Claude SEO skills and plugins, tested on a real site
Claude SEO usually means doing SEO work with Claude, Anthropic’s AI model, through plugins and skills that add audits, data and repeatable workflows. We tested four of them on our own site.
claude-seo found the most problems, 9 of 11, including content errors no other tool caught. Its audit cost about 11 times as much as SearchFit SEO’s, which found 6. SE Ranking’s skills found 4 and bring their own keyword data, and InfraNodus’ skill plans what to write. Each of the three audit tools also got something wrong, so an editor still checks the results.
Can Claude do SEO?
Yes, Claude can do SEO work, but what it can do depends on where you run it and what you connect to it. In the Claude app it drafts, rewrites and analyzes whatever you paste in. In Claude Code it works on files and runs commands on your computer. That means it can read a site’s pages, check its sitemap or run a script over a keyword export.
What Claude doesn’t have on its own is SEO data. It can’t see your rankings, search volumes or backlinks until you connect a data source.
Claude SEO can mean three things. Most often it’s doing SEO work with Claude through plugins, skills and connected data, one part of AI SEO for content teams. It’s also the name of one open-source plugin, claude-seo. And some people use it for getting your content cited in Claude’s own answers, the Claude side of LLM SEO.
What are the Claude SEO skills and plugins?
Claude SEO skills and plugins are add-ons that teach Claude an SEO task or connect it to SEO data. Four terms come up:
- Skill: a
SKILL.mdfile of instructions that Claude follows when a task matches it, or when you run it by name, such as/seo-check. - Agent: a helper Claude hands part of a job to, with its own instructions. claude-seo, for example, sends up to 17 agents to check different parts of a site at the same time.
- MCP server: a connection that lets Claude pull live data from another tool, such as rankings or keyword volumes from an SEO platform.
- Plugin: one install that bundles skills, and often agents and MCP servers too.
Agents matter because they run only in Claude Code and Cowork, not in a chat in the Claude app. A plugin built on agents, like claude-seo, needs Claude Code or Cowork.
We tested four. Two are audit plugins that work with what Claude can fetch itself: claude-seo and SearchFit SEO. The other two, SE Ranking’s and InfraNodus’ skills, work from data their makers collect.
| Tool | What it does | Price | Works in |
|---|---|---|---|
| claude-seo, by AgriciDaniel | Audits a site or page: technical, content, schema, AI search | Free, open source | Claude Code |
| SearchFit SEO, by SearchFit.ai | Audits, page checks, schema, keyword clusters, content plans | Free | Claude Code (the only install its listing offers) |
| SE Ranking’s SEO skills | Briefs, audits, backlink gaps and AI search reports from SE Ranking’s data | Skills free; data needs an SE Ranking account | Claude app, Claude Code |
| InfraNodus’ SEO skill | Finds what people search for that ranking pages don’t cover | Skill free; live data needs a free account | Claude app, Claude Code, Cursor, Codex |
As of November 20, 2026, claude-seo had 29 sub-skills and 20 agents, SearchFit SEO had 11 skills and 3 agents, and SE Ranking’s GitHub repo listed 32 skills. InfraNodus’ skill is a single skill that works through InfraNodus’ own MCP server.
Which Claude SEO skill is best? Our test results
For auditing a site, claude-seo was the best of the four we tested. It found 9 of the 11 problems we knew about on our own site, including four that no other tool found. SearchFit SEO found 6 for about a tenth of the cost, and SE Ranking’s skills found 4. InfraNodus’ skill answers a different question, so we didn’t score it against the list.
We ran each tool on contentrunbook.com on November 20, 2026, in Claude Code on a Mac. The three audit tools got the same jobs: audit the site, check one page (our GEO guide), and check its readiness for AI search. InfraNodus’ skill got one topic and the same page.
