Home › Guides

Site Health, Plainly: What a Walk of Your Site Really Checks — and the One Button Only You Can Press

Published 17 August 2026

"Site health" gets sold as a score out of a hundred, and the score usually means less than it appears to. This is what the phrase actually covers, what a proper walk of your site can and cannot see, what Google's own console knows that no outside tool ever will — and the single click in this whole trade that no software can do for you.

Every page you publish is read twice. A person reads it the way you wrote it, and a machine reads it the way a filing clerk would — following each link to see where it lands, checking the title against the titles of your other pages, looking for the one line that says what the page is about and the markup that says who and where your business is. Site health is the state of that second reading. When it goes wrong, nothing looks broken to you: the page renders, the photos load, the visitor you watch over their shoulder has a perfectly fine time. The clerk, meanwhile, walked into a door that wouldn't open, filed two of your pages under the same name, and left.

The practical questions behind the phrase come in four sizes, and it's worth having them straight before anyone shows you a score. Is each page working — does it answer when knocked on, does it have a title, does anything link into it? Is each page built properly — headings in order, images described, a stated address for the page that machines can rely on? Does the whole site hang together as a body of work rather than a pile of pages? And does it say, unambiguously, what kind of business this is and where it stands? A "health check" that only asks the first question can hand out perfect marks to a site failing the other three.

A health score is only as good as the list of things it checked — and that list ought to be thorough and clear if it is to be of any help.

Why the scores mislead

We learned this while building Fly Away Paul's own Site Health hub, which is why we're comfortable saying it bluntly. Two of the first sites our walk ever examined scored 100% — and the reason was that our early checks had never once looked at an image's alt text, a heading's order or a title's length. The sites weren't perfect; our eyesight was narrow. A perfect score meant "nothing we happened to look at was broken", which is a much smaller claim than it appears on a dashboard. When we widened the checks, the same score swung too far the other way — a page with one slightly long title was suddenly counted as dirty as a page answering 404 — and that reading was no more truthful than the first.

Both failures point at the same rule for reading any audit, ours included: ask what was checked, and ask what each finding costs. A dead link that stops the crawler cold and a description that runs a few characters over are different orders of problem, and a tool that counts them equally is producing a number rather than a reading. The other line worth keeping from the start: fixing site health removes drag — it stops you losing clicks and authority you'd already earned. It does not, by itself, move you up the rankings; that part is won by the pages themselves. Any tool implying that a tidy site outranks a useful one is overclaiming, and the rest of its report deserves reading in that light.

The list Google shows you that no audit tool can see

Open Google Search Console — Google's free reporting desk for site owners, and worth connecting whoever you are — and sooner or later it will show you a wall of "not found" errors. This week, for one site we look after, that wall was 112 URLs long. Our walk of the same site reported none of them, and the walk was right to stay silent, because not one of those addresses is linked from anywhere on the site today.

What that wall actually is: Google's memory. A crawl — anyone's crawl — can only find what today's pages point at. Google, though, re-knocks on every address it has ever known: the site as it stood two redesigns ago, paths from an old sitemap, links other sites made years back, and a surprising number of addresses Google itself assembled wrongly and now dutifully reports as missing. When we checked that 112-row list, the majority were exactly this — the skeleton of a site that no longer exists, plus phantom addresses that never existed at all. No crawler can reproduce that report, and that's no failing of the crawler — the report describes the inside of Google, not the state of your site.

The important part, which the console's red styling doesn't tell you: most of it is fine. In a best-practice world, any removed page with a true successor would have been redirected at the moment it moved — but for a page that's simply gone, with no equivalent to point at, Google's own guidance is that "not found" is the correct answer, it costs your live pages nothing, and redirecting it anyway (usually to the homepage) creates the "soft 404" problem tools rightly complain about. So the wall isn't a to-do list. Buried inside it are the addresses that deserved a redirect and never got one, and telling those apart from the correctly dead is the actual work.

