Most Claude SEO advice is a list of 50 prompts. Prompts are the least important part. The time you save using Claude for SEO comes from three things you set up once: what Claude knows about the site, how it should do a repeat task, and which live data it can reach. Get those wrong and the cleverest prompt still returns confident guesses about search volume. This guide covers that setup, five workflows that hold up in practice, and the specific ways Claude goes wrong on SEO tasks.
Scope note: this covers the Claude chat app on web, desktop and mobile. Terminal work like crawl scripts and redirect maps belongs to Claude Code for SEO.
Where Claude Actually Saves Time (and Where It Doesn’t)
Claude is strong at the work between the data and the decision: reading ten competitor pages and finding the angle none of them cover, turning a 5,000-row export into a ranked shortlist, rewriting 40 title tags to a set of rules, and drafting the client email that explains why traffic dipped. That is analyst and editor time.
It is weak at anything that needs numbers it doesn’t have. Out of the box, Claude has no search volumes, no keyword difficulty, no backlink counts and no rankings. Ask for them anyway and you may get plausible figures with no source behind them. A sound rule for Claude SEO work: Claude can reason about data you give it or connect to it, but it should never be the source of the numbers.
That is also why tools like SEO Rocket exist. It runs keyword research on real Ahrefs data, competitor gap analysis and a real-crawler technical audit inside one chat, so the reasoning sits on top of measured numbers instead of guesses.
The Three Building Blocks: Projects, Skills and Connectors
Anthropic’s own explainer on Skills draws the line cleanly: Projects tell Claude “here’s what you need to know,” Skills tell it “here’s how to do things,” and MCP connects Claude to the systems where data lives. For SEO that maps to a simple decision rule:
- Is it background that every chat about this site needs? Brand voice, target audience, service list, past audit findings. Put it in a Project.
- Is it a procedure you repeat the same way each time? Your brief template, your title-tag rules, your audit checklist. Package it as a Skill.
- Is it data that changes? Rankings, clicks, backlinks, crawl results. Connect it through an MCP connector, or upload a fresh export.
Most frustration with Claude for SEO comes from mixing these up, like pasting a stale keyword export into a Project.
Set Up One Claude Project per Site or Client
According to Anthropic’s help centre, Claude Projects are “self-contained workspaces with their own chat histories and knowledge bases,” available on every plan, with Free accounts capped at five. Each Project gets its own instructions and its own uploaded knowledge. For an agency or a freelancer, one Project per client is the setup that pays off fastest.
What to put in the knowledge base:
- A one-page site profile: what the business sells, who buys, the money pages, and the locations served.
- Brand voice rules with two or three example paragraphs you’d actually publish.
- Your last audit summary and the list of known issues already being fixed, so Claude stops “discovering” them.
- A page inventory (URL, page type, primary keyword), ideally pulled from the sitemap rather than typed by hand.
Two details matter more than they look. First, memory: Anthropic’s memory documentation says each Project has its own memory but also draws on your memory from chats outside Projects unless you switch off “Use account memory” on the Project page. If you handle competing clients, switch it off so one client’s details never leak into another’s work. Second, size: on paid plans, when Project knowledge nears the context limit, Claude switches to retrieval (RAG) mode and searches your files instead of reading all of them. The help centre advises clear, descriptive filenames, so name files “acme-brand-voice.md” rather than “notes-final-v3.docx.”
Sharing a Project with colleagues is limited to Team and Enterprise plans. Anthropic is also rolling out a new beta version of Projects, starting in Claude Code, so expect the interface to shift.
Workflow 1: Content Briefs From Live SERP Research
This is the workflow that saves the most time for content teams. Inside the client Project, turn on web search (or Research, which Anthropic’s help centre lists for paid plans and which needs web search on to work) and give Claude the target query plus your brief template.
A prompt that works: “Search for [query]. Read the top five organic results. For each, list the H2s, the search intent it serves, and one thing it misses. Then propose an outline that covers the shared ground in fewer words and adds two sections none of them have.” The “one thing it misses” instruction is the important part. It forces information gain rather than a summary of what already ranks.
Two checks. Claude’s web search is not a rank checker, so its “top five” may not match Google in your market; paste the actual URLs when that matters. And verify any statistic Claude pulls from a competitor page at its original source.
Workflow 2: Turning Search Console Exports Into Decisions
Claude can run code on files you upload. Anthropic’s help centre says the “create and edit files” feature handles CSV and spreadsheet analysis, writes Python for it, and produces charts, with a 30 MB per-file limit. That turns a Search Console performance export into a ten-minute job.
Upload the queries-and-pages export and ask for three lists: queries ranking in positions 8 to 20 with impressions above a threshold you set (striking distance), pages where two or more URLs share the same top query (possible cannibalisation), and pages whose clicks fell between two date ranges you exported. Ask Claude to show the code it ran. Computed numbers are reproducible, and you can check the filter logic.
Judgment stays with you: Claude flags a cannibalisation pair, but only you know if the two pages target different funnel stages.
