Fix what’s costing you rankings. Worst problems first.

Every check the crawl ran — 43 in all, each read off the pages it applies to or worked out from the crawl’s lists of links, redirects and blocked addresses — with how many pages each one found, the ones that found none included, so a check you fixed reads zero pages rather than vanishing from the list. Each check carries our own judgement of how much it matters — critical, warning or notice — sits under a heading such as crawlability, links or metadata, and is named by its usual term with a plain description beside it, with what it means and how to fix it one click away.

Every page it fetched, by the answer it gave: answered, redirected, missing or refused, a server error, or something else — and the addresses kept out of search results, each with the reason. Each of those figures opens the list of the pages behind it, where a page shows its title, where a redirect goes and the checks it failed, and, for a page that loaded, how far it sits from the home page and how many pages link to it. A missing or failing page opens the broken links that point at it.

The broken links it followed, from the page that carries each one to the address that fails, internal and external told apart, and the redirect chains, each from its first address to its last, hop by hop, with a chain that loops back on itself marked as a loop. A crawl keeps up to 5,000 broken links and 1,000 chains, and a list cut short says how many there were in all.

6 facts about the whole site rather than any one page: whether its certificate is valid and not about to expire, whether the plain address redirects to the secure one, whether an address that does not exist answers as missing, and whether a sitemap and a robots.txt file were found. A fact the crawler did not report reads as not measured, never as a pass.

A health score sits on top of it: the share of the pages a crawl fetched that carry no critical finding, rounded down. It is a share of pages, where the severity counts beside it are a count of findings — one page can carry three criticals — so each figure on that card is labelled with the thing it counts. A crawl that fetched nothing has no score at all rather than a zero, and the card says which of those happened.

What changed, against the crawl before it: for every check, how many pages carry it newly and how many no longer do. A fix is counted per page and per check, so a missing title that has moved to a different page is one fix and one new finding rather than no change at all. A change is only counted between two crawls that ran the same checks, so a first crawl’s findings are findings, not new findings, and a crawl that ran more checks than the one before it prints no change and says so.

• Compare: any two of the recent crawls side by side — their scores, their pages by answer, their broken links, redirect chains and blocked addresses, and every check with both counts and the change between them. A check one of the two did not run reads as not checked rather than as a zero.
• Links: beside the broken links and chains, the pages nothing on the site links to — counted only when the site has a sitemap, which is how the crawler finds them — how many clicks each page sits from the home page, and the thinly linked pages that already earn search clicks.
• Structured data: how many of the crawled pages carry microdata, and how many of them carry an error in it — both counted against the whole crawl rather than one against the other, because the two are independent answers per page and a page can carry an error without the crawler calling it microdata at all. Both are counts of pages: the upstream answer is a yes or no per page, so there is no honest count of individual errors to print.
• A history of the last several crawls, so a score going the wrong way is visible without remembering last month’s number.

After a crawl finishes, 5 of its pages get a Lighthouse run — the ones Search Console says earn the most clicks, with ties settled by URL so a repeat measures the same pages rather than a different sample. Each one reports largest contentful paint, cumulative layout shift and total blocking time, placed on the published good, needs-work and poor thresholds.

This is a lab measurement. A Lighthouse run is a synthetic page load from a machine in a data centre, not a reading from your visitors’ browsers — so there is no interaction-to-next-paint figure here, because that one needs a real person on a real device to measure. Total blocking time is the lab’s stand-in for responsiveness, and the column is headed with the name of the thing that was actually measured.

The crawler’s own fetch happens to report a layout-shift number as well, and it is deliberately left out of that chart. It is a different measurement taken by a different method, and a chart holding two instruments is a chart whose reader has no way to tell which number came from where.

A page whose own run failed is said to have failed. "Nobody has measured this crawl yet" and "every measurement was attempted and none came back" produce the same blank table otherwise, and only one of them is something to wait for.

A crawl starts from your home page and follows the site’s links and its sitemap, up to your plan’s per-crawl ceiling — 150 pages on Free, 5,000 on Scale — and never past what is left of the month’s audit pages, 150 a month on Free and 15,000 on Scale. A site larger than the ceiling is read as far as the ceiling, starting from the home page each time, and the screen says when a crawl stopped there.

One crawl counts its whole size against the month when it starts, however few pages the site turns out to have, because the crawl is charged at that size when it starts. The screen shows the ceiling and what is left of the month beside the button, so the size is known before you press it.

Crawl when you want to, or switch on a weekly crawl and it runs without being asked. If a month’s allowance runs out in week three, the weekly crawl skips that week and stays switched on: an allowance running out is not you withdrawing the instruction.

A second block on the same screen carries the findings from your last analysis run: up to 10 of the pages you already rank with, each fetched by name, and each finding written out — what it means and what to do about it. That one cannot find a page that does not exist — it only ever looks at URLs that already rank — which is exactly what the crawl above is for, and the screen keeps the two apart rather than merging them into one list.