# Claude SEO: website SEO audits and action plans (AgriciDaniel/claude-seo) — skill card for AI agents

> AgentGid skill card. Checked on 2026-11-02 against commit 6be93f4 (version 2.5.0).
> Every fact comes from the pack's own files at that commit. Numbers marked "as reported" are the author's.
> Human page: https://agentgid.com/skills/agricidaniel/claude-seo/

> Note: the pack has changed since this check; re-read its files before relying on details.

## Explain it to a non-programmer
Give your assistant a website address and it checks the site the way an SEO consultant would: speed, indexing, content quality, structured data, local search and AI search. It saves a report and a to-do list sorted from critical to nice-to-have. Separate commands plan a strategy, write content briefs and watch a page for changes.

## Who it is for
- You run or build a website and want to know why it does not show up well in Google or in AI answers
- You are an SEO agency, in-house team or freelancer who wants repeatable audits and a prioritised plan
- You use Claude Code and can run a few commands in it
- You want help with local SEO, an online shop, structured data or several languages
- Not for: You do not use Claude Code: the author ships it as a Claude Code plugin
- Not for: You want a fast deep crawler or a large backlink database: the author says crawlers and big SEO suites do that better

## What it is
- Kind: command, knowledge, scripts
- License: MIT
- Repository: https://github.com/AgriciDaniel/claude-seo

## What it does
- Detects the kind of business (SaaS, local, shop, publisher, agency) from the home page and adjusts the checks
- Crawls up to 500 pages, respecting robots.txt, and splits the work between specialist helpers: technical, content, schema, sitemap, speed, visuals, AI search, agent readiness, search experience
- Adds local, maps, Google data, backlinks, clustering, shop and change-tracking helpers only when the site or your keys call for them
- Rolls everything into an SEO Health Score from 0 to 100 and sorts fixes into Critical, High, Medium, Low
- Each recommendation says what it rests on, what it depends on and how you would know it failed
- Offers a PDF report at the end
- Never cuts: never recommends HowTo schema; Core Web Vitals use INP, never FID; a hard stop at 50+ near-identical location pages

## Skills
| Skill | Call | Changes code? | In plain words |
|---|---|---|---|
| seo | /seo [command] [url] | no | The front door: lists every command and runs the full audit |
| seo-audit | /seo audit <url> | no | Checks a whole website and writes a scored report and a fix list |
| seo-page | /seo page <url> | no | Looks at one page in depth and gives it a score card |
| seo-technical | /seo technical <url> | no | Checks crawling, indexing, security, mobile, speed and rendering |
| seo-content | /seo content <url> | no | Rates how useful and trustworthy the text is, and can tidy your own drafts |
| seo-content-brief | /seo content-brief <topic or url> | no | Writes a brief for a new or existing article or page |
| seo-schema | /seo schema <url> | yes | Finds, checks and writes the hidden labels search engines read |
| seo-geo | /seo geo <url> | no | Checks how likely AI answers (Google AI Overviews, ChatGPT, Perplexity) are to quote your pages |
| seo-agentic | /seo agentic [audit|fix|lighthouse|refresh] <url> | no | Checks whether AI agents can use the site and drafts fixes for you to review |
| seo-plan | /seo plan <business-type> | no | Builds an SEO strategy for a SaaS, local, shop, publisher or agency site |
| seo-local | /seo local <url> | no | Checks local search: Google Business Profile, address consistency, reviews, map pack |
| seo-competitor-pages | /seo competitor-pages [url|generate] | yes | Plans "X vs Y" and "alternatives to X" pages |
| seo-drift | /seo drift baseline|compare|history <url> | no | Saves a snapshot of a page and later tells you what changed or broke |
| seo-images | /seo images <url> | unclear | Checks image text, sizes and formats; can convert images and add metadata |
| seo-sitemap | /seo sitemap <url or generate> | unclear | Checks the site map file search engines read, or writes a new one |
| seo-hreflang | /seo hreflang <url> | unclear | Checks and writes language and region tags for sites in several languages |
| seo-programmatic | /seo programmatic <url|plan> | no | Plans pages made in bulk from data without thin or duplicate pages |
| seo-sxo | /seo sxo <url> [keyword] | no | Checks whether your page type matches what searchers expect |
| seo-cluster | /seo cluster <seed-keyword> | unclear | Groups keywords into topics and plans hub and spoke pages with links |
| seo-ecommerce | /seo ecommerce <url> | no | Checks product pages, product labels and Shopping visibility for stores |
| seo-backlinks | /seo backlinks <url> | no | Looks at which sites link to you and flags risky links |
| seo-maps | /seo maps [command] | no | Local rank grids, reviews and nearby competitors; deeper tiers need paid data |
| seo-google | /seo google [command] | no | Pulls your real data from Search Console, PageSpeed, Chrome field data and Analytics; makes PDF reports |
| seo-flow | /seo flow [stage] | no | Hands out ready prompts for each stage of the FLOW method |
| seo-dataforseo | /seo dataforseo [command] | no | Live search and keyword data from the paid DataForSEO service (add-on) |
| seo-image-gen | /seo image-gen [use-case] <description> | unclear | Makes social preview, hero and product images with Gemini (add-on) |

