price: Free, Apache 2.0
- Stars / last push: about 34,300 / 1 Nov 2026
- What it does: Anthropic’s role-based plugins, built for Claude Cowork and usable in Claude Code. The Marketing plugin includes a
seo-auditskill. The small-business plugin addscontent-strategyandseo-ai-visibility. None is a dedicated keyword tool, so treat these as the official baseline rather than a full keyword stack. - Install:
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install marketing@knowledge-work-plugins6. claude-skills (alirezarezvani)
- Source: github.com/alirezarezvani/claude-skills
- Price: Free
- Stars / last push: about 29,100 / 31 Oct 2026
- What it does: A very large multi-domain collection (the description claims 380+ skills). Its marketing folder includes
content-strategy,programmatic-seo, andseo-audit. Useful if you want one repo for many jobs; not specialised for keywords. - Install: follow the repo README; it supports Claude Code plus several other agents.
7. seomachine and claude-blog (writing workspaces)
- seomachine: github.com/TheCraigHewitt/seomachine, about 7,570 stars, last push 23 Oct 2026. A Claude Code workspace for researching and writing long-form SEO content. Keyword research is one stage of a writing pipeline.
- claude-blog: github.com/AgriciDaniel/claude-blog, about 2,600 stars, last push 30 Oct 2026. A blog skill suite with 30 sub-skills, from the claude-seo author.
Both fit people who want keywords to flow straight into briefs. If that is you, also see Claude Code skills for content writers.
Where to find more
Directories index thousands of community skills, including several named keyword-research: claudemarketplaces.com, claudskills.com, skills.sh, and the curated ComposioHQ/awesome-claude-skills list (about 78,500 stars). Quality varies a lot. One claudskills listing named “keyword-research” sits under Science and Research and has nothing to do with SEO, so read the file, not the title.
Paid skills and data sources for keyword research
Short answer: Most keyword skills are free. What you pay for is data: DataForSEO (pay as you go, $50 minimum top-up), Ahrefs (API and MCP access from the $129/month Lite plan), and Semrush (MCP on qualifying SEO plans). A few paid skill packs exist, but the ones checked here are prompt-and-method bundles, not data.
Paid data via MCP
| Source | Price (checked 1 Nov 2026) | Keyword data you get | How to connect |
|---|---|---|---|
| DataForSEO | Pay as you go, $50 minimum deposit, $1 trial credit; Keywords Data API from $0.0012 per task (pricing) | Google Ads volume, CPC, competition, Trends, SERPs, keyword ideas | Official server: dataforseo/mcp-server-typescript (npm dataforseo-mcp-server), or the hosted endpoint in DataForSEO’s help center |
| Ahrefs | Lite $129/mo, Standard $249/mo, Advanced $449/mo all list “API and MCP access”; Starter $29/mo and Free do not (pricing) | Volume, KD, traffic potential, SERP overview, competitor keywords | Official remote MCP server with OAuth (see Ahrefs docs) |
| Semrush | MCP on Semrush One Starter, Pro+, SEO Classic Pro or Guru (50,000 API units included); plan prices on Semrush’s site (docs) | Volume, KD%, intent, keyword gap, positions | claude mcp add semrush https://mcp.semrush.com/v2/mcp -t http, then /mcp and Authenticate |
Ahrefs notes that billing annually can lower those prices, so confirm the current figure on its pricing page before you budget.
Paid skill packs
| Pack | Price | What is in it | Keyword research? |
|---|---|---|---|
| The Ultimate Claude SEO & AEO Skills Pack (Samuel Uebersax, Contra) | $39 (listed down from $79) (Contra) | 12 skill files, 2 Webflow bonus skills, setup guide | No dedicated skill; has a Keyword Intent Classifier and a brief writer that maps keywords |
| Master Marketing Skills Bundle (inflectual, Gumroad) | $167 one-time (Gumroad) | Title claims 167 Claude marketing skills | Could not confirm |
| Keyword Research and SERP skill (virginai-skills) | Listed at $97 in search results; not confirmed on the product page | Described as pulling volume and difficulty and analysing top SERPs | Yes, as described |
An honest read: paid packs can save setup time and give you a tested method. None of them fixes the data problem on its own. If a paid skill claims to “pull volume”, ask which API it calls and who pays for that API.