We graded every finding against 11 problems we confirmed on the site by hand. We had found 3 of them before the test; the other 8 came from the tools’ reports, and we checked each one on the live site. So the list can’t include problems that no tool and none of us noticed.
| Tool | Problems found (of 11) | Wrong or misleading findings | Setup | Full audit time | Full audit cost |
|---|---|---|---|---|---|
| claude-seo | 9 | 2 | 5 min (installs Python tools and a browser) | 4.2 min | $2.93 |
| SearchFit SEO | 6 | 3 | Under 10 seconds | 1.5 min | $0.26 |
| SE Ranking’s SEO skills | 4 | 1 | Trial account and a sign-in (failed once) | 1.1 min | $0.31, plus SE Ranking credits |
| InfraNodus’ SEO skill | Not an audit tool | 0 (it flagged its own off-topic results) | Under 10 seconds; live data needs a free account | n/a | $0.19 for one page |
Costs are what each run would cost through Anthropic’s API, as Claude Code reports them. On a Claude subscription, they come out of your usage limit instead of a bill.
The answer key. These were the 11 problems on our site that we confirmed by hand:
- The homepage still showed two figures from before a correction to our AI citation study.
- A data download disagreed with the study’s corrections log.
- The
wwwaddress didn’t resolve. - The site sent no HSTS header.
- Article structured data had no publish date or image.
- Screenshots had no width and height.
- 10 of 18 sitemap entries had no last-modified date.
- One title ran to 68 characters.
- Our main GEO page didn’t link to three related guides.
- Two pages gave different advice on how often to repeat a check.
- One template page put the template seventh of eight sections.
Results also changed when we re-ran SearchFit on a larger model. It found 6 problems again, but not the same 6: it caught the missing internal links and missed the long title. We didn’t repeat runs on the same model, so we can’t say how much of that is the model and how much is chance. Either way, don’t treat one run as complete.
Every tool that ran also told us to add a named author and FAQ markup. A named author is a choice for each publication; we don’t use one. And Google says no special schema markup is needed to appear in its AI features.
The main runs used Claude Sonnet 5.5, Claude Code’s default on our Mac. We also re-ran SearchFit and InfraNodus on Claude Opus 5.5.
claude-seo
claude-seo found 9 of the 11 problems in our answer key. It was the only tool to catch the three content errors: the stale homepage figures, the data download and the conflicting advice.
- claude-seo pros: covers technical SEO, content, schema and AI search in one run; free and open source.
- claude-seo cons: the most expensive audit, about 11 times SearchFit’s; a 5-minute setup; it missed the image sizes and the long title, and its AI search check said the site has no images after reading 5 of 18 pages.
- claude-seo price: free; $2.93 in Claude usage for the full audit.
- Our take on claude-seo: use it for a monthly or pre-launch audit, if you have Claude Code and the usage to spare.
SearchFit SEO
SearchFit SEO found 6 of the 11 problems, including the image sizes and the long title that claude-seo missed.
- SearchFit SEO pros: fast and cheap: 1.5 minutes and $0.26 for the audit, 18 seconds for its
/seo-checkon one page; no setup. - SearchFit SEO cons: it missed all three content errors and got three things wrong: it put our GEO page’s meta description at 160 characters (it’s 155), told us to publish study data we already publish, and said no page was orphaned. It isn’t in Claude Code’s official plugin marketplace, so you install it from its own GitHub repository.
- SearchFit SEO price: free; $0.26 in Claude usage per audit.
- Our take on SearchFit SEO: a quick technical check before you publish a page.
SE Ranking’s SEO skills
SE Ranking’s SEO skills found 4 of the 11 problems, working from SE Ranking’s own crawl of the site.
- SE Ranking’s SEO skills pros: they bring rankings, keyword and backlink data that Claude doesn’t have, and they work in the Claude app as well as Claude Code.
- SE Ranking’s SEO skills cons: on our two-day-old site, the page and AI search checks came back with “insufficient data.” Setup took two tries: the first sign-in failed with “Failed to fetch API keys (404)”, and the account page showed 0 credits before the skills reported the trial’s 100,000.
- SE Ranking’s SEO skills price: the skills are free; data uses SE Ranking credits, with a 14-day trial that includes 100K; after the trial, data needs a paid SE Ranking plan; $0.31 in Claude usage per audit.
