Use seo-audit after a traffic decline, a redesign, or before launching important pages. A single homepage does not reveal the condition of an entire site. The URLs, Search Console exports, and access you provide determine what the agent can actually verify.
Install and invoke
Run this in your project terminal to install only this skill for Claude Code:
npx skills add coreyhaines31/marketingskills --skill seo-audit -a claude-code
Node.js and npx must be available. For Codex, replace -a claude-code with -a codex. In the agent conversation, explicitly ask it to use seo-audit and supply your material. The installation command belongs in the terminal, not in your content brief.
Describe what changed
Start with a bounded scope: the domain, five to ten important URLs, the date the problem began, and preceding changes to templates, routing, or content. Include page and query exports from Search Console when available. Mark inaccessible dashboard information as unknown.
Example brief:
Use seo-audit on the ten pages in urls.csv for example.com.
Organic clicks declined after a redesign. Read migration.md and the GSC export search.csv.
Check crawling, indexing, and canonical URLs before content. For every issue, provide the URL, observation, impact, and verification method.
Do not infer the state of data you cannot access.
First inspect whether robots rules, noindex, incorrect canonicals, or redirects obstruct important pages. Check that sitemaps point to the intended indexable addresses. A blocked primary page generally deserves attention before minor description changes. Diagnose parameter URLs, duplicates, and soft 404s using actual responses and page content.
Make the findings reproducible
Separate observations from recommendations. Performance and mobile findings need a page, test conditions, and a practical consequence rather than a generic checklist. Use examples to explain whether titles and body content match the intended query, and whether internal links make important material discoverable.
Be particularly careful with structured data. Text extraction can remove JSON-LD, and client-side code may generate it only after rendering. Its absence from one Markdown fetch does not prove the site lacks markup. Inspect the rendered page or use the appropriate testing tool. Google also requires markup to represent visible content; passing a test does not guarantee a rich result.
End with three to five prioritized fixes, each identifying where to make the change and how to recheck it. After implementation, revisit the same URLs and monitor indexing and search data over time. A same-day fluctuation is not proof of improvement. If the agent saw only a homepage, label the result as a homepage or sample audit rather than a site-wide conclusion.
Trace a visible problem to the affected pages
Start with one observed failure instead of asking the skill to enumerate every SEO item. If a new English section is not being indexed, supply representative URLs, sitemap, robots.txt, canonicals, language alternates, and Search Console status. Investigate “can it be crawled, can it be indexed, and which URL is selected?” before fine-tuning the copy. A page absent from the index is rarely fixed by a prettier meta description alone.
Request a table of issue, evidence URL or tool output, affected page count, proposed fix, and recheck method. On a bilingual site, inspect both locale pages: do canonicals refer to the intended version, are language links reciprocal, and does the sitemap list accessible URLs? One template error can affect dozens of pages; a weak title may affect only one. Prioritize by scope and consequence, not simply by checklist order.
After a fix, inspect the real page, HTTP response, and indexing report again and record the date. Structured data can be rendered in different ways, so absence from one limited crawler view is not proof it is missing; verify the rendered output. Keep “recommended,” “implemented,” and “reprocessed by the search engine” as separate states. Shipping code is not evidence that rankings have already improved.
Example: investigate English pages being grouped under Chinese canonicals
After a bilingual site adds English guides, Search Console reports that some English URLs have a different selected canonical. Prepare two affected locale pairs, page source, sitemap, and before-and-after deployment records. Ask: “Use seo-audit to inspect HTTP status, canonical, hreflang, internal links, and body language first. Attach each observation to a URL and separate template-wide from page-specific failures. Mark insufficient evidence as pending rather than assuming Google is wrong.”
A useful fix points to a reproducible difference, such as an English canonical that mistakenly names the Chinese page, or site links that always lead to the Chinese version. If the agent starts with keyword density, require it to explain the conflicting index signals first. After changing a template, sample more URL pairs and verify that alternate links and the sitemap use live absolute addresses. Then wait for recrawling; a passing local build is not restored indexation.
Turn findings into fixes, then measure
Matt Minor publicly reported running Corey’s seo-audit, receiving an issue list, and using Directus MCP to change titles, descriptions, page content, and some JSON-LD. He explicitly said the ranking impact was still unknown. Follow that discipline: attach a URL and reproducible symptom to each finding, then track “fixed,” “deployed,” and “search performance changed” in separate columns. Finishing an audit is not itself a traffic result.