Free vs paid at a glance
| Factor | Free skills (GitHub) | Paid skill packs | Paid data (MCP) |
|---|---|---|---|
| Cost | $0 | About $39 to $97 where public | $50 deposit (DataForSEO) to $129+/mo (Ahrefs) |
| Real search volume | Only with a connected source | Only with a connected source | Yes |
| Clustering and intent | Yes | Yes | Raw data only; needs a skill to organise it |
| Updates | Visible commit history | Depends on seller | Vendor-maintained |
| Transparency | Full SKILL.md before install | Often after purchase | API docs public |
| Best for | Most people starting out | Teams wanting a packaged method | Anyone making real prioritisation calls |
Takeaway: Spend money on data before you spend it on skills. A free skill plus a $50 DataForSEO balance beats a paid pack with no data.
How to connect real data: MCP servers and GSC exports
Short answer: There are two ways to give a skill real numbers. Connect an MCP server (DataForSEO, Ahrefs, Semrush) so Claude can query live data, or export a CSV from Google Search Console, Keyword Planner, or your SEO tool and drop it in the project folder. Exports are free and private; MCP is faster and covers keywords you do not rank for yet.
Option A: an MCP server
Claude Code registers MCP servers with claude mcp add. For Semrush the vendor gives the exact command (above). For DataForSEO, the official npm package reads your API login and password from environment variables. A typical registration looks like this, but check DataForSEO’s current setup page, since guides disagree on syntax:
claude mcp add dataforseo \
-e DATAFORSEO_USERNAME=your_login \
-e DATAFORSEO_PASSWORD=your_password \
-- npx -y dataforseo-mcp-serverRun claude mcp list to confirm it registered. Then a skill such as claude-seo’s DataForSEO extension can call it directly. Ryan Doser’s walkthrough of this setup is a useful second reference (ryandoser.com).
Option B: a Google Search Console export
GSC tells you what Google already shows your site for, with real impressions, clicks, and average position. It does not give total search volume, and it hides some rare queries for privacy. Still, for an existing site it is the most honest data you can get for free.
- In Search Console, open Performance, set a date range (16 months is the maximum), and export queries as CSV.
- Save it as
data/gsc-queries.csvin your project. - Ask Claude Code: “Using the keyword research skill, cluster
data/gsc-queries.csv, tag intent, and flag queries with high impressions and position 8 to 20.”
That last filter is the classic “striking distance” list: queries where you already appear but rank low enough that a better page could move you up.
Which source when
| Situation | Best source |
|---|---|
| New site, no rankings | DataForSEO or Ahrefs/Semrush MCP |
| Existing site, refreshing content | GSC export first, then MCP for gaps |
| Tight budget | GSC plus Keyword Planner export plus a free skill |
| Agency, many clients | Ahrefs or Semrush MCP (one login, many projects) |
The Claude Code skills for SEO guide covers the wider MCP setup if you also want audits and rank checks.
A keyword research workflow in Claude Code, step by step
Short answer: Run five passes: expand seeds, pull real metrics, cluster, tag intent, then build a topical map with priorities. Keep each pass as a separate file so you can check Claude’s work between steps.
This is a method, not a reported result. Adjust thresholds to your niche.
Step 1: seed expansion
Give Claude your product, audience, and 3 to 10 seed terms. Ask for modifiers (best, vs, alternative, for [role], template, price), question forms (how, what, can, does), and the jargon your customers use. A good skill will also ask for competitor URLs. Output: 01-seeds-expanded.csv with a source column showing where each idea came from.
Step 2: pull metrics
Send the list through your MCP source or join it against an export. Add volume, difficulty, CPC, and the top-ranking URLs. Drop zero-volume terms only if you have a reason; low-volume, high-intent terms are often the easiest wins. Output: 02-with-metrics.csv.
Step 3: clustering
This step gets its own guide: keyword clustering skills. The short version: there are two methods, and it helps to know which your skill uses:
| Method | How it works | Strength | Weakness |
|---|---|---|---|
| Semantic | Claude groups by meaning | No extra data cost | Can merge terms Google treats as separate pages |
| SERP-based | Keywords sharing several top-10 URLs go together | Mirrors how Google groups intent | Needs SERP data for every keyword |
claude-seo’s /seo cluster command is described as SERP-based. If your skill is semantic only, spot-check clusters against a live search.
Step 4: intent tagging
Tag each cluster informational, commercial, transactional, or navigational. Better skills look at what ranks: if the top results are all product pages, a blog post will struggle no matter how good the intent label looks on paper.
Step 5: topical map and priorities
Turn clusters into pillar pages and supporting pages, link them, and map them against URLs you already have so you are not creating duplicates. Score each by business fit, difficulty, and volume. Output: 05-topical-map.md plus a prioritised list.
For a non-technical version of this setup, Claude Code skills for non-developers walks through the terminal basics.
Common mistakes with AI keyword research
Short answer: The biggest mistakes are trusting unsourced numbers, clustering by meaning when Google splits by SERP, mixing difficulty scales from different tools, ignoring your existing pages, and installing skills without reading them.