And be clear about how big that buried portion can be. The bulk minting of ghosts happens at a site rebuild: a redesign launches with a fresh structure, nobody writes the redirect map, and a hundred addresses that spent years earning links and visitors all start answering "not found" on the same day. The site we mentioned above had lived through exactly that — the majority of its wall was old pages with no home on the new site, which means the traffic they had built was lost with them, invisibly, at launch. Of everything in this guide, that is the failure with the most money attached. It is also more recoverable than it looks, because the record of what the old site had never went away — which is what the next section is about.

Reading your site's past without borrowing Google's

If you're wondering whether the answer is plugging your audit tool into Search Console, that was our first instinct too — and we decided against it, because we'd rather Fly Away Paul stand on evidence it can gather itself than ask for the keys to another product. It turns out Google's memory is not the only memory of your old site. The Internet Archive's Wayback Machine has been photographing the web for decades, and its public index holds the addresses your site used to have. So our walk now reads that inventory, knocks on every former address, and sorts the dead ones by the only question that matters — is anything still arriving here? Run after a rebuild, that read is the recovery map for the loss described above: it names which of the retired addresses were still earning, so the redirect map that never got written at launch can be written late. Below is the full field guide — every state a dead or dying address can be in, what the evidence means, and what each one deserves. Tap any row for the why.

A console shows you the wall; a useful reading tells you which three of the hundred deserve a redirect and why, next to the reassurance that the rest are dying as they should.

What only Google can still tell you

Some of what Search Console shows, no outside tool can reconstruct — ours included. It reports Google's decisions: whether it chose a different canonical address for a page than the one you declared, whether a page was crawled but left out of the index, which searches your pages actually appeared for. Those are facts about Google's private ledger, not about your site, and no amount of crawling — ours or anyone's — can observe them from outside. The residue of its 404 memory that no archive holds sits in the same category, though as it happens that residue is made of the addresses nothing ever points at, which is precisely the kind that costs you nothing.

The practical stance we'd recommend to anyone: keep Search Console connected to your site and glance at it occasionally — it's free, it's yours, and it's the only window into those verdicts. Then make the verdicts boring. A site that declares a clean canonical on every page gives Google nothing to choose differently; a site whose sitemap is accurate and whose dead addresses are redirected or correctly buried gives the index nothing to puzzle over. Prevention reads the same as omniscience from the outside, and it's available to everyone.

The one button only you can press

There remains a single lever in this whole trade that no tool can pull on your behalf — not ours, not the expensive suites, none of them. When you've fixed something Google flagged, or published a page it hasn't picked up, Search Console has a button for it: "validate fix" on a flagged issue, "request indexing" on an inspected URL. It nudges Google to come and look sooner than its own schedule would. Google exposes no programmatic way to press it — the click lives in the console's interface and nowhere else, which is almost certainly deliberate, since a button robots could press would be pressed by robots all day long.

Two things follow from that. First, the button moves the inspector's schedule and nothing else — whatever you claimed to fix still has to hold up when Google arrives. Second, a well-kept site rarely needs it. Google finds changes on its own by re-reading your sitemap — one whose dates actually change when pages change is the standing "come and look" signal — and by following links from the pages it visits often, which is why new work linked prominently from your homepage gets found fast. The button is for the exceptions: a page that's clean, listed, linked and still sitting unindexed weeks later, or a fix you'd like off the console's red list before the next monthly sweep. When that's the situation, log in and press it. It takes ten seconds, and it is the one contribution to your site's health that will always be yours.

The button is a doorbell, not a repair — it brings the inspector round early, and the fix still has to be real when they arrive.

Where this lands

Site health, done properly, is less a score to chase than a standing answer to a short list of questions: is anything broken, is anything leaking value you already earned, is the site legible to the machines that decide when to show it — and is what remains on the list actually worth doing, or just red styling on housekeeping? You can hold any tool to that standard, including ours: it should say what it checked, grade what it found by cost, tell you the past is dying cleanly when it is, and admit which facts live inside Google where nobody else can read them.