Claude has become the default AI assistant for a lot of SEO work. In Keyword.com's 2026 survey cited by Search Engine Land, 78% of respondents said they use Claude, ahead of ChatGPT at 57%. The sample was small (97 responses), but the direction matches what I see: founders and SEOs now ask Claude to plan content, cluster keywords and audit pages.
The search results for "Claude SEO skill" show what happened next. People started packaging SEO workflows as Claude skills, sharing them on GitHub and Reddit, and recording YouTube videos about ranking #1 in ten minutes.
Most of those guides skip the part that matters. A skill teaches Claude a method. It does not give Claude your data. Without your Search Console, real keyword volumes or a live view of Google's results, Claude still works from what it can read on a page and what it remembers, and that is where confident but wrong SEO advice comes from.
So the useful question is not "which Claude SEO skill is best?" It is what does Claude need to do real SEO work on your site, and which part does a skill actually solve?
I build RankBuddy, which connects SEO data to Claude through MCP, so I have a stake in this and I will say where it fits. For this guide I read Anthropic's documentation on skills and connectors, reviewed the most popular SEO skills, and ran a real question about my own site through Claude with SEO data connected.
Quick answer: a Claude SEO skill is a folder with a SKILL.md file that tells Claude how to do an SEO task, such as an audit or keyword clustering. Skills give Claude a repeatable method, but not data. To work with real numbers, connect an SEO data source through MCP: in the Claude app you add it as a custom connector (one is allowed on the Free plan), and in Claude Code skills can call it directly. The setup that works best for most founders is Claude + an SEO data connector + a person who decides what goes live. Add skills when a task repeats.
TL;DR: skills, Claude Code and MCP
- Claude on its own is good at drafting, clustering lists you give it and explaining exports. It cannot see your Search Console, search volumes or backlinks.
- A Claude SEO skill packages a method (an audit checklist, a clustering process) so Claude does it the same way every time. It adds no data by itself.
- Claude Code runs skills on your machine with access to files and scripts. Use it for site code, bulk changes and crawls. It also makes it easier to break things.
- MCP connects Claude to real data: Search Console, keyword metrics, Google's top 10, backlinks. This is the part that turns guesses into analysis.
- A person still decides what gets published, redirected or rewritten.
What is a Claude SEO skill?
A skill is a folder that contains a SKILL.md file. The file starts with a short YAML header, then plain instructions. Claude reads the name and description of every installed skill, and when your request matches a description, it loads the full instructions and follows them.
Here is a minimal SEO skill. It checks pages that lost clicks and uses an SEO data connector for the numbers:
---
name: content-decay-check
description: Finds pages that lost Google clicks in the last 28 days and explains why. Use when the user asks which pages dropped, lost traffic or need a refresh.
---
# Content decay check
1. Pull Search Console clicks and impressions by page for the last 28 days and for the 28 days before.
2. List pages whose clicks or impressions fell by more than 30%. Ignore pages with fewer than 20 impressions in both periods.
3. For each page, show the queries that lost the most impressions and the current position.
4. Check whether the page is still indexed.
5. Suggest one fix per page: refresh, merge, retitle or leave alone. Say which fix you are least sure about.
Never invent numbers. If the data source is not connected, stop and say so.
The last line matters more than it looks. A skill without data does not stop Claude from answering. It stops Claude from answering with numbers it cannot see only if you tell it to.
According to Claude Code's documentation, personal skills live in ~/.claude/skills/<skill-name>/SKILL.md and project skills in .claude/skills/<skill-name>/SKILL.md, which you can commit to share with a team. In the Claude app, skills are available on Pro, Max, Team and Enterprise plans, need code execution turned on, and are uploaded as a ZIP file under Settings, then Capabilities.
The Claude SEO skills people actually use
| Skill | What it does | How to install | Where its data comes from |
|---|---|---|---|
| claude-seo (open source, MIT) | 25 sub-skills and 32 /seo commands: full audits, technical SEO, schema, content quality, local, hreflang, AI search |
Claude Code plugin, plus Python 3.10+ | The pages you point it at. Optional extensions for Google APIs, DataForSEO, Ahrefs, SE Ranking and others need your own accounts |
| SE Ranking SEO skills | 7 skills: content briefs, AI search visibility, backlink gaps, ads intelligence, content audit, clustering | Claude Desktop plugin or Claude Code plugin | SE Ranking's MCP server, which needs an SE Ranking account |
| Your own skill | Your exact process, your rules | A folder with SKILL.md |
Whatever connector or files you give it |
claude-seo is the most popular open-source option, with about 15,100 GitHub stars when I checked in October 2026. You install it in Claude Code with two commands and then run /seo setup:
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo
Its own README is honest about the limits. Without credentials, Core Web Vitals are lab estimates rather than field data, and indexation is inferred from page signals rather than read from Google. That is exactly the gap between a skill and data.