- Trusting a number with no source. Covered above, and still the top problem.
- Over-merging clusters. “Keyword research tool” and “keyword research guide” sound alike but often rank different page types.
- Mixing KD scales. Ahrefs KD and Semrush KD% are calculated differently. Pick one tool per project.
- Ignoring cannibalisation. Without your URL list, Claude will happily propose a new page for a topic you already cover.
- Skipping the SERP check. Intent labels from text alone miss SERP features like forums, videos, or AI Overviews.
- Installing blind. A skill can include scripts that run on your machine. Read
SKILL.mdand any scripts first, as several directories themselves advise. - Too many overlapping skills. Three keyword skills with similar descriptions make it harder for Claude to pick the right one. Keep one per job.
How to install Claude Code skills explains how to check and remove what is installed.
Who this setup is for (and who should skip it)
Short answer: It suits SEO leads, content strategists, and marketers who already use a keyword tool and want faster clustering and mapping. It is a poor fit if you need a one-click volume lookup and nothing else; your SEO tool’s own interface is quicker for that.
| Profile | Fit | Why |
|---|---|---|
| In-house SEO with Ahrefs or Semrush | Strong | Already paying for data; MCP makes it queryable in plain English |
| Content marketer on a budget | Good | Free skill plus GSC export covers a lot |
| Freelance writer | Good | Pairs with writing skills; see Claude Code skills for copywriters |
| Agency running many sites | Strong | Same workflow, repeatable per client |
| Someone wanting quick lookups | Weak | A browser tab is faster |
Marketers who want the wider picture can start at Claude Code skills for marketers, and anyone combining keywords with site health checks should read Claude Code skills for SEO audits.
Frequently asked questions
Can Claude do keyword research?
Yes, for ideation, clustering, intent tagging, and planning. It cannot give reliable search volume or difficulty on its own. Connect a data source or provide an export.
Does Claude know search volume?
No. It has no live access to Google query data. Numbers it states without a tool call or file are guesses.
Is AI keyword research accurate?
The organising work can be accurate. The metrics are only as accurate as the tool behind them. Several published tests found chatbot volumes did not match real tools.
What is the best Claude Code skill for keyword research?
For a dedicated skill, kostja94’s keyword-research or manojbajaj95’s keyword-research-and-clustering. For a full SEO suite with clustering and data extensions, AgriciDaniel’s claude-seo.
Are there free Claude Code skills for keyword research?
Yes. Every GitHub skill named in this guide is free. Data sources are where costs appear.
Can I use Ahrefs with Claude Code?
Yes, through Ahrefs’ official MCP server. The pricing page lists API and MCP access on Lite ($129/mo) and above.
Does Semrush have an MCP server?
Yes. Semrush documents a remote MCP endpoint and a claude mcp add command. It needs a qualifying SEO plan with API units.
What is the cheapest way to get real keyword data into Claude?
A Google Search Console export (free) or DataForSEO’s pay-as-you-go API with a $50 minimum top-up.
How do I install a keyword research skill?
Use npx skills add owner/repo for skill repos, or /plugin marketplace add then /plugin install for plugins. Skills land in ~/.claude/skills or .claude/skills.
Can Claude cluster keywords?
Yes. Semantic clustering works with any list. SERP-based clustering needs ranking data for each keyword.
Can Claude build a topical map?
Yes. Give it clustered keywords and your existing URLs, and ask for pillars, supporting pages, and internal links.
Can I use Google Search Console data with Claude Code?
Yes. Export the queries report as CSV, save it in the project, and point the skill at it.
Is ChatGPT or Claude better for keyword research?
Both have the same limit: no native volume data. In Claude Code the advantage is skills plus MCP, which turns a chat into a repeatable workflow on your files.
Will AI replace Ahrefs and Semrush?
Not for data. These tools collect the data that AI workflows depend on. AI changes how you query and organise it.
Are paid Claude skill packs worth it?
Sometimes, for a packaged method. Check whether the pack includes a real keyword skill and whether it relies on a data API you must pay for separately.
Sources and methodology
GitHub stars and last-push dates were pulled from the GitHub API on 1 November 2026. Install commands come from each repo’s README. Prices come from vendor pages fetched the same day: DataForSEO pricing, Ahrefs pricing, Semrush MCP docs, and the Contra listing. Where a price could not be confirmed, it is marked. Reliability concerns about AI-stated volumes draw on SpyFu and Ross Simmonds. This guide does not report hands-on benchmarks of the skills.
Your next step
Pick one free skill, connect one data source, and run the five-step workflow on a single topic. If the clusters hold up against a live search, scale it. If they do not, fix the data rule in SKILL.md before you go further.
