Claude can do much of the analytical and drafting work in SEO: interpreting data, finding patterns, writing briefs, drafting titles and descriptions, and summarising reports. It needs real data to be reliable, and people still need to set strategy and approve changes.

How do I connect Claude to SEO tools?

Through the Model Context Protocol. Claude Desktop and Claude Code can connect to MCP servers from SEO tools and to community connectors for Search Console and analytics, by adding the server to the app's configuration.

Is Claude good at keyword research?

It is good at expanding and organising keyword ideas and grouping them by intent, but it does not know search volumes or difficulty on its own. Connect a keyword data server or paste data from your tool for numbers you can trust.

Should Claude write full blog posts for clients?

It can draft, but publishing unedited AI drafts is risky for accuracy and quality. Use it for briefs, outlines, research summaries and first drafts that a subject expert edits, and check every factual claim.

Is it safe to give Claude access to client data?

Use a business plan with appropriate data controls, connect only the data the task needs, prefer read-only access, and follow each client's data agreements. Do not paste credentials or personal data into conversations.

Claude for SEO: 12 Workflows Agencies Use Every Week

Most SEO teams have tried asking an AI assistant for keyword ideas or a meta description. Fewer have built it into their weekly work in a way that saves real time. The difference is data and structure: an assistant working from real Search Console, ranking and crawl data, with repeatable prompts, is far more useful than one guessing from memory.

This guide covers twelve workflows agencies use with Claude every week, the data each one needs, example prompts, and where to be careful. The same principles apply to other assistants that support tools.

Setting up: data first

Claude is only as good as the data it works with. There are two ways to give it data.

Paste or upload it. Export a CSV from Search Console, a crawl or a rank tracker and attach it. Simple and safe, but manual.
Connect it through MCP. Add MCP servers for your SEO tools so Claude can fetch data directly. Faster and repeatable, but needs setup and careful permissions. See the complete guide to MCP and the best SEO MCP servers compared.

Start with read-only access. For anything that changes a site, route drafts through an approval step. See human-in-the-loop SEO automation.

Workflow 1: Diagnose a traffic drop

Data: Search Console performance by page and query for two comparable periods; ideally rankings for the same period.

Prompt: "Compare the last 28 days with the previous 28. List the 15 pages that lost the most clicks. For each, tell me whether impressions, CTR or average position changed most, and which queries drove the change."

Why it works: The pattern of the drop points to the cause. Lost impressions suggest ranking or indexing problems; stable position with lower CTR often suggests a new AI Overview or SERP feature. See how much traffic AI Overviews really take.

Check: Verify the top two or three findings directly in Search Console.

Workflow 2: Find keyword cannibalisation

Data: Search Console query and page data.

Prompt: "Find queries where two or more of our pages each get at least 10% of impressions. Group them, show which page Google prefers, and suggest whether to merge, differentiate or re-link."

Check: Cannibalisation is sometimes fine, for example a category page and a product page for different intents. Judge each case.

Workflow 3: Spot quick wins

Data: Search Console queries with position and impressions.

Prompt: "List queries where we rank between positions 5 and 15 with at least 200 impressions a month. Group by page. For each page, suggest what to change to better answer those queries."

Why it works: Pages already close to the top often move with modest improvements to titles, headings and content.

Workflow 4: Build a content brief

Data: Target query, top-ranking pages' headings (from a crawl or SERP tool), "people also ask" questions, and the client's own knowledge.

Prompt: "Build a brief for a page targeting [query]. Include the search intent, the questions the page must answer, a suggested heading structure written as answer-first sections, facts that need a source, internal links to [list], and keywords to avoid because other pages own them."

Why it works: A brief that lists keywords to avoid prevents new content from competing with existing pages. See answer-first content.

Workflow 5: Audit a page

Data: The page's raw HTML, or a crawl record for the URL.

Prompt: "Audit this page for technical and on-page SEO. Check the title, meta description, headings, canonical, robots directives, structured data, internal links and whether the main content is present in the HTML without JavaScript. Prioritise issues by impact."

Check: Confirm rendering issues by fetching the page without JavaScript yourself. See can AI crawlers read JavaScript?.

Workflow 6: Review robots.txt for AI crawlers

Data: The site's robots.txt.

Prompt: "Review this robots.txt. List which AI training crawlers, AI search crawlers and user-triggered agents are allowed or blocked, flag any rule that blocks AI search by mistake, and suggest a corrected version that blocks training but keeps AI search access."

