The problem this solves
Most site audits are really homepage audits with extrapolation. The tool crawls what it can reach quickly, the reviewer spot-checks the pages everyone already knows about, and the long tail of the site, where thin content, orphaned templates, and stale metadata actually accumulate, never gets read by anyone. The fix list that comes out of that process is honest about the pages it saw and silent about the hundreds it skipped.
The silence is expensive because the long tail is where sites decay. The pricing page gets attention every quarter; the 2023 integration pages, the abandoned landing page variants, and the blog posts with broken embeds get attention never. Search engines and AI assistants read all of it, weigh all of it, and form their judgment of the site from the whole corpus, including the parts no human has opened in a year.
This mission closes the gap mechanically: it discovers every URL the site exposes, reads every one of them as rendered content, and only then writes the fix list. Coverage stops being a claim and becomes a count you can check.
How the mission runs
- Map the full URL inventory. The site mapper walks your sitemap files, follows sitemap indexes, and falls back to link discovery where sitemaps are incomplete, producing the definitive URL inventory. The count is reported up front, so the audit's denominator is explicit before any judgment is made.
- Read every page. The crawler fetches each URL and converts it to clean readable content, capturing titles, headings, metadata, and body text page by page. Batches proceed through the whole inventory rather than a sample, which is the difference between this audit and the spot-check it replaces.
- Layer on the technical audit. The technical audit runs alongside the content read: indexation directives, canonical consistency, structured data, and metadata completeness are checked per page, so a technical defect and a content defect on the same URL land in the same row of the findings.
- Rank the findings into one list. Findings are graded by severity and reach: a template-level defect touching 200 pages outranks a typo on one, and thin pages are ranked by how much search demand their topics carry. The output is one ordered list, so the first item is always the highest-value fix.
- Deliver the evidence file. A CSV lands with one row per page: URL, word count, title and metadata status, technical flags, and the finding that applies. The prioritized fix list references those rows, so every recommendation traces to a page someone can open.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
One prioritized fix list covering the entire site, backed by a per-page CSV of findings. The list opens with the highest-reach defects, names the affected URLs for each, and states what fixed looks like, so the work can be handed to a developer or a writer without a translation step.
Make it yours
- Scope the crawl to one section, such as /blog or /docs, when you want a deep read of a single content surface before committing to the whole site.
- Schedule it quarterly and diff the CSVs: the interesting story after the first run is which findings appeared, which were fixed, and which have been on the list for two quarters.
- Pair it with the pull-request playbook so the mappable technical findings arrive as code changes while the content findings go to the writing queue.
Frequently asked questions
How large a site can this cover?
The mapper handles sitemap indexes and caps discovery at the scale a single mission can read carefully, in the low hundreds of pages per run. Larger sites are covered by scoping runs per section and letting the schedule walk the whole property over a week.
Does it read the rendered page or the raw HTML?
The crawl reads served content converted to clean text, which covers the large majority of sites. For pages that only exist after client-side rendering, the mission can escalate individual URLs to the rendered browser and read what a visitor actually sees.
What makes this better than my SEO tool's crawl report?
Crawl reports enumerate technical attributes; they do not read the writing. This mission does both in one pass, so "this page has no meta description" and "this page says nothing a buyer would cite" arrive as one prioritized judgment made by an agent that actually read the page.