SE Ranking's skills are the opposite design: the skills are thin, and the value comes from SE Ranking's data through its MCP server. That makes sense if you already pay for SE Ranking. Without the account, the skills have nothing to work with.
Writing your own is often the best choice once you have a process you repeat every week. Five to ten small skills, each doing one job well, beat one giant "do my SEO" skill that tries to cover everything.
Claude Code for SEO: when you need it and when you don't
Most SEO skills on GitHub assume Claude Code, the command-line version of Claude that works inside your project folder. It can read and edit files, run scripts and crawlers, and use plugins.
You need Claude Code when the work touches your site's files:
- Bulk changes to titles, meta descriptions or internal links across a Next.js, Astro or Hugo site.
- Crawling and rendering your own pages with a headless browser.
- Running scripts against exports, sitemaps or logs.
You do not need it to analyze performance, research keywords, check competitors or plan content. The Claude app with a data connector does those without a terminal.
Claude Code also makes it easier to break things. Will Scott described in Search Engine Land how Claude, asked to target new keywords, cloned his homepage into new URLs and only changed the title and H1. The clones got zero impressions while the original kept ranking. His rule is the one I follow: let Claude research and draft, and keep a person responsible for anything that goes live.
Why a skill without data guesses, and how MCP fixes it
Ask Claude "which of my pages lost traffic?" without a data source and one of two things happens. It tells you it cannot see your analytics, or, worse, it gives you a plausible list based on what your pages look like.
MCP (Model Context Protocol) is the standard way to give Claude tools that read real data. For SEO, the useful ones are:
| MCP data source | What Claude gets | What you need |
|---|---|---|
| RankBuddy (my product) | Your Search Console, keyword volume and difficulty, Google's live top 10, competitor keywords and gaps, backlinks and Domain Rating | A RankBuddy account. Connects as a custom connector, no code or API keys |
| Ahrefs MCP | Ahrefs data: keywords, backlinks, site metrics | Ahrefs Lite or higher (API units included) |
| Semrush MCP | Semrush reports | An eligible Semrush plan; each request uses API units |
| SE Ranking MCP | SE Ranking keyword, backlink, audit and AI visibility tools | An SE Ranking account |
| DataForSEO MCP | Raw SERP, keyword and backlink data | A pay-as-you-go DataForSEO account |
| Open-source Search Console MCP servers | Your Search Console data | Running the server yourself and setting up Google API credentials |
I compared these data connectors in more detail in my guide to SEO chat tools. The short version: if you already pay for Ahrefs or Semrush, start with their MCP. If you do not, or you mainly want your own Search Console plus market data in one place without running a server, that is the gap RankBuddy was built for.
How to connect RankBuddy to Claude
This takes about two minutes and works in the Claude web app, the desktop app and mobile. According to Anthropic's help center, custom connectors are available on Free, Pro, Max, Team and Enterprise plans, and Free users are limited to one custom connector.
- Create a RankBuddy account and, optionally, connect Google Search Console for your site so Claude can see your own clicks, impressions and positions.
- In Claude, open Add custom connector: click + in the message box, then Connectors, Add connector and Add custom connector. You can also go to Settings, then Connectors.
- Name it and paste the URL: name
rankbuddy, URLhttps://mcp.rankbuddy.io/mcp, then Continue. - Keep the default sign-in settings. Claude detects RankBuddy's sign-in on its own. Leave the detected options as they are and confirm.
- Sign in to RankBuddy and click Allow. The screen lists what Claude gets access to. Everything is read-only.
- Turn it on in your chat: +, then Connectors, and switch rankbuddy on.
Optionally, set RankBuddy's tools to Always allow in Settings, then Connectors, so Claude stops asking before each lookup. Every RankBuddy tool only reads data, so it cannot change your site, publish anything or see your billing.
Two details worth knowing. One question in Claude uses one RankBuddy request, however many tools Claude calls to answer it, and a failed tool uses none. Claude's own usage counts against your Claude plan, not RankBuddy's.
SEO audit in Claude with live data: a real example
To show the difference data makes, I asked Claude, with RankBuddy connected, two questions about rankbuddy.io in October 2026.
