Skip to content
Climb Google Maps: Google Business Profile optimization.

Technical SEO

Google's Modern Web Guidance 1.0 Teaches AI Coding Agents Current HTML and CSS. It Won't Do Your SEO

Chrome's skill set for AI coding agents left preview on Nov. 10 with 203 guides, a legacy code scanner, and Firefox, Edge and MDN as co-maintainers. We read every guide in the package: strong on speed, forms and accessibility, almost silent on search.

Two monitors showing code above a keyboard on a developer's desk

Key takeaways

  • Google's Modern Web Guidance 1.0, a free skill set that gives AI coding agents current web platform guidance, was released on Nov. 10, 2026 with 203 use-case guides.
  • Version 1.0 is maintained by the Chrome, Edge, Firefox and MDN teams and adds a legacy-scan command that flags outdated code patterns in an existing project.
  • Google's own evals report Claude Code passing 94% of checks with the guidance versus 58% without; those are Google's numbers, not our test.
  • In the 1.0.0 package we counted 26 performance entries and Core Web Vitals named in a handful of guides, but no guide covers titles, canonical tags, structured data or crawling.
  • Site owners building with AI agents should install it for speed and accessibility, and keep their own SEO checks running on every build.

Google's Chrome team released Modern Web Guidance 1.0 on Nov. 10, 2026. It is a free, open source set of skills that an AI coding agent such as Claude Code, Codex or Google's Antigravity reads before it writes HTML, CSS or browser JavaScript, so it stops reaching for patterns from years ago. Version 1.0 takes it out of preview, ships 203 guides, and adds a scanner that finds outdated code in an existing site.

If you build or edit websites with an AI agent, it is worth installing today. But we read the whole package, and one thing should be clear before you rely on it: it makes agents write better front-end code, not better SEO.

What's new in version 1.0

From the v1.0.0 release notes on GitHub:

  • Out of preview, with 203 use-case guides covering 250 web platform features.
  • New maintainers: the project is now run by the Google Chrome, Microsoft Edge, Mozilla Firefox and MDN teams, with the developer community.
  • A legacy code scanner: npx modern-web-guidance@latest legacy-scan [path] scans a codebase for old patterns and maps each finding to the guide with the modern replacement. The notes say it covers roughly 20% of guides, the ones with rules simple enough to detect automatically.
  • Smaller and offline: search now runs on a small built-in model plus keyword matching, works offline, and the npm download fell from 23.5 MB to 8.8 MB.
  • 15 new guides and 98 updated ones, nearly 70 of them with accessibility changes.

Google also published eval results in the release notes. It reports that Claude Code with Claude Opus 5.5 passed 58% of automated checks without the guidance and 94% with it; Codex CLI went from 50% to 86%, and Antigravity CLI from 60% to 98%. Those are Google's numbers from Google's own test tasks. We haven't reproduced them.

How do I install Modern Web Guidance in Claude Code?

Chrome's documentation lists one command per agent:

  • Claude Code: claude plugin install modern-web-guidance@claude-plugins-official
  • Any agent, through Chrome's own installer: npx modern-web-guidance@latest install
  • Vercel's skills tool: npx skills add GoogleChrome/modern-web-guidance
  • Copilot CLI: /plugin marketplace add GoogleChrome/modern-web-guidance, then /plugin install modern-web-guidance@googlechrome

Chrome's docs say that without a setting, the guidance will "default to Baseline Widely available," meaning features that have worked across major browsers for a while. To target newer features, add a line such as "This project's Baseline target is Baseline 2024." to your CLAUDE.md or AGENTS.md file.

One detail worth knowing from the skill's own instructions file: the agent is told to run npx -y modern-web-guidance@latest for every search, so each lookup fetches whatever version is newest on npm. That is convenient, and it also means your agent runs fresh third-party code during the build. If your team pins dependencies for everything else, ask whether you want the same here.

What we found in the 203 guides

We downloaded the 1.0.0 package from npm and read its guide index and guide files. The index lists 208 entries (the 203 guides plus overview pages) across 17 categories. The biggest are UI behaviors (40), performance (26), UI components (22), visual design (22), forms (19), CSS (18) and security (17).

What that means for a site that lives on search:

  • Page speed is well covered. The performance guides include prioritizing the image that sets Largest Contentful Paint (LCP), breaking up long tasks that hurt Interaction to Next Paint (INP), and prefetching likely next pages. LCP is named in eight guide files, INP and Cumulative Layout Shift (CLS) in six each. Google Search Central says Core Web Vitals, with other page experience aspects, align with "what our core ranking systems seek to reward."
  • Accessibility and forms are strong, including autofill-friendly sign-up, address and payment forms.
  • There are three WebMCP guides, for exposing a site's forms and functions as tools that AI agents can call. That is early, Chrome-led work, and it is the part of this package closest to "agent-ready websites."
  • SEO basics are absent. No guide covers titles, meta descriptions, canonical tags, structured data, sitemaps, robots.txt or crawling. The word "SEO" appears once in the whole package: a line in the drill-down menu guide telling agents to use real <a href> links rather than clickable buttons, because real links get "SEO for free." No guide mentions Googlebot or search engines.

That isn't a flaw. The project is about the web platform, and it says so. But an agent that follows it closely can still ship a fast, accessible page with no description, a duplicate title or broken structured data.

Our view: install it, and keep your SEO checks

This site is built and run by Claude Code (with Claude Opus 5.5), so this is a tool for exactly how we work. We haven't run it on this site yet; it is what we are testing next. Here is what we'd tell anyone building sites with an agent:

  1. Install it in the agent you already use, and add a line to your CLAUDE.md or AGENTS.md naming your Baseline target.
  2. Run legacy-scan on an older site before asking an agent to modernize it. Fix what it finds in small batches and test each one, because a rewrite of a menu or modal can change what users see and what links crawlers can follow.
  3. Keep an SEO check on every build. Titles, descriptions, one H1 per page, canonicals, alt text and internal links still need their own test. Our guide How to build a website with Claude includes a 30-line script that does this on a static build.
  4. Measure Core Web Vitals on real pages in PageSpeed Insights or Search Console after the agent's changes, rather than trusting any pass rate.

What we don't know yet

  • Whether Google's eval gains show up on real sites. The tasks are Google's; there are no independent results yet.
  • How much legacy-scan finds on a typical CMS site. It covers about a fifth of the guides by Google's count, so a clean scan is not a clean bill of health.
  • Whether search features will be added. Nothing in the release notes says so.

Sources

Photo of Kiran Dasgupta

About the author

Kiran Dasgupta, Editor

Kiran runs Kihan Marketing, a web design, SEO and ads agency, and manages dozens of websites across real estate, home services and healthcare. He writes about SEO, AI search and paid ads news, and what it means for the people who run websites.