Claude SEO Plugin: Optimize for Google and AI Search

What This Workflow Actually Does

Claude Code can turn a website audit into a reviewed implementation workflow. The open-source Claude SEO plugin gives it repeatable commands for crawling pages, checking technical SEO, reviewing content and schema, reading Google data when authorized, and producing a prioritized report. When Claude also has access to the website code, it can prepare the fixes instead of stopping at a checklist.

Simon Pittman's demonstration follows a useful sequence: install the plugin, verify the site in Search Console, give Claude a brand and positioning brief, run the audit, inspect the action plan, implement selected changes, deploy a preview, and validate the result. The method is strong because it joins evidence, business context, and implementation. The weak version would be to accept every generated recommendation or promise that a schema block will put a company at the top of AI answers.

Watch the Original Demo

Credit: this guide analyzes I Used Claude to Optimise My Site for AI Search & SEO!, published by Systems Made Better on 20 June 2026. Simon Pittman demonstrates the workflow on his own website. His audit score and implementation results are a creator test, not a ranking guarantee.

SEO, AEO, and GEO: One Foundation, Different Surfaces

AEO usually means answer engine optimization, while GEO means generative engine optimization. They are useful labels for discussing visibility in answer engines and generated responses. For Google, however, the distinction is smaller than the marketing language suggests.

Google's current AI optimization guide says its generative experiences are rooted in the core Search ranking and quality systems. Pages still need to be crawlable, indexed, eligible to show a snippet, technically clear, and genuinely useful. Google also says there is no special schema for AI search and no need to rewrite pages only for machines.

The video adds an llms.txt file near the end. That can be maintained for services that support it, but Google now states that it ignores the file and that it does not affect Google rankings or AI-search visibility. Treat it as an optional interoperability artifact, not a core SEO task.

The Toolkit Used in the Video

The Claude SEO repository is MIT licensed. At the time of this review, it documents 26 sub-skills, 19 sub-agents, and optional integrations for services such as DataForSEO, Firecrawl, Ahrefs, Bing, and Matomo. That inventory has already changed since the video, so use the repository's live documentation rather than an old screenshot.

Install and Connect It Safely

The repository currently documents this Claude Code marketplace route:

/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo
/seo setup
/seo doctor

This is a third-party marketplace. Anthropic's plugin documentation warns that installed plugins can run code with the user's privileges. Review the repository, install at the narrowest useful scope, inspect updates before accepting them, and avoid handing a plugin production credentials it does not need.

For Search Console, first verify the domain or URL-prefix property. Programmatic access to private user data requires OAuth 2.0. Follow the official Search Console API authorization guide, keep client secrets and tokens outside the website repository, and start with read access. The video uses Claude to guide the DNS TXT verification and OAuth setup, but Claude's instructions should be checked against the current Google interface before every credential or DNS change.

1. Verify ownership of the exact site property in Search Console.
2. Create or select a Google Cloud project and enable the required API.
3. Create the appropriate OAuth client and use the minimum scope needed.
4. Store credentials outside source control and exclude local token files.
5. Ask Claude to confirm which property it can read before running an audit.

Give Claude the Business Context Before the Audit

The most transferable part of the video is not a technical command. Simon creates a project brief explaining who he is, the audiences he serves, his three areas of work, his preferred language, and the claims he wants the site to make. He also supplies voice rules so generated metadata and page copy do not collapse into generic AI language.

A useful brief should contain:

• the business, person, products, services, and geographic scope
• primary audiences and the problems they are trying to solve
• approved positioning, proof, credentials, and first-party sources
• topics the brand should and should not be associated with
• voice, terminology, spelling, and prohibited claims
• the primary conversion action for each important page

Keep this context in a small project brief and decision log rather than relying on one very long chat. That makes a later session easier to resume and gives reviewers a stable source for checking Claude's copy.

Run the Audit, Then Separate Facts From Recommendations

Simon asks Claude to run a full SEO and AEO audit with the plugin. On the newly launched site, the report returns a score of 66 out of 100 and identifies missing or weak elements: no submitted sitemap, little Search Console history, no dedicated About page, and opportunities to clarify identity and page-level structure. It also notes that image alt text is already in reasonable shape.

