How I Optimized My SEO with Claude: A Step-by-Step Workflow

The short answer: I stopped exporting CSVs from Google Search Console and pasting them into a chat. I connected Claude directly to my real SEO data through Rankeero, and now I ask questions in plain English. Claude reads Search Console, Google Analytics 4 and PageSpeed, runs the analyses an SEO would run by hand, and hands me a ranked list of actions. Then I schedule the whole thing so it lands in my inbox every week as a PDF.

This article walks through the exact workflow, step by step, so you can reproduce it on your own site or your clients' sites. If you are new to the idea, start with the guide to SEO with AI.

Why SEO is really a data problem

Most of the SEO work that moves the needle is not mysterious. It is a loop:

1. Look at what Google already shows you for.
2. Spot where you are close to a win (page two, weak titles, competing pages).
3. Fix it.
4. Measure again.

The hard part is not the thinking. It is the plumbing. Search Console caps its interface exports, Analytics lives in another tab, PageSpeed in a third, and every analysis means filtering, sorting and cross-referencing by hand. By the time you have the answer, you have lost an afternoon.

A large language model like Claude is very good at the thinking half: reading a table of 500 queries, noticing the pattern, and explaining what to do. What it lacks by default is access to your data. Paste a screenshot or a CSV and you get an answer about a snapshot that is already stale.

The fix is to give Claude a live, structured connection to the data instead.

What Rankeero adds to Claude

Rankeero is an MCP connector. MCP (Model Context Protocol) is the open standard that lets an AI assistant such as Claude call external tools. You add Rankeero to Claude with a single URL, sign in with Google, and Claude gains a set of SEO tools that read your accounts:

• Google Search Console: queries, pages, positions, click-through rates, indexing status and sitemaps.
• Google Analytics 4: sessions, conversions and landing pages.
• Page speed: Lighthouse audits through PageSpeed Insights, and real-user data from the Chrome UX Report (CrUX).
• Reports and email: PDF reports laid out for you, and Gmail drafts to send them.

Search Console is open today. Google Analytics 4 and Gmail are coming soon, while Google verifies the app, so the Analytics step and the emailed report below apply as they open.

Two design choices matter for a professional workflow:

• Projects and workspaces. A workspace holds your projects (one per site or brand), and each project groups its sources: a Search Console property, a GA4 property and so on. If you manage several clients, each one stays separate, and you can ask a question across all of them.
• Your own accounts. Rankeero reads through the Google account you connect. It does not copy your data into a separate database for Claude to browse.

From there, you just talk. Here is what I ask, in the order I ask it.

Step 1: Get a baseline

Before changing anything, I want to know where the site stands and what moved recently.

"Give me a performance overview of my site for the last 28 days compared with the previous 28. What grew, what dropped, and which queries and pages are behind it?"

Claude calls the Search Console tools that compare two periods, and comes back with the totals (clicks, impressions, CTR, average position), the biggest gainers and the biggest losers. Instead of eyeballing two charts, I get the explanation: which pages lost clicks, and which queries went with them.

Two follow-ups make the baseline much more honest:

• Brand vs. non-brand. Branded searches flatter your numbers. I ask Claude to split brand and non-brand traffic using my brand name, so I can see whether the growth comes from people who already know me or from new visitors discovering me through generic queries.
• Position distribution. How many queries sit in positions 1 to 3, 4 to 10, 11 to 20, and beyond? And how many entered or left the top 10 since last period? This tells me whether I am climbing, stalling or leaking.

This first step usually takes under a minute, and it gives me a map for everything that follows.

Step 2: Find the quick wins

The cheapest SEO gains hide in pages that already rank but under-perform. Two analyses find them.

Striking distance: queries on the edge of page one

"Which queries am I ranking between position 4 and 20 for, with real impressions? Sort them by how many extra clicks I would get if they reached the top 3."

Rankeero's striking-distance analysis lists queries with meaningful impressions and an average position between 4 and 20, and estimates the upside against the click-through rate your own site gets in the top 3. These are the pages where a better title, a stronger internal link or an extra section can push you onto page one.

