Ask an AI agent to "improve SEO" and you usually get a pile of keyword ideas, a rewritten meta description and a confident promise about rankings. The indexing problem that is actually costing you traffic stays exactly where it was.
Website Optimizer makes your agent work like a senior optimisation team — technical SEO, performance, accessibility and conversion — and makes it back every finding with evidence and rank every fix.
Where agent SEO advice goes wrong
- It rewrites content before checking whether the page can even be indexed.
- It chases a Lighthouse score instead of real users.
- It recommends
llms.txt, "AI schema" or keyword density as ranking levers. - It adds structured data that doesn't match what is on the page.
- It mixes measurements, guesses and opinions into one list.
How the skill works
- Understand the site first — audience, markets, languages, and the one conversion that matters.
- Check crawling and indexing — robots, sitemaps, canonicals, redirects, hreflang, and content that only exists after JavaScript runs.
- Review content, intent and structured data — one clear intent per page, real trust signals, and schema that matches the visible page.
- Measure Core Web Vitals against LCP ≤ 2.5s, INP ≤ 200ms and CLS ≤ 0.1, with a likely cause for each problem.
- Audit accessibility and conversion — WCAG 2.2 AA including keyboard checks, then calls to action, friction and trust, without dark patterns.
- Prioritise — every issue scored on impact, confidence and effort, and labelled P0 to P3.
It chooses one of three modes: Audit for a public site, Implementation when it has your repository (it makes the fixes and runs your build and tests), and Strategy for a site that doesn't exist yet.
What you get back
An executive summary, a scorecard across nine areas, a prioritised issue table, page-level recommendations and a 30-day plan. An illustrative line from the issue table:
| Priority | Issue | Evidence | Fix | Effort |
|---|---|---|---|---|
| P0 | Pricing page is not indexed | noindex in the rendered HTML | Remove noindex, add a self-referencing canonical | Trivial |
Knows your stack
Framework-specific checks for Next.js, React SPAs, WordPress, Shopify and Odoo websites: metadata APIs, sitemap routes, hydration cost, plugin conflicts and duplicate schema.
Who it's for
Developers, agencies and founders who use Claude Code or Codex on real sites and want audits they can act on, not a list of generic tips.
What it won't do
No keyword stuffing, doorway pages, fake reviews or ranking guarantees. It keeps verified findings apart from assumptions, and it won't break analytics, forms or checkout to chase a score.