### seo-audit
- Asks for: The address of the site
- Asks for: Optional: your Google, Moz or Bing keys for real data
- Gives: FULL-AUDIT-REPORT.md
- Gives: ACTION-PLAN.md, sorted Critical > High > Medium > Low
- Gives: One findings file per area (technical, content, schema, speed, visuals)
- Gives: Desktop and mobile screenshots
- Gives: An optional PDF report
- Say: "/seo audit https://example.com"

### seo-page
- Asks for: The address of one page
- Gives: Overall score out of 100
- Gives: Scores for on-page, content, technical, schema and images
- Gives: Issues sorted Critical to Low
- Say: "/seo page https://example.com/pricing"

### seo-technical
- Asks for: The address of the site
- Gives: A technical score; areas it could not measure are marked "not measured"
- Gives: A pass, warn or fail table per area
- Gives: A read-only check for a server stack with a known critical security hole
- Say: "/seo technical https://example.com"

### seo-content
- Asks for: A page address, or your own draft for cleanup
- Gives: A content quality score out of 100
- Gives: Experience, expertise, authority and trust scores
- Gives: On request: a cleaned draft with hidden characters and stock AI phrases swapped out
- Say: "/seo content https://example.com/blog/my-post"

### seo-content-brief
- Asks for: A topic or keyword, or a page to improve
- Asks for: Optional: the page type (blog, service page)
- Gives: Search intent
- Gives: A competitor table with scores and gaps
- Gives: An outline with word counts per section
- Gives: Where to place the main keyword
- Say: "/seo content-brief best running shoes for flat feet"

### seo-schema
- Asks for: A page address or the thing to describe (business, product, article)
- Gives: SCHEMA-REPORT.md with a status per label
- Gives: generated-schema.json, ready to paste
- Say: "/seo schema https://example.com/about"

### seo-geo
- Asks for: The address of a page or site
- Gives: GEO-ANALYSIS.md with a readiness score out of 100
- Gives: Which AI crawlers may read the site, for training and for search, listed apart
- Gives: llms.txt status
- Gives: Brand presence on Wikipedia, Reddit, YouTube, LinkedIn
- Gives: Answer-ready passages
- Say: "/seo geo https://example.com"

### seo-agentic
- Asks for: The address of the site
- Asks for: Your choice on AI training access, before it drafts robots.txt
- Gives: An agent-readiness report
- Gives: Drafts of robots.txt signals, llms.txt, ai-catalog.json and form tools, shown for review
- Say: "/seo agentic https://example.com"

