How to use Claude to audit article SEO locally
All seven audited URLs failed at least one check, including a missing H1, meta-description problems, and overlong titles.
dev.to
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Audit [URLs] locally with Browser Use and Claude. Extract title, meta description, H1 count, canonical tag, and flags as JSON; check up to [link limit] links per page. Save resumable state and route login or access failures to [human review queue].
Reach people exploring AI agents
See placements →What happened
All seven audited URLs failed at least one check, including a missing H1, meta-description problems, and overlong titles.
Reported by dannwaneri on dev.toHow the agent did it
- Built a local browser audit agent with structured Claude extraction.
- Added concurrent broken-link checks and resumable JSON state.
- Implemented skip, retry, quit, and needs_human access handling.
Where a person stepped in
The author tested their own content; access problems can prompt a human or be logged in needs_human for later review.
A person approved each step before the agent continued.
The author built a local SEO audit agent and tested it on seven URLs from their own published content. Every URL received a failing result. The checks found a missing H1 on the Hashnode profile, description problems on freeCodeCamp pages, and overlong titles on DEV.to articles. The author distinguishes platform-controlled description templates from titles they wrote themselves.
The system uses Browser Use to navigate rendered pages in visible Chromium. Claude Sonnet reads page snapshots and returns structured JSON containing title status, description status, H1 count, canonical information, and flags. A separate httpx component checks links with concurrent HEAD requests, capped at fifty links per page and a five-second timeout.
Flat JSON state records completed work so an interrupted run can resume. Access problems receive explicit handling: an interactive run can skip, retry, or quit, while automatic mode records affected URLs in needs_human and continues. The post also describes a Windows scheduling option, though the demonstrated result is the author's own seven-URL audit. The implementation consists of seven Python files and runs on a Windows laptop.
Who it's for: Content owners and agency operators who want repeatable metadata checks with a queue for access issues.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: dev.to · dannwaneri
Frequently asked questions
What did the seven-URL audit find?
It flagged all seven URLs. Findings included a missing H1 on the Hashnode profile, description issues on freeCodeCamp pages, and overlong DEV.to titles.
How are inaccessible pages handled?
Interactive mode offers skip, retry, or quit. Automatic mode skips the URL, records it in needs_human, and continues processing.
More details
Remaining steps
- Ran the audit on seven of the author’s own published URLs.
About this account
The agent navigates pages, extracts SEO fields, checks links, and records progress and access exceptions.
Platform: dev.to. Attribution: https://dev.to/dannwaneri/i-built-a-local-ai-agent-that-audits-my-own-articles-it-flagged-every-single-one-pkh. 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 30 Mar 2026
Source checked 14 Oct 2026
Tools and topics
Browser Use · Claude API · Chromium · httpx · Python · JSON state files
SEO audit · Local automation · Human review
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.