CTR opportunities: good rankings, weak titles

"Which pages rank well but get a click-through rate far below what I would expect for their position?"

Here Claude compares each row to your site's own typical CTR at that position and flags the ones at or below half of it. These are title and meta description rewrite candidates: Google already likes the page, searchers just are not clicking.

What I like about asking Claude instead of reading a table is the next step. It does not stop at the list. It can open the page's top queries, read the current title, and propose new title and description options aimed at the queries that page actually ranks for.

Step 3: Fix cannibalization and content decay

Two problems quietly drain rankings, and they are hard to see in the Search Console interface.

Keyword cannibalization

When two of your pages compete for the same query, Google may alternate between them, and neither ranks as high as one strong page would.

"Show me queries where two or more of my pages split the impressions or clicks. For each, tell me whether to merge, redirect or differentiate."

Rankeero finds queries where several pages each take a meaningful share and returns the split per page. Claude then reads that split and gives a verdict: consolidate into one page, or sharpen each page's angle.

Content decay

Pages that used to perform and slowly stopped are the best place to spend refresh time.

"Which pages have lost the most clicks over the last few months, and why?"

The decay analysis looks at consecutive time windows and reports, per page, how many clicks were lost since its best window. It also classifies the pattern (sudden drop vs. gradual slide) and the likely cause: lost rankings, lower demand, lower CTR, or a page that disappeared. It checks seasonality against the same period a year earlier, so a Christmas page does not get flagged in July. It even lists the queries each page lost.

That last detail changes the refresh brief: instead of "update this article", I know which queries to win back.

Step 4: Plan content from real queries

SEO tools often hand you keyword lists disconnected from your site. Here I start from what Google already associates with me.

• Topic clusters. Claude groups my queries by shared words into topics, with their combined clicks, impressions and average position. It flags topics split across several pages, topics where only the homepage ranks, and topics stuck beyond page one.
• Content briefs. For one topic, say "technical SEO audit", Claude collects the matching queries, the questions people ask, the subtopics, the pages ranking today, and the gaps: queries that the topic attracts but that no page of mine answers.

"Build a content brief for the topic 'internal linking'. Which of my pages owns it, what questions am I missing, and what should a new or updated page cover?"

The result is an outline anchored in real demand, not in a generic "People also ask" scrape. Claude is also well placed to write the first draft, since it already holds the queries, the questions and the current page.

Step 5: Check the technical side

Content is only half the story. If a page is not indexed, or loads slowly, the best copy will not rank.

Indexing. Rankeero exposes Google's URL Inspection. I can ask:

"Inspect these 20 URLs. Which are not indexed, which are blocked by robots.txt, and where does Google's chosen canonical differ from mine?"

Claude returns a compact table: verdict, coverage, last crawl, canonicals with a mismatch flag, and rich-result status, plus a tally of problems.

Sitemaps. Claude can list the sitemaps Search Console knows about and, with your permission, submit a new one. Writes are kept separate from reads, so asking questions never changes anything on your property.

Speed. Two tools cover page performance:

• A PageSpeed (Lighthouse) audit gives lab metrics and the failing audits for any URL.
• Chrome UX Report data gives real-user performance for an origin or URL, with history, so you can see whether Core Web Vitals are trending the right way for actual visitors, not only in a lab.

I combine them: "Audit my five most-visited landing pages. Which have poor real-user metrics, and what are the top fixes for each?" Claude reads the field data, runs the lab audit, and writes the fixes in order of impact.

Step 6: Join search data with Analytics

Rankings are a means. Traffic that converts is the goal. Because Rankeero reads both Search Console and GA4 for the same project, Claude can join them, which is tedious to do by hand.

"Which landing pages bring me Google organic traffic and which of those actually convert? Show me pages with high impressions but weak engagement or conversions."

This separates two very different kinds of page:

• Pages that rank and convert: protect them and build links to them.
• Pages that rank and bounce: the intent is mismatched, or the page has no call to action.

It is also how you catch a ranking win that does not matter. Moving from position 9 to 4 on a query that never converts is not a priority, however nice the chart looks.

