Running an audit

Starting a Website Audit in RankX AI, the modes available, how long it takes, and how to read the report when it finishes.

Start an audit from the Website Audit screen or over MCP. It is dispatched rather than instant: RankX AI returns a job immediately and the crawl runs in the background over minutes. When it completes, the findings appear with a score, a severity breakdown and the pages each issue affects.

The modes

Mode	What it does	When to use it
Full crawl	Follows your site from its entry points up to the page count you choose	The periodic health check, and the first thing to run on a new Website
Quick check	One page	Confirming a fix, or checking a single important page
Specific pages	A list of pages you name	Re-checking a set you have just changed. It runs a browser, so it sees things a plain crawl cannot

The specific-pages path is worth knowing about because it is not merely a narrower crawl: it renders the page, which means it can produce verdicts for checks the plain crawl structurally cannot. See what "a check did not run" means.

Scheduled audits

A Website can run its audit on a schedule as well as on demand. The cadence has a floor: your plan's, or 7 days while on trial, whichever is slower. Scheduling is in the Website's automation settings, and turning it off there stops the scheduled run without affecting manual ones.

How long it takes

Minutes, and it scales with the page count. There is no useful average: a hundred-page site on a fast host and a hundred-page site behind a slow one are different jobs.

If you are driving it from an assistant, poll rather than wait: the dispatch returns a job id and get_site_audit_status reports when it is complete. A client that reads the dispatch response as the result will report "no issues found" immediately, which is the most common integration mistake on this surface.

Reading the report

Start with severity, not with the score. The score is a summary; the issues are the work. Issues first, warnings second, and the affected-page counts tell you which of them is one page and which is the whole site.

Read the not-assessed list. A large not-assessed count usually has a single cause, most often that the crawl did not reach enough pages, and fixing the cause fills the column. It is not a list of things wrong with your site.

Check the page counts against your expectation. If you know your site has four hundred pages and the crawl reached forty, the report describes forty. See page selection and cost.

Site-wide findings are not diluted. A missing sitemap or a missing robots.txt takes its full weight when it fires, rather than being divided by the number of pages crawled. That is deliberate: the alternative made real findings vanish behind a large page count.

Only the most recent completed audit is reported

The issues view reads the latest completed audit. A crawl that is still running does not partially overwrite the last one, and a crawl that failed does not replace a good report with an empty one.

So a report that looks stale after you started a new audit means the new one has not finished yet, which the status will tell you.

Turning findings into work

Two routes, and both keep the finding attached to its evidence:

Create tasks. Findings become items on the Tasks board, deduplicated so one page does not appear four times, ordered by opportunity rather than by category. A task carries a copy of what it was created from, so it survives the audit being superseded.

Apply the fix, where the Website has a connected WordPress site and the fix is one RankX AI can make. It dry-runs first and shows you the exact before and after; applying is a second, explicit step. See the WordPress integration.

The Task Creation Assistant

For a finding you want turned into a properly written task rather than a one-line note, RankX AI can draft one: what the problem is, why it matters, the steps, and a definition of done. It spends credits and says so before it runs. See the Task Creation Assistant.

Driving this from an assistant

"Start a site audit on my main website. Tell me my credit balance and the cost first, then poll until it finishes and give me the issues grouped by severity, with the not-assessed ones listed separately."

start_site_audit → get_site_audit_status → get_site_audit_issues. See the tool reference.