### seo-plan
- Asks for: Business type, audience, competitors and goals
- Asks for: Budget and timeline
- Asks for: The current site, if there is one
- Gives: SEO-STRATEGY.md
- Gives: COMPETITOR-ANALYSIS.md
- Gives: CONTENT-CALENDAR.md
- Gives: IMPLEMENTATION-ROADMAP.md in 4 phases
- Gives: SITE-STRUCTURE.md
- Gives: Targets at 3, 6 and 12 months
- Say: "/seo plan saas"

### seo-local
- Asks for: The address of the business site
- Gives: A local SEO score out of 100
- Gives: Business Profile checklist
- Gives: Review snapshot
- Gives: Name, address and phone consistency
- Gives: Top 10 actions
- Gives: What it could not check and which paid tools fill the gap
- Say: "/seo local https://example-plumber.com"

### seo-competitor-pages
- Asks for: Your site and the competitor to compare with
- Gives: COMPARISON-PAGE.md with sections and a feature table
- Gives: Word count targets, at least 1,500 words
- Gives: comparison-schema.json
- Gives: Keywords to target
- Say: "/seo competitor-pages generate my app vs Competitor"

### seo-drift
- Asks for: The page address, before and after a change
- Gives: A saved snapshot: title, description, headings, structured data, speed, status code
- Gives: A comparison flagged by severity
- Gives: A change history
- Say: "/seo drift baseline https://example.com, then after your release: /seo drift compare https://example.com"

## Know before installing
- /seo setup installs a private Python environment and a headless Chromium browser. It runs only when you ask; installing the plugin itself runs no package managers.
- Your pages and data go to Claude like anything else in Claude Code. Some checks also read public sources such as Wikipedia, Reddit and Common Crawl.
- A full audit uses more of your Claude plan: five helpers run on Opus, the rest on Sonnet. You can change any helper's model in its file.
- After big reports it adds a footer with links to the author's free and paid community.
- Scores are its own estimates, not Google's ranking data; the skill tells the assistant to say so in reports.
- Fetched pages are treated as data: text on a page that addresses the assistant is reported, not obeyed.

## Install (official commands)
- Claude Code (plugin): `/plugin marketplace add AgriciDaniel/claude-seo` && `/plugin install claude-seo@agricidaniel-claude-seo` && `/seo setup` — Needs Claude Code 1.0.33+ and Python 3.11 to 3.14. /seo doctor checks readiness without changing anything. Remove: `/plugin uninstall claude-seo@agricidaniel-claude-seo`.
- Claude Code (manual, macOS / Linux): `git clone --depth 1 https://github.com/AgriciDaniel/claude-seo.git` && `bash claude-seo/install.sh` — Copies skills to ~/.claude/skills/seo*, helpers to ~/.claude/agents/seo-*.md and builds ~/.claude/skills/seo/.venv. The schema check after edits only runs with the plugin install. Remove: `bash claude-seo/uninstall.sh (from a fresh clone)`.
- Claude Code (manual, Windows): `git clone --depth 1 https://github.com/AgriciDaniel/claude-seo.git` && `powershell -ExecutionPolicy Bypass -File claude-seo\install.ps1` — Inspect install.ps1 before running it.
- Optional extensions: DataForSEO, Firecrawl, Banana (Gemini images), Ahrefs, SE Ranking, Profound, Bing Webmaster, Matomo and Unlighthouse each have their own install.sh / install.ps1 under extensions/ and need your own account or key.

