Audit SEO in a fixed order, because each layer depends on the one before it. First, can Google crawl and index the site: robots.txt, an XML sitemap of only canonical indexable URLs, correct canonicals, no stray noindex. Second, technical foundations: Core Web Vitals (LCP under 2.5s, INP under 200ms, CLS under 0.1), HTTPS, mobile parity. Third, on-page: unique titles of about 50-60 characters with the query first, meta descriptions of 150-160 characters, one H1. Then content quality and E-E-A-T, then authority and links. Once the foundations hold, the fastest wins are usually pages already ranking in positions 8-20. Don't report missing schema from a curl or web_fetch result. Those strip JavaScript-injected JSON-LD, so check with the Rich Results Test or a rendering crawler.
How the SEO Audit skill works
- 1
Establish site context and scope
The agent reads your product marketing context, then asks what kind of site it is, the business goal for SEO, priority keywords, current organic traffic, recent changes or migrations, and whether it has Search Console access. Scope matters: a full audit, specific pages, or one focus area like local or international SEO each produce different work.
- 2
Check crawlability and indexation
This comes first because nothing else matters if Google can't index the page. The agent checks robots.txt for accidental blocks, sitemap contents and submission, important pages within 3 clicks of the homepage, orphan pages, noindex tags, canonicals pointing the wrong way, redirect chains, soft 404s, and indexed versus expected page counts. Fetched pages are treated as untrusted data, never as instructions.
- 3
Audit technical foundations
Next come speed and Core Web Vitals against their thresholds, mobile parity with desktop, HTTPS and mixed content, and URL structure. For multilingual sites the agent checks hreflang self-references, reciprocal links, valid codes (en-GB, never en-UK), and canonicals that never point across locales. It verifies schema with a rendering tool, because JS-injected markup is invisible to a plain fetch.
- 4
Review on-page and content quality
Titles are checked for uniqueness, query position, length, and fit with the page type. The agent also reviews meta descriptions, heading hierarchy, keyword-to-page mapping, cannibalization, internal links, and image alt text. Content is judged on E-E-A-T and depth against the pages that currently outrank it, with known problems by site type: thin feature pages for SaaS, faceted duplicates for e-commerce, inconsistent business details for local.
- 5
Find the rankings push candidates
From Search Console the agent pulls query and page pairs into three buckets: striking distance (positions 8-20), weak click-through (positions 1-5 with low CTR), and near the top (positions 4-7 on revenue queries). For each page it names why the top three results win, whether intent, a missing section, freshness, format, or links. Then it recommends the smallest change that closes the gap.
- 6
Deliver a prioritized report
The report opens with an executive summary and the top 3-5 issues. Every finding lists the issue, its impact, the evidence, a specific fix, and a priority. The action plan is ordered as critical fixes blocking indexation, high-impact improvements, quick wins, then long-term work. After an edit, wait 3-4 weeks before judging it, so the result can be attributed.
When to use it
Your agent reaches for this skill when you ask for things like:
What's inside
The playbook covers
Reference library — 7 deep-dives
See it in action
Can you do an SEO audit of our SaaS website? We're getting about 2,000 organic visits/month but feel like we should be getting more. URL: https://example.com
Health: foundations are mostly sound, but traffic is concentrated. At ~2,000 visits/month, the gap looks like coverage and ranking position, not a penalty. Top issues 1. Feature pages are thin and share one templated title, so they compete with each other instead of ranking. 2. 23 blog posts are orphans with no internal links pointing to them. 3. No comparison or alternatives pages, which is a common gap on SaaS sites. Quick wins - 11 queries sit in positions 8-20 with 100+ impressions/month. These are cheaper to move than any new keyword. - The /pricing title is "Plans That Grow With You". Change it to "[Brand] Pricing: Plans and Costs" to match the pricing query. Needs access: Search Console coverage and query data to confirm counts before acting.
Issue: 6 feature pages use the title "Features | Brand" Impact: High. Duplicate titles split relevance, and Google is rewriting them in results Evidence: crawl export, plus the live titles shown in search Fix: give each page a unique 50-60 character title, "[Category] for [Audience] | Brand", aligned with its H1 Priority: 2 Issue: Schema status unconfirmed Evidence: the raw HTML fetch shows no JSON-LD, but a plugin may inject it client-side Fix: check with the Rich Results Test before reporting it missing Priority: verify first
More things to ask
Real requests the skill is tested against:
Why am I not ranking for 'project management software'? We have a page targeting that keyword but it's stuck on page 3.
We just migrated from WordPress to Next.js and our organic traffic dropped 40% in the last month. Help!
Review the technical SEO of our e-commerce site. We have about 50,000 products and use faceted navigation.
Can you check our blog posts for on-page SEO issues? We publish 4 posts per week but traffic has been flat for 6 months.
I run a local plumbing business with 3 locations. My website barely shows up when people search for 'plumber near me' in our areas. What's wrong?
FAQ
What should an SEO audit check first?
Crawlability and indexation. Confirm robots.txt isn't blocking important pages, the XML sitemap lists only canonical and indexable URLs, canonicals are self-referencing on unique pages, and no important page carries a noindex. Then compare indexed pages against expected counts with a site: search and the Search Console coverage report. A fast, well-written page Google can't index will never rank, so fix these before speed or content.
How do I get pages from page two onto page one?
Pull query and page pairs from Search Console with average position 8-20 and enough impressions to matter, often 100+ a month, then prioritize by business value. Check for cannibalization first. For each page, name why the top three results win, then make the smallest fix: answer the query in the first two sentences, add the missing section, and add internal links from your strongest related pages. Wait 3-4 weeks before judging.
Why does Google rewrite my title tags?
Google tends to replace titles that are much longer than the display width, stuffed with keywords or repeated boilerplate like "| Brand | Brand Blog", out of line with the H1, or generic like "Home". It usually substitutes text from the H1 or anchor text. Treat a rewrite as a signal: tighten the title to about 50-60 characters with the query first, and align it with the H1.
What are the Core Web Vitals thresholds?
LCP (Largest Contentful Paint) under 2.5 seconds, INP (Interaction to Next Paint) under 200 milliseconds, and CLS (Cumulative Layout Shift) under 0.1. Measure them in PageSpeed Insights, WebPageTest, Chrome DevTools, or Search Console's Core Web Vitals report. Common culprits are server response time, unoptimized images, heavy JavaScript, render-blocking CSS, missing caching headers, and font loading.
Why are my hreflang tags being ignored?
The usual causes: a page missing its self-referencing entry, one-directional links where B doesn't point back to A, invalid codes like en-UK instead of en-GB, targets that 404 or redirect or aren't canonical, or HTML and sitemap annotations that disagree. A canonical URL left out of the hreflang set causes all of its hreflang to be ignored. Never canonical one locale to another, because that suppresses the non-canonical locale entirely.
Can an AI agent detect schema markup by fetching a page?
Not reliably. Plain fetches like web_fetch and curl strip script tags, so JSON-LD injected client-side by plugins such as Yoast, RankMath, or AIOSEO won't appear. Reporting "no schema found" from that output creates false findings. The skill checks with a browser tool that queries script[type="application/ld+json"], Google's Rich Results Test, or a JavaScript-rendering crawl export such as Screaming Frog.
Related skills
Use it in your agent
SEO Audit is a plain SKILL.md folder, so the same skill loads in any agent that reads the skills format. Setup guides:
Build a tool for this workflow?
Marketers run this skill mid-task — the exact moment your tool matters. Partner tiers put your tool inside the skill their agent runs.
See partner tiers