Technical SEO audit

Goal: a short, ranked list of problems that are real, shown with the evidence (a command result), the source that says why it matters, and the fix. Nothing in the report comes from folklore. A check the tools cannot run is listed under "Not checked", with how the user can check it.

Inputs

• A domain or homepage URL, and optionally 3 to 5 key templates (home, a category, an article or product, a landing page).
• Optional: Search Console access (see the setup skill). Without it, the audit covers what a crawler sees, not what Google has indexed.

Start with a crawl

npx sitetiller crawl https://example.com --max 100 --json

One request at a time (1 s apart by default), robots.txt respected, same host only. Output: score (overall and categories for indexability, onpage, links, content, structured, with the arithmetic in method), findingCounts, siteFindings (duplicate titles and descriptions, broken internal links with the pages that hold them, links to redirects) and pages[] with each page's status, depth and findings. The score is a checklist ratio for tracking progress between audits, not a Google measure; report the findings, not the number alone. Use the steps below to confirm and explain what the crawl found.

Order of work

Run in this order. A blocker at an early step makes later steps moot, so fix it first.

1. Can crawlers get in? (robots.txt and AI crawlers)

npx sitetiller crawlers example.com --json
npx sitetiller crawlers example.com/some/section/ --json

Read status, found, problems, sitemaps, and each bot's allowed, rule, line. Googlebot blocked on a path that should rank is a blocker. For the AI crawler decisions use the ai-crawlers skill. Facts that matter (Google's robots.txt documentation, see the ai-crawlers sources): 4xx on robots.txt means no rules, 5xx pauses crawling. A robots.txt block stops Google reading a noindex on that page.

2. Sitemap

npx sitetiller sitemap example.com --sample 20 --json

Read found, kind, urlCount, inRobots, lastmod (a count of <lastmod> tags), issues, and sample[] (status, noindex, redirectedTo). Every sampled URL in a sitemap should answer 200, be indexable and not redirect. Details: the sitemaps skill.

3. Status codes and redirects on the key templates

npx sitetiller page https://example.com/ --json
npx sitetiller page http://example.com/some-page --json
npx sitetiller page https://www.example.com/some-page/ --json

Test the four variants (http, https, www, non-www) of the homepage. Read chain[] (each hop's status), finalUrl, status. Expectations from Google's documentation:

• One permanent hop to the final URL. Googlebot follows up to 10 hops, but the site move guide advises chains ideally no more than 3 and fewer than 5. The page command follows at most 6 hops.
• 301 and 308 are strong signals that the target is canonical; 302, 303, 307 are weak signals.
• 4xx removes indexed URLs from the index; persistent 5xx eventually does too.
• A "not found" page that answers 200 is a soft 404. Redirect work: the redirects-migration skill.

4. Indexability signals on every template

From the same page output read noindex, robotsMeta, xRobots, canonical, canonicalCount, robotsTxtGooglebot.

• noindex true on a page that should rank: blocker.
• canonicalCount above 1, a canonical pointing to another URL, or to a redirecting or noindex URL: fix (see canonicals).
• robotsTxtGooglebot.allowed false: Googlebot cannot crawl it.
• The tool reads the HTML only. It does not read Link: header canonicals and does not run JavaScript. If the page builds its canonical or robots meta with JavaScript, say so: Google says noindex in the raw HTML may stop rendering, so JavaScript cannot undo it.

5. On-page basics across many pages

npx sitetiller pages example.com --pages 40 --json

Read summary: missingDescription, shortDescription, longDescription, duplicateTitles, duplicateDescriptions, noindex, multipleOrNoH1, invalidSchema, schemaTypes. Per page: title, titleLength, descriptionLength, h1Count, canonical, words.

• Every page should have a <title>. Google puts no limit on title or description length; it truncates to the device width. Duplicate or boilerplate titles make pages hard to tell apart. So report these without a length rule: missing, duplicate, boilerplate.
• The tool's shortDescription (under 70 characters) and longDescription (over 160) are the tool's own rules of thumb, not Google limits. Mark them "low priority".
• multipleOrNoH1 follows a convention. Google's title link guidance suggests the main title in the first visible <h1>; it does not make several H1s an error. Low priority.
• A page the tool marks error: true could not be read; list it.
• Titles may show HTML entities (for example | for |). That is how the page source encodes them.

6. Internal links and orphans

npx sitetiller links example.com --pages 50 --json

Read orphans[] and weak[] (url, linkedFrom, anchors). Google: every page you care about should be linked from at least one other page, with crawlable <a href> links. The tool samples --pages pages, so a page listed as an orphan may be linked from pages it did not crawl. Raise --pages or verify with the live site before reporting. The note field in the output states the sample size.

7. Structured data

npx sitetiller schema https://example.com/some-article --json

Read types and findings[] (level, type, message, doc). The linter covers Article, SoftwareApplication, Product, Event and LocalBusiness and flags FAQPage (Google stopped showing FAQ rich results on 2026-05-07). Markup must describe visible content, and valid markup does not guarantee a rich result. For anything else, use Google's Rich Results Test.

8. Core Web Vitals

npx sitetiller cwv https://example.com/ --json
npx sitetiller cwv https://example.com/ --form DESKTOP --json
npx sitetiller cwv https://example.com/ --origin --json

Needs CRUX_API_KEY. Thresholds and fixes: the core-web-vitals skill.

9. Mobile parity (manual)

Google indexes the mobile version, so the same content, robots meta tags, structured data, titles and descriptions must exist on mobile. The CLI fetches one version. Ask the user to compare the mobile and desktop page, or to use Search Console's URL Inspection.

Findings table (sample output)

Score impact (H/M/L) by effort (H/M/L). Order: high impact and low effort first. Never invent impact; "H" means the page cannot be indexed or found, or many pages are affected.

Output to the user

1. Verdict in two sentences: blockers first.
2. The findings table, sorted, at most 15 rows.
3. "Not checked": Search Console indexing data, JavaScript rendering, mobile parity, server logs, backlinks, content quality. Say how to check each.
4. The first three fixes to do this week.

Guardrails

• Report evidence from commands only. Every claim carries a command result or a source in references/sources.md.
• Do not claim a length, count or ratio limit that Google does not publish (title, description, H1, internal links per page, word count).
• A sample is a sample: state how many pages were read and the total listed.
• Do not edit the user's site. Recommend fixes only; changes belong to the user.
• Do not call something a penalty. Google documents manual actions separately; a missing ranking is not evidence of one.

References

• examples/pages.json, examples/crawl.json: real output of this skill's command on sitetiller.com (2026-11-13), see examples/README.md.
• references/sources.md: Google, web.dev and protocol documents behind each check, with access dates.