How to use Claude to fix website metadata
The author reports moving from four failing URLs to four passes with zero flags, spending under $0.15 across two audits and a rewrite run.
dev.to
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Inspect [website URLs] with Playwright after JavaScript renders. Check titles, descriptions, H1 headings, and canonical tags. Identify unrelated inherited metadata, propose page-specific rewrites, and rerun the same checks after I deploy approved changes.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- dannwaneri
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports moving from four failing URLs to four passes with zero flags, spending under $0.15 across two audits and a rewrite run.
- Human involvement
- The author deployed the changes, reset audit state, and reran the audit to verify all four URLs.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Audit the four URLs using the rendered DOM.
- Run the rewrite agent for metadata suggestions.
- Deploy shorter titles and page-specific descriptions.
The author tested a custom SEO agent on two owned domains and an invoice-tool subdomain. Its Playwright browser inspection read the rendered DOM rather than just static HTML. That distinction exposed homepage copy appearing as the description of an unrelated invoice generator after JavaScript execution.
The first audit marked all four URLs as failures. Findings included long homepage titles and descriptions, an overlong comparison-page title, and missing page-specific metadata on the invoice subdomain. The audit used tiered routing, with deterministic checks handling most pages and Haiku inspecting the invoice page more closely.
The author then ran the rewrite agent. Deterministic logic shortened titles, Haiku suggested descriptions, and Sonnet supplied voice-matched opening copy for the invoice tool. Deployed changes included a shorter homepage title, a trimmed homepage description, and a description tag in the React SPA's static index.html head.
After resetting state, the author reran the audit and reported four passes with zero flags. The post gives a combined cost below $0.15 for two audit passes and the rewrite run. Who it's for: website owners investigating metadata problems that appear only after client-side rendering.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: dev.to · dannwaneri
Frequently asked questions
Why inspect the rendered DOM?
The author found that homepage copy leaked into the invoice subdomain after JavaScript executed; inspecting only raw HTML would have shown an empty description.
Where was the invoice metadata fix deployed?
The author added a page-specific description tag to the static HTML head in index.html rather than placing the fix in JavaScript.
More details
Remaining steps
- Reset state and rerun the audit for verification.
About this account
The agent discovers rendered metadata problems and proposes rewrites through deterministic and Claude model tiers.
Platform: dev.to. Attribution: https://dev.to/dannwaneri/i-ran-my-own-seo-agent-on-my-two-domains-it-went-from-04-to-44-pass-in-one-afternoon-39an. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: dannwaneri on dev.to
Source published
Added to Agents In Use
Source checked
Tools and topics
Playwright · Python · Claude Haiku · Claude Sonnet
SEO · Metadata · React
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.