How to use Claude to triage SEO audits
The author reports that eight of 50 URLs reached Sonnet while the rest resolved at Tier 1, reducing that run from about $0.30 to about $0.05.
dev.to
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Audit [project URL list] using deterministic checks for title length, description presence, H1 count, and canonical tags. Escalate ambiguous pages to Haiku, then semantic edge cases to Sonnet. Produce severity-sorted findings and rewrite suggestions using [voice sample].
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 that eight of 50 URLs reached Sonnet while the rest resolved at Tier 1, reducing that run from about $0.30 to about $0.05.
- Human involvement
- The author compared a 50-URL run and incorporated public architectural feedback; review of individual suggestions is not described.
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
- Restructure the repository into core and premium modules.
- Run deterministic checks before model escalation.
- Route ambiguous pages through Haiku and Sonnet.
The author rebuilt an SEO audit agent after a comment thread challenged the need to use a language model for mechanical checks. Title length, description presence, H1 count, and canonical tags became the deterministic first tier. Haiku handles cases needing a closer look, and Sonnet is reserved for semantic ambiguity.
The repository kept its original core modules and added a premium layer for routing, isolated client projects, enhanced reports, and rewrites. The routing function runs the first tier before escalating to Haiku, then to Sonnet when Haiku flags semantic judgment. The reporter produces PDFs with screenshots, severity ordering, and suggested fixes. A rewrite component uses deterministic title truncation, Haiku description suggestions, and Sonnet opening paragraphs, with an optional writing sample.
A unified command reads a project's input CSV, audits URLs, generates rewrite suggestions, writes the report, and appends audit history. In the author's reported run of 50 URLs, eight reached Sonnet and the rest resolved at the first tier. Reported cost fell from about $0.30 to about $0.05. Who it's for: developers operating repeated client SEO audits who want model calls concentrated on judgment rather than character counting.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: dev.to · dannwaneri
Frequently asked questions
Which checks avoid a model call?
The first tier uses Python for mechanical checks such as overlong titles, missing descriptions, and missing H1 headings.
What does the rewrite component produce?
The post describes deterministic title truncation, Haiku-generated meta description suggestions, and Sonnet-written opening paragraphs, optionally informed by a writing sample.
More details
Remaining steps
- Generate rewrite suggestions and a PDF audit report.
- Append the completed run to audit history.
About this account
The audit routes URLs through deterministic checks and model tiers, then generates suggestions and reports.
Platform: dev.to. Attribution: https://dev.to/dannwaneri/i-was-paying-0006-per-url-for-seo-audits-until-i-realized-most-needed-0-132j. 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
Python · Claude Haiku · Claude Sonnet · WeasyPrint
SEO · Model routing · Reports
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.