The exact score is less important than the evidence beneath it. Ask for a report with four layers:

1. Observed: status code, title, canonical, schema, sitemap, internal links, performance data, and Search Console output.
2. Inferred: likely intent mismatch, entity ambiguity, weak information hierarchy, or missing proof.
3. Recommended: the proposed change, dependency, owner, and expected benefit.
4. Measured: the metric and review date that would show whether the change helped.

This structure stops a plausible model suggestion from being presented as a measured defect. It also makes the action plan easier to prioritize by impact, confidence, and effort.

Implement Changes Through a Review Gate

Because Claude can reach Simon's codebase, it drafts an About page, revises metadata and page copy, adds structured data, updates the sitemap, improves internal paths, and deploys through Vercel. For a site built in Framer, Squarespace, or another managed platform, the same process can produce a copy-and-code handoff for manual entry.

The demo also catches a very ordinary failure: the newly deployed About URL returns a 404. Claude diagnoses the route and repairs it. That moment is a better lesson than a flawless montage. A generated change is not complete when the code exists; it is complete when the live URL works and the expected artifact can be inspected.

Structured data must describe content that visitors can actually see. Do not invent awards, credentials, reviews, locations, relationships, or services in JSON-LD. Use stable entity identifiers, but only connect entities when the relationship is real and supported by the page.

Validate the Result and Measure the Right Things

Use both validation tools because they answer different questions:

• Google Rich Results Test checks whether a page is eligible for Google-supported rich-result types.
• Schema Markup Validator checks general Schema.org markup without Google's feature-specific rules.

Then test the live site itself: request every changed URL, check mobile rendering, inspect titles and canonical URLs, confirm the sitemap contains the final routes, and verify that Search Console accepted the sitemap. Allow time for crawling and indexing before judging performance.

Measure outcomes in layers. Start with deploy health and indexability, then impressions and query coverage, then qualified clicks, leads, or conversions. Google now provides a generative AI performance report in Search Console where available. External answer engines may require separate manual checks or specialist monitoring, and attribution can remain incomplete.

Reusable Starter Prompts

1. Create the brief

Create a compact SEO project brief for [site]. Use only the sources I provide.
Include audiences, offers, positioning, proof, entities, preferred language,
prohibited claims, conversion goals, and open questions. Mark unknown facts.
Do not audit or rewrite the site yet.

2. Audit without editing

Use the Claude SEO plugin to audit [site]. Read the project brief first.
Separate observed facts, inferences, and recommendations. For every action,
show evidence, impact, confidence, effort, dependencies, and a success metric.
Do not edit files, change DNS, submit a sitemap, or deploy anything.

3. Prepare an approved change set

Implement only the approved items from the audit in a new branch or preview.
Preserve the existing design and voice. Do not add claims unsupported by the site.
Show the changed files and a review checklist. Do not deploy to production.

4. Verify after deployment

Verify every changed live URL, status code, title, canonical, internal link,
sitemap entry, visible content, and structured-data block. Report failures first.
Distinguish valid schema from Google rich-result eligibility.
Do not claim ranking or AI-citation success without measured evidence.

Video Chapters

Sources and Resources

• Systems Made Better: I Used Claude to Optimise My Site for AI Search & SEO!
• Claude SEO repository, installation, commands, and MIT license
• Anthropic: Claude Code plugins overview and trust model
• Google: optimizing for generative AI features on Search
• Google: succeeding in AI search
• Google Search Console API authorization; Search Console
• Google structured-data testing guide; Rich Results Test; Schema Markup Validator
• Cowork OS; Notion Agent OS; Better Creating resources
• Wispr Flow referral featured in the video; Notion referral featured in the video; Vercel

The video was published on 20 June 2026. The plugin repository, Anthropic plugin guidance, Google AI-search guidance, Search Console authorization, and schema validation documentation were checked on 26 September 2026. Plugin commands, supported agents, interfaces, and search features can change; verify the current primary documentation before granting access or implementing changes.