Modern Web Guidance gives AI coding agents a new way to use current browser features, and version 1.0 adds an audit for existing code. Search Engine Journal reported that Google released the free skill set for development teams, with Firefox and Edge now joining the project.
The guidance packages recommendations that an agent can consult while writing code. Philip Walton, a senior staff developer relations engineer at Google, announced the release in a Chrome for Developers post.
Modern Web Guidance Adds a Code Audit
The audit reviews a project’s HTML, CSS, JavaScript files, and package.json. It identifies areas developers may safely update, then guides the agent through those changes. This first version uses static analysis and checks a limited number of code patterns, according to the post.
Google says the project now targets “hundreds of verified AI knowledge gaps” across its guides. Each guide is reviewed and approved by a domain tech lead. User-facing feature guides also receive a dedicated accessibility review from experts who use assistive technologies daily.
The project began as an early preview at Google I/O in May. Teams can install it in a project directory with the command shown in the announcement, or find it through marketplaces for popular coding agents. A governance document uploaded to the project’s GitHub repository on Nov. 9 outlines six owner seats: two for Chrome, one each for Microsoft Edge and Mozilla, and two for independent web platform experts. One independent seat is open, and Apple has no seat.
Tests Show Different Results With Guidance
Google’s public dashboard runs coding agents from Anthropic, OpenAI, and Google through the same tasks nightly, with and without the guidance. A browser checks each finished app’s DOM, styles, and source code. In the dashboard results for runs completed Nov. 8, guided runs passed 88% of automated checks, compared with 56% for unguided runs. Each agent’s combined results covered 199 to 215 tasks.
The dashboard reviews accessibility features such as keyboard support, focus, and form labels, but is not a full WCAG audit. Task prompts do not mention the guidance, so agents must decide when a guide applies. In most recent suites listed, guided runs took longer and used more tokens. This differs from reporting on Google and ChatGPT search volume, while addressing coding performance rather than search use.
Japan’s creator platform note inc. tested two agents across six user interface tasks and 180 test runs on clone versions of its production code base. Scores on 19 browser checks rose after the guidance was installed, then rose again when prompts explicitly named it. Agent One scored 60.8%, 76.5%, and 94.0% across the three conditions. Agent Two scored 75.8%, 86.2%, and 90.0%.
Google also reported that MetaMask used the guidance in automatic reviews of pull requests for its open-source browser extension. Through Oct. 21, during the first five weeks, the bot identified 102 issues across 57 feature and fix pull requests. Engineers fully resolved 43 and partially resolved three. Sixteen were declined or false positives, 20 went unaddressed, and the outcome of another 20 was unknown.
Guidance Still Needs Review
Google plans to expand audits of existing code and explore audits in which an agent reasons about the code under review. The company also suggests pairing the guidance with Chrome DevTools so agents can run created code and test it in a browser. Earlier coverage of Google’s unconfirmed search update shows why reported changes should retain their stated certainty; here, measured results come from Google and customer testing it highlighted.
For teams considering the guidance, the reported tests point to a practical next step: try it against a project, then keep field data and manual accessibility testing in the review. That approach matters as publisher SEO audits address discoverability, where technical checks and human review both shape what teams can assess.
