Technical SEO: the four things that actually decide it.

A site can look completely fine and still be unreachable, duplicated, uncrawlable or slow enough that most visitors leave before it loads. None of those are visible from the page itself, which is why they survive for years.

The pattern is consistent. Somebody rewrites the copy, then rebuilds the design, then changes agencies, and the ranking does not move, because none of those touched the thing that was actually wrong.

The most common single cause is duplication. The same page reachable at four addresses, with and without the www, with and without a trailing slash, over both protocols, and nothing telling a search engine which one is the real one. Every one of them splits the same page's value four ways.

The second most common is a page that is quietly told not to be indexed. Usually a leftover from when the site was being built, sometimes a plugin setting nobody knew was there, occasionally a rule left over from a staging site.

Four things decide it: whether pages can be reached, whether they are allowed to be indexed, whether the right version is the canonical one, and whether the site loads fast enough on a phone. Almost everything else on a standard audit is noise.

This is worth saying plainly because the industry does the opposite. An audit tool will hand you three hundred issues, most of which have no bearing on anything, and the length of that list is usually the sales pitch.

A missing alt attribute on a decorative image is on that list. So is a heading used out of order. Both are cosmetic. A canonical tag pointing at the wrong page has cost plenty of businesses a ranking, and it sits on the same list at the same severity.

The four that actually decide it

• Can it be crawled.

  Nothing in the robots file, the firewall or the site's own settings stopping a search engine reaching the page.

• Is it allowed to be indexed.

  No noindex, deliberate or accidental, and it is more often accidental.

• Is there one canonical version.

  One address per page, everything else redirecting to it, and the canonical tag agreeing.

• Does it load on a phone.

  Under about two and a half seconds on mobile data, which most sites do not manage.

Speed matters, but not the score. What matters is how long a person on a phone waits before they can read something and before the page stops moving under their thumb. A perfect score with a five second wait is worth nothing.

The measurement people quote is usually a laboratory number taken on a fast connection, and it can be excellent while the real experience is poor. What is worth watching is the field data, which is what actual visitors experienced.

In practice the causes are almost always the same three things. An enormous hero image nobody compressed, a stack of tracking scripts loading before the content, and a font that arrives late and reflows the whole page after the reader has started reading.

Fixing those three usually does more than any amount of work on the rest of the list, and all three are changes to what an existing page loads rather than a rebuild.

What a redirect chain quietly costs

When a site changes address, or moves platform, the old addresses usually end up pointing at the new ones through two or three hops. Each hop loses a little, and a chain that ends in a page that no longer exists loses all of it. This is worth checking after any rebuild, and almost nobody does.

Crawl the site the way a search engine does, list what is genuinely blocking it, fix those in order of what is costing the most, and then check the fix actually took. Not a three hundred item report.

The crawl is the part that produces facts rather than opinions. It finds the pages that cannot be reached, the ones pointing at the wrong canonical, the ones nobody has linked to from anywhere, and the redirects that go nowhere.

Orphaned pages are worth naming separately because they are so common and so quietly damaging. A page that exists, is in the sitemap, and has no link pointing at it from anywhere on the site is a page telling a search engine that the site itself does not think it matters.

What is genuinely changing in search, and what has not moved at all, is set out in Australian search in 2027.

The order it happens in

• Crawl it properly first.

  Facts before opinions. A real crawl, not a score out of a hundred.

• Fix what is actually blocking.

  Indexing, canonicals, redirects, reachability. In that order, because each one can hide the next.

• Then the speed work that matters.

  Images, scripts and fonts. Not the score.

• Then check it took.

  A fix that was not verified is a fix that might not have happened.