Workflow 3: Titles, Meta Descriptions and Rewrites at Scale
Language models are unreliable at counting characters in their heads, and a title or meta description that runs long gets cut off in results. The fix is to make Claude check its own output with code. Ask it to write the 40 titles, then run a character count on each, and rewrite any outside your limits before returning the table.
The same pattern works for refreshes. Give Claude the current page, the Search Console queries it already earns impressions for, and your voice rules from the Project, then ask for a rewrite that answers those queries more directly without dropping the sections that already rank.
Workflow 4: Package Repeat Work as Skills
If you’ve pasted the same brief template or audit checklist into Claude five times, it should be a Skill. Per Anthropic’s help centre, Skills are folders of instructions, scripts and resources that Claude loads when a task calls for them. They’re available on Free, Pro, Max, Ultra, Team and Enterprise plans, and they require code execution to be switched on. You can upload your own as a ZIP file under Customize, then Skills.
Good SEO candidates are narrow and repeatable: a content-brief Skill with your template and quality checks, a title-tag Skill with your length rules and the counting script, or a “monthly report” Skill that turns an analytics export into your client report format. We go deeper on structure and examples in Claude Skills for SEO.
Workflow 5: Connect Live SEO Data With MCP Connectors
Connectors fix the “Claude has no numbers” problem. They use the Model Context Protocol, and Anthropic’s help centre says custom connectors to remote MCP servers work on every plan, with Free users limited to one. On Pro and Max you add one under Customize, then Connectors, then “Add custom connector,” and paste the server URL. On Team and Enterprise an owner adds it in organisation settings.
Ahrefs, Semrush and community Search Console servers are the obvious choices for live keyword, backlink and click data; our roundup of SEO MCP servers compares them, and there is a step-by-step guide for connecting Search Console. If you want a free starting point, SEO Rocket runs a public MCP server at app.seorocket.ai/mcp. It needs no account and is read-only, and it gives Claude four tools: search our SEO guides, find a free SEO tool, define an SEO term, and map any site’s structure from its XML sitemaps. It doesn’t touch your projects or rankings.
Anthropic’s help centre also carries a warning worth repeating: only connect servers from organisations you trust, because a malicious server can hide instructions aimed at making Claude act in ways you didn’t intend.
Claude vs ChatGPT for SEO: What Actually Differs
The honest answer is that the gap is smaller than either fan base claims. Both products now have project workspaces with files, instructions and project-scoped memory; OpenAI documents a “project-only” memory option for new projects that works much like Claude’s account-memory switch. Both can analyse uploaded spreadsheets and search the web.
The deciding factors for most SEO teams are practical: which assistant your clients and colleagues already use, which data tools offer a connector for it, and whether you want reusable Skills in the same place you chat. Skills and custom connectors on every plan are the strongest workflow arguments for Claude. We compare the two head to head in Claude vs ChatGPT for SEO.
Where Claude Gets SEO Wrong, and How to Guard Against It
- Invented metrics. Without a data connector, any search volume or difficulty score is a guess. Ban them in your Project instructions: “Never state search volume, difficulty or rankings unless they come from an attached file or connector.”
- Outdated guidance. Training data lags, and SEO changes fast. For anything about Google features or algorithm updates, make Claude search and cite the current Google Search Central page.
- Over-confident audits. Claude reading a page’s HTML is not a crawl. It won’t see rendering problems, redirect chains or orphan pages across a whole site. Use a real crawler and let Claude interpret the output; see how to run an SEO audit with Claude.
- Generic content. No brief means the average of the internet. Feed it your angle and first-hand detail, then edit.
The flip side of Claude SEO is being found by Claude. Anthropic’s help centre lists three crawlers: ClaudeBot for training, Claude-User for fetches a user triggers, and Claude-SearchBot for search indexing. It notes that blocking the last two can reduce your visibility in Claude’s answers. Check your robots.txt doesn’t block them by accident, then track whether you’re actually cited, which is what SEO Rocket’s AI-visibility tracking does.
Claude SEO FAQ
Can Claude do keyword research?
It can cluster, map intent and prioritise keywords well, but it has no native search-volume data. Connect an SEO data source or upload an export first. Our guide to keyword research with Claude shows the full workflow.
Do I need a paid plan to use Claude for SEO?
No. Projects (up to five), Skills, file analysis and one custom connector are all on the Free plan, per Anthropic’s help centre. Paid plans add Research, RAG for large Projects and more usage, which matters once you run this daily across several clients.
Is Claude SEO the same as optimising for Claude’s answers?
The phrase covers both. This guide is about using Claude to do SEO work. Visibility inside Claude’s answers depends on crawl access, clear and citable content, and authority, much like other AI search; see our guide to AI agents for SEO for the wider picture.
Is it safe to upload client data to Claude?
Use one Project per client, turn off account memory for those Projects, and check your plan’s data settings with whoever owns your client contracts. Only add connectors from vendors you trust.
Start small this week. Build one client Project, write the “no invented metrics” rule into its instructions, and run Workflow 2 on a fresh Search Console export. Once that saves you an hour, add a Skill and a connector. That order follows a playbook proven across 1,000,000+ ranking pages: process first, tools second, and the numbers always come from real data.