- Our take on SE Ranking’s SEO skills: worth it on an established site that has rankings to analyze.
InfraNodus’ SEO skill
InfraNodus’ SEO skill finds topics people search for that ranking pages don’t cover. On “AI content operations for content teams” it listed five gaps.
Three were topics already on our own plan: what AI content operations is, who owns what in an AI content team, and AI usage rules for content teams. The other two were about digital asset management, which isn’t a problem our readers have. On our GEO page it drifted off topic and said so in its report.
- InfraNodus’ SEO skill pros: it says plainly when it lacks live data or goes off topic; $0.17 to $0.19 in Claude usage per run; it also works outside Claude.
- InfraNodus’ SEO skill cons: no search volumes, so you still need a keyword tool; without an account it falls back to web search.
- InfraNodus’ SEO skill price: the skill is free; live data needs a free InfraNodus account.
- Our take on InfraNodus’ SEO skill: use it on a topic when you plan content, next to a keyword tool.
How we run SEO on Claude skills
We run this site’s SEO and content work on four skills of our own in Claude Code. Each skill is a SKILL.md file in the site’s repository, so every page goes through the same steps:
- seo-brief: pulls keyword volumes from DataForSEO, reads the top results in real Google through a browser, and writes a brief and outline. An editor approves the outline before any drafting.
- write: builds a fact ledger first, with every number and quote and its source, then drafts only from that ledger. Anything it can’t source gets a marker for the editor to check.
- qa: checks the draft against our writing and SEO rules, opens every cited source, and recomputes our own figures from the raw files. The judgment pass runs in a separate agent that didn’t write the draft.
- package: takes an approved page live, then checks the live page and submits it to Google, Bing and IndexNow.
Tracking AI citations isn’t a skill yet. Each round, we put the same 25 questions to every AI tool we track and log which sites each answer cites. Scripts collect most answers; Google’s AI Overviews are read in a browser.
On the four articles with complete logs, as of November 20, 2026, Claude’s time and the keyword data cost little next to the editor’s review:
| Article | Claude’s time (min) | Editor review (min) | Keyword data cost |
|---|---|---|---|
| How to track AI citations | 23.7 | not logged | $0.012 |
| AI search citation study | 27.5 | 62 | not logged separately |
| LLM SEO | 40.2 | not logged | $0.029 |
| GEO vs SEO | 50.9 | 90 | $0.059 |
Claude’s work on an article took between 24 and 51 minutes, and the keyword data for a brief cost between 1 and 6 cents. The editor’s review took longer than Claude’s work on both articles where we logged it. The brief was the longest Claude stage on three of the four.
Claude’s time is wall-clock time while it works, logged by a script at the start and end of each stage. The editor’s minutes are what the editor reported after each review round. On two of the four articles the editor’s time wasn’t logged, so we leave it out rather than estimate it.
An editor still reviews every draft in comment rounds and approves every page before it goes live. On the last article we published, that review was one round of 15 minutes.
How to install a Claude SEO skill
You install a Claude SEO skill in one of two places: the Claude app, from Customize > Plugins, or Claude Code, with the /plugin command. Data comes through an MCP server, which you connect separately.
Before you start, you need a paid Claude plan for plugins from the directory. For Claude Code, you need it installed on your computer. claude-seo also needs Claude Code 1.0.33 or newer and Python 3.11 to 3.14. A plugin can run code on your computer with your user permissions, so read what it installs first.
Step 1: Pick the app or Claude Code
Choose Claude Code if the tool relies on agents or local scripts; claude-seo and SearchFit SEO both do. Agents also run in Cowork. Choose the Claude app if you want skills and an MCP server without a terminal; SE Ranking’s and InfraNodus’ skills both work there.
Step 2: Install the plugin or skill
In the Claude app, open Customize > Plugins, select Add > Add marketplace, and enter the plugin’s GitHub repository, such as seranking/seo-skills. Install the plugin once its marketplace loads. It appears in your skills list, and it also syncs to Claude Code when you sign in with the same account.