Check: Compare against the vendors' current documentation, as agent names change. See AI crawler user agents.

Workflow 7: Draft titles and meta descriptions

Data: Page URLs, current titles and descriptions, the page content and the target query.

Prompt: "Draft a new title (under 60 characters, brand at the end) and meta description (140–160 characters) for each page. Use only facts that appear on the page. Flag any page where the content is too thin to describe accurately."

Check: Review every draft before it goes live, or review a sample and approve in batches for low-risk pages. See writing meta descriptions with AI at scale.

Workflow 8: Suggest internal links

Data: A list of pages with titles and main topics, and the target page.

Prompt: "Suggest ten existing pages that should link to [target page], with the sentence where the link fits naturally and the anchor text to use."

Why it works: Internal linking is tedious to do by hand across large sites, and the suggestions are easy to check.

Workflow 9: Check AI visibility for a prompt set

Data: The client's prompt set, and AI answers collected from each engine, either pasted in or from a tracking tool.

Prompt: "For each answer, record whether [brand] is mentioned, cited or absent, which competitors appear, and which third-party domains are cited. Summarise mention rate, citation rate and the domains cited most often."

Why it works: Turning dozens of answers into a clear share-of-voice summary is slow by hand. See AI share of voice.

Workflow 10: Map entity inconsistencies

Data: The client's canonical facts and text from its main profiles and listings.

Prompt: "Compare these profiles with the canonical facts. List every inconsistency in name, description, services, locations, hours and contact details, with the profile it appears on."

Why it works: Inconsistent facts weaken how AI assistants describe a brand. See entity SEO for LLMs.

Workflow 11: Draft the monthly report summary

Data: The month's metrics for outcomes, visibility, traffic and work completed.

Prompt: "Write a 150-word executive summary for the client: what happened, why, what needs attention and what we will do next. Lead with business outcomes. Explain any traffic change in plain language."

Check: The account lead edits the final version. See our SEO client report template.

Workflow 12: Prepare a migration redirect map

Data: Lists of old and new URLs with titles.

Prompt: "Match each old URL to the most relevant new URL. Mark matches you are less than confident about, and list old URLs with no good match."

Check: Review every low-confidence match and test the final map on staging. See our website migration SEO checklist.

Prompting tips that make the difference

Give it the data. Answers from real data beat answers from memory every time.
State the output format. Tables, lists with specific columns, or JSON for later processing.
Set constraints. Length limits, facts that must come from the page, terms to avoid.
Ask it to flag uncertainty. "Mark anything you are unsure about" surfaces weak spots.
Save prompts that work. Turn them into templates the whole team uses, or MCP prompts so they run the same way every time.
Verify the important parts. Check the top findings against the source before acting.

Where to be careful

Search volumes and difficulty need real keyword data; Claude does not know them on its own.
Facts about the client must come from the client, not from the model.
Client data should only be shared under appropriate plans and agreements.
Changes to sites should always go through review and approval.

A worked example

On a Monday morning, a specialist asks Claude, connected to Search Console and a crawl tool for one client, to run Workflow 1. It finds that a services page lost 40% of clicks while its position held, on queries that now show AI Overviews. The specialist asks for Workflow 5 on the page, which finds that the pricing section is loaded with JavaScript and invisible in the raw HTML. Claude drafts an answer-first rewrite of the section and a ticket for the developers. The rewrite goes into the approval queue, the developer fix is scheduled, and the client gets a two-sentence explanation in that week's update. The whole sequence takes under an hour.

How Rankwyn helps

Rankwyn's MCP server gives Claude portfolio-wide tools for audits, AI visibility checks and on-page fixes across every client site. Read tools work instantly; every change Claude drafts lands in the approval queue as a dry-run diff. Setup takes a few minutes; see the MCP setup guide.

Key takeaways

Claude is most useful for SEO when it works from real data, pasted in or connected through MCP.
High-value workflows include traffic diagnosis, cannibalisation, quick wins, briefs, audits, robots.txt reviews, on-page drafts, internal links, AI visibility, entity checks, report summaries and redirect maps.
Save prompts that work as shared templates so results are consistent across the team.
Verify important findings at the source, and keep facts about clients coming from clients.
Route every site change through review and approval.

The bottom line

Claude saves SEO teams the most time on analysis and drafting, provided it works from real data and its output is checked. Connect it to the tools you already use, start with read-only workflows, turn good prompts into team templates, and keep a human approval step between every draft and every live site.