Question 1: which pages lost visibility in the last 28 days? Claude pulled Search Console data by page for two periods, 13 August to 9 September and 12 September to 9 October 2026, and compared them:
| Page | Impressions before | Impressions after |
|---|---|---|
| Best AI blog writer for Next.js | 14 | 1 |
| GSC clicks vs impressions | 8 | 1 |
| Auto SEO software: AI agents vs traditional tools | 7 | 1 |
| Automated SEO monitoring | 4 | 2 |
It also pointed out something a human analyst would: the whole site had only 179 impressions by page in the last 28 days, so these drops are too small to judge click-through rates, and the right reading is "visibility is still forming" rather than "these pages are failing". That caveat came from the data itself, not from a guess.
Question 2: which keywords does a competitor rank for that I do not? Claude compared chatseo.app with rankbuddy.io and found 178 keywords where chatseo.app ranks and RankBuddy does not, led by "seo agent" (27,100 monthly searches, KD 19, chatseo.app at position 30) and alternative queries such as "ahrefs alternatives" (590, KD 2) and "semrush free alternative" (480, KD 4).
Without a data connector, Claude cannot answer either question. With a skill but no connector, it can describe how to answer them. With the connector, it answers in a minute and shows its sources. A skill like the content-decay example above then makes the same check run the same way every week.
What I did not test: I did not benchmark claude-seo or SE Ranking's skills on the same questions, so I describe them from their documentation rather than claim results.
Prompts to try with Claude and SEO data
Swap in your own domains:
- "Which of my pages lost clicks over the last 28 days, and are they still indexed?"
- "Which keywords does competitor.com rank for that mysite.com does not? Show only those with KD under 30."
- "Compare the backlinks and Domain Rating of site-a.com, site-b.com and site-c.com."
- "Find keyword ideas around 'your topic' and show who ranks in the top 10 for the best one."
- "Which of my queries are on positions 5 to 15 with the most impressions? Suggest a title change for the top three."
- "Check the title, meta description, canonical and robots tags on this URL and tell me what is wrong."
For a full process from crawl to fixes, my technical SEO audit tutorial shows how to run an audit with AI step by step, and GSC clicks vs impressions explains how to read what Claude finds.
Where Claude and SEO skills go wrong
- Cloning instead of creating. Asked for new pages, Claude may copy an existing one and change the headline. Check every new page against what already exists.
- Inventing numbers. Without a data connector, search volumes, difficulty scores and traffic are estimates at best. Tell your skills to stop when data is missing.
- Trusting the HTML. Many SaaS sites render content with JavaScript. A skill that only reads raw HTML can report missing content that Google actually sees.
- Robots.txt, canonicals and redirects. These are easy to misread and expensive to get wrong. Verify with Search Console's URL inspection before changing anything.
- Claims about live URLs. Anything Claude says about whether a page is indexed or ranking should come from a tool, not from memory.
- Letting it publish. Keep publishing, redirects and deletions behind a human approval.
Can you rank in Claude's answers?
Partly, and it is a separate question from using Claude for SEO. When Claude searches the web, it cites pages that are easy to find and clearly answer the question, so the fundamentals are the same as for Google and other assistants: crawlable pages, clear answers, and being mentioned on sites that already rank. I cover what works in how to rank in ChatGPT, and most of it applies to Claude.
To be clear about my own product: RankBuddy's AI visibility tracking currently checks ChatGPT, Gemini and Perplexity. It does not track Claude's answers yet.
How I researched this
I read Anthropic's documentation on skills in Claude Code and the Claude apps and on custom connectors, the README of the claude-seo plugin, SE Ranking's page about its Claude skills, and the guides that rank in Google for "claude seo skill" and "claude for seo", including Search Engine Land and StudioHawk. I checked plan availability and setup steps in October 2026.
The audit example is real data from rankbuddy.io, pulled through RankBuddy's MCP tools in October 2026. I did not run the same questions through claude-seo or SE Ranking's skills, so I do not compare their output.
I build RankBuddy, which I describe above, and I have tried to be clear about what it does not do: it does not change your site, it is not a skill library, and it does not track Claude's answers yet.
My verdict
Skills are worth it, but they are the second step. A skill makes Claude consistent. It does not make Claude informed.
Data is the first step. Connect Search Console and an SEO data source through MCP before you build skills on top. If you already pay for Ahrefs, Semrush or SE Ranking, use their MCP. If you want your own Search Console plus keywords, Google's top 10, competitors and backlinks in one read-only connector, that is what I built RankBuddy for. My AI agent for SEO guide explains where this goes next.
Use Claude Code only when the work touches your files. For analysis and planning, the Claude app with a connector is enough.
Keep a person in charge of what goes live. Claude is a fast researcher and a good drafter. It is not the one who should decide to redirect your homepage.
That is the difference between asking Claude about SEO and SEO execution.