Step 7: Get one ranked action list

After all those analyses, the question becomes: where do I start? Instead of juggling six reports, I ask for a roadmap.

"Build an SEO roadmap for the next month. Rank actions by estimated impact and effort."

Rankeero's roadmap tool runs four analyses at once (CTR rewrites, striking distance, content refreshes, and merging cannibalized pages) and returns a single prioritized list. Each action names the page, the query, an estimated click gain, an effort level (low, medium or high), a score that weighs impact against effort, a plain sentence saying what to do, and the evidence behind it.

A word on the numbers: the estimates are ceilings, calculated from the data over the period you choose. Treat them as a way to rank work, not as a forecast. The value is in the ordering.

Automate it: scheduled tasks and PDF reports

Running this workflow once is useful. Running it every week without thinking about it is where it compounds.

Claude supports scheduled tasks: you write a prompt once, choose when it runs, and Claude executes it on that schedule using the tools you have connected. Because Rankeero is one of those tools, a scheduled task can read fresh Search Console and Analytics data every time.

Here is the weekly job I use, in plain language:

"Every Monday at 8:00, for my site: compare the last 7 days with the previous 7, list the three biggest movers, run the SEO roadmap, and pick the five best actions. Build a PDF report with key figures, a chart of clicks, and the action list. Create a Gmail draft to me with the PDF attached."

Rankeero has a PDF tool built for this. Claude writes the report as structured blocks (headings, paragraphs, key figures, tables, charts and callouts), and Rankeero lays it out in a clean, consistent style and returns a download link that expires after seven days. Gmail tools can attach that PDF to a draft, so the report arrives ready to review and send.

If you work for clients, the same pattern produces a recurring client-ready report with the recommendations already written in. Claude reads the data and decides what the report says, and Rankeero only formats and delivers it.

Here is what one run looks like, from the check-up to the report sent by email (the Gmail step opens soon, as noted above):

What Claude does not do for you

An honest version of this workflow includes its limits:

• It does not replace judgment. Claude proposes titles, briefs and priorities. You decide what matches your brand, your audience and your business goals.
• It does not change your site. Claude can tell you which title to rewrite, but you (or your developer or CMS) publish it.
• It is only as good as your data. A new site with few impressions gives thin analyses. Give Search Console time to collect data.
• Backlinks are out of scope. The workflow here is about on-site SEO, search performance and technical health, based on your own Google data.
• Estimates are estimates. SEO outcomes depend on competitors, algorithm updates and many factors outside any dataset.

FAQ

Can Claude do SEO on its own?

Claude can analyze data, find opportunities and write recommendations and drafts, but it needs access to your data. Without a connector it only knows what you paste into the chat. Connected to Search Console and Analytics, it can run real analyses on your site.

How do I connect Claude to Google Search Console?

You can use an MCP connector such as Rankeero. You add the connector's URL to Claude, sign in with Google, and choose the Search Console properties to expose. After that, Claude can query them in conversation.

Is my data safe?

Rankeero reads through the Google account you connect, and access is scoped to your workspace. Reads are separate from writes, so asking questions never changes your Search Console property. See Rankeero's Privacy Policy for the details.

Can I use this for several clients?

Yes. A workspace holds several projects, one per site, each with its own sources. You can ask a question about one project or compare across them.

What is the best first prompt to try?

Start with: "Compare the last 28 days to the previous 28 for my site and explain what changed. Then list the queries on the edge of page one with the most upside." It exercises the baseline and the quick-win analysis in one go.

Can Claude run these analyses automatically every week?

Yes. With a scheduled task, Claude runs the same prompt on a schedule, reads fresh data through Rankeero, and can produce a PDF report and a Gmail draft each time.

Does this work with ChatGPT or other assistants?

Rankeero is built on MCP, the open standard, and works with Claude as well as other MCP-compatible assistants such as ChatGPT. The workflow in this article is the same.

Try it on your own site

The shortest path from "I should look at my SEO" to "here is my action list" is three steps: connect Claude to Rankeero, point it at your Search Console property, and ask for a roadmap.

Get started with Rankeero and see what Claude finds on your site. If you want to compare plans first, see pricing.