Skip to content

Use Claude for SEO, from real data to one useful change

Updated Sep 30, 2026
Short answer

Connect Claude to Search Console, choose one question, and ask for evidence before a recommendation. Use the workflows below to find an existing page worth improving, investigate a traffic drop, check overlapping pages, or draft a specific edit. Review the change before publishing and record what you changed.

On this page

The most useful first SEO task for Claude is small: identify one page that deserves attention, explain the evidence, and produce a change you can review. A generic request to “improve my SEO” leaves too much unspecified.

This guide gives you four research workflows and a review checklist. The prompts work with a connected assistant; the examples are illustrative, not customer results. We build Percy and the free Search Console MCP used here.

Give Claude the right inputs

Start with your website, what it sells or helps people do, and your priority. For example: “We sell scheduling software to independent tutors. A qualified trial signup matters more than visits to our general study tips.” That context helps Claude distinguish useful traffic from a large but irrelevant audience.

Next, supply evidence. There are two practical routes:

  • Connect Search Console through MCP. Claude can request page and query performance, compare dates, and inspect indexing status. Follow our Search Console connection guide for installation, or choose your client on the MCP setup page. Anthropic documents how Claude Code connects to MCP tools.
  • Provide exports. Export comparable periods from Search Console. Include page and query tables, dates, and filters. Ask Claude to identify what the files can and cannot establish. Separate page and query exports do not establish which query belongs to which page.

Before asking for recommendations, have Claude confirm the property and date windows. Keep search type and any country or device filters consistent. Use complete periods; a partial current day is a poor comparison with a complete previous day.

Search Console gives you Google search performance. It does not tell you a visitor's revenue, every competitor's traffic, or whether an AI assistant cited your brand. Google's Search Analytics API documentation also notes that query results are subject to data limits. Treat returned rows as evidence with known coverage, rather than an exhaustive census.

1. Find SEO quick wins on existing pages

Use this when you have existing pages and limited time. The goal is a short list of changes you can actually make.

Find up to three SEO quick wins on this site. Compare the last 28 complete days with the previous 28, then inspect the relevant queries and pages. Prioritize business relevance and implementation effort, not average position alone. Give me the evidence, one next action per page, and any important uncertainty. If the data is too sparse, say so rather than inventing click forecasts.

Look for a recommendation that names the page, shows the relevant queries, and explains why the page is valuable to your business. A ranking at position 11 alone is not enough. The search intent might be wrong, the page might need substantial work, or the apparent opportunity might rest on only a handful of impressions.

Ask a follow-up: “Which one would you do first if I have an hour, and what evidence would change your mind?” If the data is sparse, a reasonable answer can be to fix a clear page problem and collect more evidence. For more detail, see how to prioritize SEO quick wins.

2. Investigate a traffic drop

Use this when clicks have fallen and you need to understand where to look first.

Investigate this site's Google organic traffic. Compare the last 28 complete days with the previous 28 and identify the pages and queries contributing most to lost clicks. Check relevant page evidence and separate observed changes from possible causes, including query mix and seasonality. Do not infer an algorithm penalty or an AI Overview effect from dates alone. If there is no meaningful decline, tell me. End with the first check or action worth taking.

A useful answer separates what changed from what might have caused it. For example, a page can lose clicks because impressions fell, CTR fell, or its query mix changed. A date that coincides with a Google update does not establish that the update caused the decline.

Ask Claude to identify the largest contributors in absolute lost clicks, then check the affected queries and pages. Keep seasonal comparisons in view. Google's traffic-drop troubleshooting guide is a useful reference for that investigation. Our traffic-drop guide includes a diagnostic you can work through before editing.

Use this when two URLs appear for similar queries and you are considering consolidating them.

Check this site for possible keyword cannibalization using query-to-page Search Console data across comparable complete periods. Find up to three meaningful overlaps, inspect the candidate pages, and distinguish healthy multi-page visibility from competing pages serving the same intent. Show URLs and evidence, state uncertainty, and recommend either keeping, differentiating, or investigating consolidation. Do not recommend redirects or canonicals from shared keywords alone.