In Claude Code, add the marketplace, then install:
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo
The install command opens the plugin’s details and asks for a scope. Local scope turns the plugin on for you in the current folder only. That keeps a test away from your other projects. When it works, typing / lists the plugin’s commands.
Some plugins need one more setup command. For claude-seo it’s /seo setup, which installs its own Python tools and a headless browser. It took 5 minutes in our test, and /seo doctor then reports whether everything is ready.
Step 3: Connect your data
Skills that use a data provider need a sign-in or an API key; without one, some fall back to weaker data without asking. InfraNodus’ skill, for example, ran on web search in our test.
For SE Ranking, type /mcp in Claude Code, pick the SE Ranking server and sign in. For InfraNodus, add https://mcp.infranodus.com as a connector in the Claude app under Customize > Connectors. Then sign in with the API key from a free InfraNodus account. claude-seo connects free Google APIs, including Search Console, with /seo google setup.
In Claude Code, /mcp lists every MCP server with a check mark once it’s connected.
Step 4: Run it on one page first
Run the tool on one page you know well before a whole site, such as /seo page https://yoursite.com/pricing in claude-seo. Compare its findings with what you know is true about that page: wrong findings on a page you know mean wrong findings on pages you don’t. Then run the full audit.
Which Claude model is best for SEO?
Claude Opus 5.5 is the best Claude model for SEO judgment work, and Sonnet 5.5 is enough for checklist audits. We run this site’s own skills on Opus 5.5.
For plugins, our test shows the trade-off. SearchFit’s audit on Opus 5.5 cost $1.20 against $0.26 on Sonnet 5.5 and took 5.2 minutes instead of 1.5. It found the same number of problems, so for a checklist audit the smaller model was enough.
The SEO plugins themselves mix models by task. claude-seo runs its technical, schema and sitemap agents on Sonnet, and its content, AI search and topic agents on Opus. We’d follow the same split. Use the larger model for judgment work such as briefs, drafts and content reviews. Use a faster model for checks that follow a fixed list.
Price is the other factor when you call Claude through the API. In our citation pilot, the same question with web search cost $0.051 on Claude Sonnet 5.5 and $0.013 on Claude Haiku 5.5, about a quarter. That pilot measured cost, not answer quality, so it doesn’t tell you which model writes the better brief.
What Claude can’t do for SEO
Claude can’t replace an SEO data tool, a fact-check or an editor. These are the limits we ran into:
- No live search data on its own. Claude doesn’t know your rankings, keyword volumes or backlinks.
- The data tools can be wrong too. On November 16, 2026, we found that DataForSEO’s Google results didn’t match what US searchers saw, because its crawler got a different page. When we re-collected the results in a browser, the share of sites cited in AI Overviews that also rank on Google’s first page went from 62% to 78%. We now read Google’s results pages in a browser and use DataForSEO only for keyword volumes. The citation study explains the change.
- Plugins cost tokens even when idle. An enabled plugin is part of every session, and each skill’s name and description count toward your usage on every turn. Turn off plugins you’re not using.
- Plugins run as you. A plugin can run code with your user permissions. Anthropic reviews plugins listed in its directory, but not ones you add from a GitHub repository.
- Drafts need an editor. On our first two articles, QA found blocking problems in both drafts, and neither was ready to publish until the editor’s review.
Is Claude better than ChatGPT for SEO?
Neither Claude nor ChatGPT is better for SEO across the board. For the tools in this article, the difference is where they run.
The SEO plugins in our test are built for Claude Code and the Claude app. claude-seo needs Claude Code, and SearchFit SEO’s listing installs only there. InfraNodus’ skill also runs in Cursor and Codex, so its method isn’t tied to Claude.
As sources of AI answers, they behave differently too. In our AI citation study round of November 19, 2026, ChatGPT searched the web in 62% of its answers and Claude Sonnet 5.5 through the API in 46%. Claude shared only 5–8% of its cited sites with each other AI tool.
That’s one round for Claude, so treat it as a first reading, not a trend. Getting cited by either one is a separate job, covered in our guide to LLM SEO.