The SEO Plugin That Checks Whether AI Can Cite You, Not Just Whether Google Ranks You

A plugin just passed 25,000 stars on GitHub by doing SEO audits inside Claude Code. That sounds like a niche, and it is, but the pitch on the README is a sentence worth reading twice: rank in Google, get cited by AI.
That second clause is the actual product. Claude SEO, by Agrici Daniel, ships 26 sub-skills and 19 specialist agents, runs a full audit in 10 to 15 minutes, and is MIT licensed. It hit about 5,000 stars in the past week, which puts it fourth on the weekly trending list behind much bigger names.
What a run looks like
You install it in three commands inside Claude Code:
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo
/seo setup
You need Claude Code 1.0.33 or newer. /seo setup builds an isolated Python runtime and a headless Chromium without touching the rest of your system, and /seo doctor tells you whether the environment is ready. There are manual install paths for macOS and Linux if you would rather review a shell script before running it, which I would.
Then:
/seo audit https://example.com
Up to 17 agents fan out across the site at once. Technical SEO covers crawlability, indexing, sitemaps, robots.txt, redirects, page speed and Core Web Vitals. Content covers E-E-A-T, thin pages and topic coverage. There is a schema agent that detects, validates and writes Schema.org markup, and a sitemap architect, and a visual QA agent, and a backlink analyst pulling from Moz, Bing and Common Crawl.

The two commands that are not about Google
/seo geo estimates how likely Google AI Overviews, AI Mode, ChatGPT, Perplexity and Gemini are to cite your pages. Fair warning: nobody outside those companies can measure this precisely, and any tool claiming otherwise is modelling a proxy. But the proxy is at least the right shape — it looks at whether your content is structured in extractable chunks, whether claims are attributable, whether there is machine-readable context.
/seo agentic is the more unusual one. It scores agent readiness: Lighthouse's Agentic Browsing score, whether AI agents can access the site at all, whether there is an llms.txt, whether pages have Markdown equivalents, whether WebMCP is present.
This is a real shift and it is worth naming. For twenty years, SEO meant optimizing for a crawler that indexed pages so a human could click them. Now a growing share of traffic is an agent fetching a page so it can answer a question, and those are different requirements. A page that renders beautifully behind three JavaScript bundles and a cookie wall is fine for a human and useless for an agent.
The drift monitor is the sleeper feature
/seo drift baseline snapshots your site into SQLite; compare and history show what changed since. Agencies do this manually with spreadsheets. Having it as a command that writes to a local database is unglamorous and genuinely useful, because most SEO damage is not a collapse, it is a template change that quietly dropped 400 pages of structured data six weeks ago.
There is also /seo plan with presets for SaaS, local, ecommerce, publisher and agency, a /seo cluster command that does SERP-based semantic clustering, and /seo sxo for search experience work — page type, personas, user stories. Extensions cover Search Console, PageSpeed, CrUX, GA4, IndexNow submission and a local multi-page Lighthouse runner.
The multilingual corner is stronger than it looks
One command I did not expect to matter: /seo hreflang. It audits and generates hreflang annotations for international sites, which is the single most commonly broken piece of technical SEO on any site that publishes in more than one language.
If you run a blog in eleven locales, you know the failure mode. Somebody adds a twelfth language, the canonical tags point at the English version, the sitemap lists the same URL four times with different hreflang values, and Google picks one at random. You will not notice for months, because the pages still load.
Having an agent walk the alternates and cross-check them against actual page existence is exactly the kind of boring, exhaustive, error-prone work that language models are good at, and exactly the kind of work no human wants to do on a Friday. It is a small feature that quietly justifies the install for anyone running a multilingual site — including, I would guess, most of the people reading this.
The results section, read carefully
The README's proof point is the project's own site: 10.8K Search Console clicks and 324K impressions in its first 12 weeks, with a 99 mobile PageSpeed score and 100 for SEO.
Those are real numbers and I do not doubt them. They are also the numbers of a small, fast, freshly built documentation-style site with a single author who cares about this stuff. A 99 on mobile PageSpeed for that site tells you the plugin can identify and fix the obvious things. It does not tell you it will move a 200,000-page ecommerce catalogue, where the problems are template debt, duplicate faceted URLs and a CMS nobody can deploy on a Friday.
Which is the general trap with audit tools. They are very good at the class of problem that can be detected from the outside, and most of the money is in the class of problem that cannot.
What it costs you beyond the licence
Nineteen agents, several of them on Opus. A full audit with 17 running in parallel is a token event, and the README's own timing — 10 to 15 minutes — describes a run that is doing serious work. If you are an agency planning weekly audits for every client, do the arithmetic before you promise that to anyone.
There is also a subtler cost: report volume. Every agent produces findings, every finding gets a priority, and a 30-command plugin can generate an action plan longer than anyone will read. The value is entirely in the ranking, and ranking is the part an LLM is least reliable at. Read the top five items. Treat the rest as a backlog you will never clear, because you will not.
Should you install it
If you run a content site, a docs site, or a SaaS marketing site and you already pay for Claude Code, this is the cheapest competent second opinion you will get this month. Run /seo audit once, read the top five findings, fix them, and then stop. Run it again in a month to see what drifted.
If you run a large site with a real engineering team, use /seo agentic and /seo geo as instrumentation and ignore the rest. The agent-readiness score is the part of this that is genuinely new, and it is the part most teams have not thought about at all.
The bigger point is what the star count says. A year ago an SEO plugin would have been a scraper with a dashboard. This one treats "can an AI agent read and cite this page" as a first-class audit category, and 25,000 developers starred it in a matter of weeks. That is not enthusiasm for SEO. That is a lot of people noticing that their audience changed.
Related articles
Chinese text in video always turns to mush. Dreamina handed the typesetting to code
Dreamina launched Motion Studio on November 10. It no longer generates frames; the model writes code that lays out text, shapes and pacing. Chinese titles stop blurring, but whether the content is right is still on you.
A community spreadsheet of taps that turns off the ads inside Chinese Android builds
A no-root tool that switches off the ads Chinese phone makers ship inside the system UI. Its method is a hand-written tap script per phone model, maintained by volunteers every time a ROM moves. It works, and the maintenance is the point.
GPT Image 2.6 held the token price and raised the bill
GPT Image 2.6 kept every token rate from 2.5 and still costs more per image, because more of the work now happens in output tokens. Eleven days into the ramp, the pricing structure says more than the leaderboard does.
Someone rebuilt IDM from the outside and wrote down what they did not touch
A Rust reimplementation of Internet Download Manager claims 5x the speed and 86 languages. The more interesting document is the paragraph explaining how it was built without copying anything.