## What runs on the user's machine
- Code on the machine: yes (plugin and manual installs; Python scripts run when a command needs them)
- Network: yes
- After every Edit or Write in Claude Code (plugin install): A Node launcher starts a Python check on .html, .jsx, .tsx, .vue, .svelte, .php and .ejs files. It reads the file and flags placeholder, deprecated or broken structured data back to Claude. It writes nothing and makes no network calls.
- /seo setup (only when you run it): Creates a Python virtual environment, pip-installs requirements.txt and downloads Playwright Chromium into the plugin data folder (manual install: ~/.claude/skills/seo/.venv).
- During audits: Python scripts fetch and render the pages you name, take screenshots and write reports into a {domain}-audit/ folder in your working directory.
- With your keys only: Scripts call Google (Search Console, PageSpeed, CrUX, GA4, Indexing, Ads, NLP, YouTube), Moz, Bing / IndexNow, Keywords Everywhere, Matomo and DataForSEO. Keys and tokens are kept in ~/.config/claude-seo/ or environment variables.
- Always available: Common Crawl is queried for backlink data without a key; /seo flow sync downloads prompts from api.github.com.
- /seo drift: Stores snapshots in ~/.cache/claude-seo/drift/baselines.db.

## Settings
- `CLAUDE_SEO_PYTHON`: Path to a Python 3.11 to 3.14 interpreter to use
- `CLAUDE_SEO_DATA_DIR`: Where /seo setup puts its Python environment and browser
- `CLAUDE_SEO_LOCAL_TARGETS`: Allowlist of private addresses (for a self-hosted Matomo)
- `CLAUDE_SEO_TAG`: Release tag the manual installer clones (default v2.5.0)
- `~/.config/claude-seo/google-api.json`: Google API key, service account path, default Search Console property and GA4 property

## Examples (from the pack's own files)
- The start of a schema audit report: output format defined in README.md
- A single-page score card: output format defined in skills/seo-page/SKILL.md

## Author's numbers (as reported, not run by AgentGid)
- A full audit takes 10 to 15 minutes. Source: AgriciDaniel/claude-seo/README.md
- The project's own site got 10.8K Search Console clicks in its first 12 weeks and a 99 mobile PageSpeed score, audited with Claude SEO from day one (The author's own site; screenshot of Search Console and PageSpeed in the README). Source: AgriciDaniel/claude-seo/README.md
- Up to 17 agents work on the site at the same time. Source: AgriciDaniel/claude-seo/README.md

## Steps only the person can do
- Create a Google Cloud project with an API key and, for Search Console and GA4, a service account (optional) (skills/seo-google/SKILL.md (/seo google setup)) Then: The assistant runs /seo google commands with real Search Console, PageSpeed, CrUX and GA4 data
- Add the service account's email as a user in Search Console settings (Google Search Console > Settings > Users) Then: Search Console queries and URL inspection stop failing with an access error
- Open a DataForSEO account and enter its username and password in the extension installer (optional, paid per call) (https://app.dataforseo.com/register) Then: The assistant can use /seo dataforseo, and shows a cost estimate before expensive calls
- Decide whether AI companies may train on the site before robots.txt signals are drafted (/seo agentic fix robots) Then: The assistant drafts the matching signal for review
- Confirm you own the site before the technical audit requests paths beyond the audited URL (/seo technical) Then: The assistant checks the server stack for known security holes, read-only

## Mistakes to avoid
- Not: Read the scores as Google's own ranking data. Instead: Treat them as the pack's estimates and say so in reports. Why: The skill marks its scores as heuristics, not Google signals
- Not: Expect real Core Web Vitals and indexing status with no keys. Instead: Connect the free Google APIs with /seo google setup when you need field data. Why: Without keys the speed numbers are lab estimates and indexing is guessed from the page
- Not: Use /seo google index to push any page into Google. Instead: Use it only for job posting and live video pages, and tell the person about the limit. Why: The Indexing API is officially limited to those page types, with 200 requests a day
- Not: Run /seo audit for every small question. Instead: Use /seo page or a single sub-command when one page or one area is enough. Why: A full audit runs several Opus helpers and uses more of the Claude plan
- Not: Install manually and expect the structured-data check after each edit. Instead: Install as a Claude Code plugin if you want that check. Why: Manual installs copy the hook files but do not switch the hook on

