# The Technical SEO Audit Checklist We Run in 90 Minutes With Claude Code

**Author:** John Morabito (Founder, /winston)
**Published:** June 12, 2026
**Reading time:** 14 minutes
**Canonical:** https://www.winstondigitalmarketing.com/playbooks/technical-seo-audit-90-minutes-claude/

Our technical SEO audit checklist runs in about 90 minutes with Claude Code: 15 minutes to crawl and pull the data, 45 minutes for an AI pass that scores seven categories and flags every issue by severity, and 30 minutes for a human to verify and rank the fixes. The same audit by hand takes a day or more, almost all of it spent collecting and reformatting data instead of deciding what actually matters.

## Why the technical SEO audit checklist takes a day by hand

Open any traditional audit and watch where the time goes. You export a crawl from one tool, page-speed numbers from another, schema validation from a third, indexation data from Search Console, and then you spend hours pasting it all into a spreadsheet so the patterns line up. The analysis (the part that needs a brain) is maybe 20% of the clock. The other 80% is data wrangling that a machine should do.

Claude Code collapses the wrangling. Point a crawler at the site, hand the output to a Skill that holds the rubric, and the model reads thousands of URLs against the same standard you would, only without getting tired on row 1,400. You arrive at the judgment call with the data already sorted, scored, and deduplicated.

## The seven categories we score

Each category gets a pass, warn, or fail, with the exact URLs attached so the finding is actionable.

1. **Crawlability and indexation.** robots.txt sanity, sitemap coverage, accidental noindex tags, canonical chains, orphan pages, and the gap between pages submitted and pages indexed. The fastest wins usually hide here.
2. **Site architecture and internal linking.** Click depth from the homepage, internal link distribution, anchor text, and whether money pages are linked like money pages.
3. **Core Web Vitals and page speed.** LCP, CLS, and INP at the template level, separating real-user data from lab data.
4. **Structured data and schema validity.** Is there schema, is it the right type, and is it connected with stable @id references? See https://www.winstondigitalmarketing.com/playbooks/schema-markup-for-ai-engines-2026/.
5. **On-page fundamentals.** Title and meta uniqueness, heading hierarchy, duplicate and thin content, canonical correctness, H1-to-query match.
6. **Mobile and rendering parity.** Does the rendered DOM match the served HTML, or is content injected by JavaScript a crawler may never run? Same failure mode as the Dutchie iframe problem: https://www.winstondigitalmarketing.com/playbooks/dutchie-iframe-seo-problem/.
7. **The AI-engine layer.** Server-rendered content, llms.txt, markdown twins, citable chunk structure. The standard lives in https://www.winstondigitalmarketing.com/playbooks/how-to-get-cited-by-chatgpt-in-2026/.

## The Skills that automate each step

| Phase | What runs | Time |
|---|---|---|
| Crawl + data pull | Headless crawler exports URLs, status codes, titles, headings, schema, internal links; Search Console and Core Web Vitals pulled via MCP connectors | ~15 min |
| AI scoring pass | A Skill holding the seven-category rubric scores each category pass/warn/fail and groups issues by severity with affected URLs | ~45 min |
| Human verification | Spot-check the highest-severity findings, kill false positives, write the prioritized fix list in revenue order | ~30 min |

The data pull leans on the connector pattern in https://www.winstondigitalmarketing.com/playbooks/mcp-servers-for-marketing-teams/: the assistant reaches into Search Console and the crawler directly instead of exporting CSVs by hand. The scoring Skill is a written rubric plus instructions, which is why it is reproducible run to run.

The honest limit: the machine flags forty issues, but it cannot tell you which three are worth your quarter. Deciding that a slow LCP on a high-converting template beats a dozen orphan pages on a dead blog is judgment, and judgment does not automate. The list is the cheap part.

## What you can DIY (and where it stops paying)

This checklist is deliberately open. Run a crawler, hand the export to Claude Code with the seven categories written out as a rubric, and you get a credible first-pass audit on your own site. For a single small site, that is most of the value.

Where DIY stops paying is volume and prioritization: running this across a dozen client sites monthly, keeping the rubric current as the AI-engine layer shifts, and turning forty flags into a ranked fix list. We built this pipeline in a couple of focused sessions, the same way we built the site in https://www.winstondigitalmarketing.com/playbooks/how-we-built-this-site-in-6-hours/, and it feeds the broader review in https://www.winstondigitalmarketing.com/playbooks/the-complete-geo-audit-methodology/.

## FAQ

**How long does a technical SEO audit take?** About 90 minutes on a small-to-mid site when the crawl and analysis are automated (15 min crawl, 45 min AI scoring, 30 min human verification). By hand it is a day or more, most of it data collection.

**What is included in a technical SEO audit checklist?** Seven categories: crawlability and indexation, site architecture and internal linking, Core Web Vitals and page speed, structured data and schema validity, on-page fundamentals, mobile and rendering parity, and the AI-engine layer. Each scored pass, warn, or fail with the exact URLs.

**Can you do a technical SEO audit yourself with Claude Code?** Yes, most of it. The crawl, data pull, and first-pass scoring automate with a crawler plus a written rubric. What does not automate is judgment: deciding which flagged issues actually move revenue.

Service: https://www.winstondigitalmarketing.com/services/seo/
Audit: https://www.winstondigitalmarketing.com/contact/#audit