Claude can help you explore SEO data when you give it a clear question and a reliable dataset. Start with exports you control. A spreadsheet workflow works without connecting your accounts to a third-party service.
1. Prepare the data
Export the reports you need from Search Console and your analytics tool as CSV files. Include the date range, property, filters, and definitions of the metrics. Use comparable periods and exclude personal information that is not needed for the analysis.
Useful inputs include query and page performance, landing-page traffic, and a log of content changes. Keep the original exports so you can verify the output.
2. Explain what each file contains
Where your Claude account supports file uploads, provide the CSV files and describe each column. Otherwise, work in a spreadsheet and share a small, non-sensitive sample. Do not provide account passwords, API keys, or authentication cookies.
Analyse only the attached data. State the date range, filters, and missing fields first. Separate measured findings from hypotheses. Do not invent traffic, rankings, conversions, or audit scores.
3. Find opportunities worth reviewing
Quick-win candidates
From the query and page export for the last 28 days, identify queries with average positions between 5 and 15 and at least 200 impressions. Show the associated URL, clicks, impressions, and CTR. Explain why each page is worth inspecting before recommending a change.
These thresholds are a starting point, not a ranking formula. Review the actual page and search intent before editing it.
Referral traffic from AI tools
Using the source and landing-page data supplied, group referrals from identifiable AI tools. Compare two equal periods and show the landing pages with the largest changes. Label unattributed visits and missing referrer data as unknown.
Referral visits do not measure all brand mentions or citations. A recommendation can happen without sending a click to your site.
Manual visibility log
Summarise this manually collected prompt log by tool, date, and prompt. Calculate how often our brand appears in the observations and list the cited URLs. Do not generalise this sample to all users or answers.
Collect the observations first. An analytics export cannot reveal answers that were never recorded.
Declining pages
Compare the two supplied periods and rank pages by clicks lost. Show the underlying values, then suggest possible causes to investigate. Consider seasonality, changed demand, tracking changes, and page edits; do not present a cause as proven without evidence.
Competing pages
Find queries associated with several URLs in the query-page export. Identify possible overlap and the evidence for it. Suggest which pages to inspect for consolidation, differentiation, or internal-link changes.
Several URLs appearing for one query is not automatically a problem. Check whether they serve different purposes before merging or redirecting anything.
Technical review
Review the supplied crawl report and page excerpts. List observed indexing, link, title, and structured-data issues. Quote the evidence for each finding and separate untested checks from confirmed problems.
Use a crawler, browser, or measurement tool to collect technical evidence first. Do not treat an AI-generated score as a measured result.
4. Turn findings into a reviewable plan
For each proposed change, record the page, evidence, expected benefit, effort, owner, and measurement date. Make a small batch of changes, keep a change log, and compare results after a suitable observation period.
Optional: use a connector
A connector may provide access to live data, but availability, permissions, and behaviour depend on the provider and your account. Verify the provider's documentation, request only the access needed, and inspect the tools exposed before use. Never assume that an MCP connector is read-only.
The CSV workflow above remains available if you do not want to connect accounts. Review the measurement lesson and technical audit guide for the underlying checks.