## What leaves the machine
- Requests for the pages you audit → The site you name, Every audit or page check. Opt out: Only audit sites you mean to check
- URLs, your Search Console and GA4 queries, keywords and page text for language analysis → Google APIs (PageSpeed, Search Console, CrUX, GA4, Indexing, Ads, Natural Language, YouTube), Only after you add Google credentials and run /seo google commands. Opt out: Do not run /seo google setup
- Domains you analyze → Common Crawl (public dataset query); Moz, Bing and Keywords Everywhere when their keys are set, Backlink checks. Opt out: Skip /seo backlinks; leave the paid backlink keys unset
- URLs, domains, keywords, brands or image prompts → DataForSEO, Firecrawl, Gemini, Ahrefs, SE Ranking, Profound, Bing Webmaster / IndexNow or your Matomo, Only when you install that extension and run its commands. Opt out: Do not install the extensions
- A request for the public FLOW prompt files (no data about your sites) → api.github.com, When you run a Flow sync. Opt out: Do not run the Flow sync

## Lines in the pack aimed at agents (ask the person first)
- "Offer a local run: `npx lighthouse@latest <url> --only-categories=agentic-browsing --output=json`, then `--from-json`." (skills/seo-agentic/SKILL.md): Ask the person first and pin a Lighthouse version instead of @latest
- "curl -fsSL https://raw.githubusercontent.com/AgriciDaniel/claude-seo/main/install.sh | bash" (docs/INSTALLATION.md): Do not pipe the script to a shell; clone, read install.sh, then run it, or use the plugin install
- "Prompt for your DataForSEO username and password" (extensions/dataforseo/README.md): The person types the credentials into the installer; the assistant never asks for them in chat
- "After completing any **major deliverable**, append this footer as the very last output:" (skills/seo/SKILL.md): The footer is promotional links to the author's community; drop it from reports meant for clients if the person prefers

## May I use it like this?
- commercial-app: yes (MIT: keep the license notice)
- client-work: yes (MIT allows audits and reports for clients)
- redistribute-files: yes (MIT: keep the copyright and permission notice in copies)

## Which one to pick
- If You want full site audits, a scored action plan and about 30 SEO commands in Claude Code: skill:AgriciDaniel/claude-seo
- If You mainly care about being cited in AI search answers: skill:zubair-trabzada/geo-seo-claude
- If You want your agent to work from your real Search Console, Analytics and Bing data: skill:iannuttall/seo
- If You only need page speed and Core Web Vitals fixes based on Lighthouse: skill:addyosmani/web-quality-skills

## Works well with
- agent:claude-code: It ships as a Claude Code plugin with slash commands, helper agents and a hook
- tool:playwright: /seo setup installs Playwright and its Chromium to render JavaScript pages and take screenshots
- tool:lighthouse: /seo agentic reads the Lighthouse Agentic Browsing category
- api:firecrawl: An optional extension adds Firecrawl crawling and scraping

## Works in and needs
- Works in Claude Code: plugin
- Works in Cursor: rules folder (copy skills and agents into .cursor/rules)
- Works in Codex CLI, Gemini CLI, Cline, Aider: AGENTS.md in the repo root
- Needs Claude Code 1.0.33 or later and Python 3.11 to 3.14: The plugin and its Python scripts
- Needs Headless Chromium (Playwright): Renders JavaScript pages and takes screenshots; installed by /seo setup
- Needs Node.js: The plugin's after-edit hook is started with node
- Needs Google API key or service account (optional: Real field data from /seo google
- Needs DataForSEO account (optional: /seo dataforseo live search data

## How to tell it works
After install and /seo setup, run /seo page on one page of your own site. You get a score card with five part scores and a list of issues from Critical to Low. If /seo alone prints the list of commands, the plugin is loaded.

## Safety
Install from github.com/AgriciDaniel/claude-seo. The author advises git clone and reading the installer instead of piping a script from the web. The marketplace also lists a second plugin, seo-cockpit, fetched from another branch and pinned commit; it was not part of this check. The technical audit's security check is read-only and asks whether you own the site before probing other paths.