The critical follow-up is to read both pages. Two pages can legitimately serve different intents even when some query terms overlap. Shared keywords alone are not a reason to redirect one, add a canonical, or delete useful content.

Ask for the URLs, the shared queries, the performance of each page across comparable periods, and the difference in purpose. The answer should explain whether keeping both, differentiating them, or investigating consolidation makes sense. See the Search Console cannibalization workflow for a fuller decision process.

4. Draft one page improvement you can publish

Use this after choosing a page. The output should be concrete enough to review beside the existing text.

Find one existing page worth improving using the last 28 complete days of Search Console data and the previous 28. Explain why it matters to this business. Read the live page before drafting a specific title, introduction, or section change. Preserve the URL, existing search intent, and useful content. Show the evidence, the current text, the proposed replacement, and what to check after publishing. Do not invent traffic forecasts or claim the change has been published. If you cannot read the page, ask me to paste it before drafting.

Claude needs the actual page content as well as search data. The Search Console MCP does not fetch your page copy. If your assistant cannot open the page through another available tool, paste the relevant section yourself.

An acceptable draft names the current text, the proposed replacement, and the reason for the change. Reject invented testimonials, unsupported product claims, or promises about future rankings. Preserve useful coverage and the page's existing purpose. A title rewrite cannot rescue a page that fails to answer the query.

Review the answer before publishing

Use this short checklist for any of the four workflows:

  1. Evidence: Does the answer identify the property, dates, pages, and queries it used? Can you trace important numbers back to the data?
  2. Business value: Does the action help the audience you want to reach? Which meaningful next step should the page support?
  3. Specific change: Can you point to the exact edit, or is the recommendation still “improve content quality”?
  4. Preservation: Does the change retain the URL, useful content, search intent, and working canonical unless there is a reason to change them?
  5. Follow-through: Have you recorded the edit and the publication date so you can revisit performance?

After publishing, verify that the live page contains your intended change. Revisit comparable periods once enough data is available, keeping seasonality and other changes in mind. A before-and-after difference is not proof that your edit caused the result. For a young site, the first useful outcome may be a better page and a reliable measurement habit.

Choose where to keep the work

Claude with a connected MCP is a good fit when you already work in an assistant and want direct access to your own search data. Keep a simple log of the recommended action, its evidence, what you published, and when to review it.

Percy provides a dedicated workflow for that work: connect a site, investigate a question, draft a page change, and save a next action. The saved action is linked to your site's plan so you can return to it after the conversation. You review and publish the change yourself.

If you are comparing Percy with ChatSEO, use the same site and question in both. ChatSEO advertises broader integrations, including CMS publishing. Check the current features and pricing against what you need. We have not run a controlled head-to-head product test, so this guide does not claim Percy produces better results.

Try the page-improvement workflow in Percy. Your seven free days start with your first completed chat answer. No card is required; continuing with one site costs $29/month. Trial usage limits apply. For other approaches, read our AI SEO tool comparison.

Frequently asked

+Can Claude access Google Search Console?
With an authorized connection, yes. A compatible MCP server lets Claude request Search Console data for properties you can access. You can also provide exports. Without either, Claude cannot know your site's actual clicks, impressions, or queries.
+Is the Search Console MCP server free?
Percy's open-source search-console-mcp-server is free and MIT licensed. Claude may have separate subscription or usage costs. The free MCP does not require a paid Percy account.
+Can Claude publish SEO fixes through this MCP?
No. This server reads Search Console data. It does not edit your website or submit pages for indexing. Publishing requires your CMS or a separate development workflow, with your review.
+Should I use Claude, ChatSEO, or Percy?
Use Claude with the free MCP if you prefer doing research in your existing assistant. Evaluate a dedicated SEO product if you want connected research and follow-through in one workspace. Test the same real question in each product and compare evidence quality, usable output, and cost.