AI-Powered SEO Audit: How to Analyse and Fix Your Site in 2026

This article is part of the series "Rebuilding a Website with AI". Previous episode: How to Add Smooth Page Transitions.

Why Run an SEO Audit?

After adding internationalisation, 50 blog posts, new pages and animations, the site had changed significantly in terms of structure. And structural changes mean SEO risk. URLs that break, metadata that no longer matches, orphaned pages.

I asked Claude Code to run a full SEO audit. What it found surprised me.

The RSS URLs Were Broken

The RSS script (scripts/rss.mjs) was still generating URLs based on the old route structure, without the locale prefix. The result: every link in the RSS feed led to a 404. Claude Code overhauled the script to generate correct multilingual URLs, with the right locale prefix for each article.

The Sitemap Was Incomplete

The sitemap only listed French pages. With the addition of English and Portuguese, two thirds of indexable pages were missing. Claude Code rebuilt app/sitemap.ts to list all pages in all 3 languages, add hreflang attributes to indicate alternative versions, include the new service pages (SaaS, Sprint), and exclude articles marked as draft: false.

Structured Data Was Missing

No Schema.org structured data on the site whatsoever. Google couldn't understand that AsuOs is an organisation, that it offers services, or that blog posts are articles. Claude Code added Organization markup (name, logo, URL, social networks), Service (for the Sprint and SaaS pages), BlogPosting (for each article) and BreadcrumbList (for navigation).

robots.txt and Metadata

The robots.ts file still referenced the old route structure and didn't include a link to the new sitemap. Fixed. The <title> and <meta description> tags were in French everywhere, even on English and Portuguese pages. Claude Code integrated getTranslations() into the generateMetadata() functions of each page to produce metadata in the appropriate language.

Tags Were Mixed Across Languages

On the /tags page, you could see tags from all articles across all languages. An English-speaking user would see tags from French articles they couldn't read. Claude Code filtered tag-data.json by locale: each language version of the site now only displays tags from articles in its own language.

The same problem existed on the homepage: "recent articles" weren't filtered by language. A visitor on /en would see French articles. Fixed with filtering based on the active locale.

Bonus Fixes

Beyond the pure SEO issues, Claude Code also identified and fixed the mobile menu not closing after navigation (a classic but annoying UX bug), insufficient contrast on Sprint cards (non-compliant with WCAG accessibility criteria), dead pages left over from the migration to the [locale] structure, the tag-data.json format that wasn't aligned with Prettier and caused phantom diffs on every build, the Contentlayer configuration that needed a computed locale field and support for .en.mdx / .pt.mdx files, and it created a translation script (scripts/translate.mjs) to streamline future article translations.

What This Reveals

An SEO audit is the kind of task you know is important but never get around to because it's tedious. You have to check every URL, every meta tag, every link in the sitemap, every piece of structured data.

Claude Code completed this audit in a matter of minutes: reading all relevant files, identifying inconsistencies, and fixing them on the spot. All documented in a report (docs/seo-audit-2026-03-12.md) for the record.

The result: a technically clean site, correctly indexable in all 3 languages, with structured data that helps Google understand exactly what we offer.

Tomorrow, the final episode. We step back for the big picture: what did this experience teach me about working with AI? What works, what doesn't, and what does it concretely change when you're a freelance developer in Switzerland? The conclusion, and a few surprises.