The technical SEO that pays for itself, and the kind that does not

Technical SEO attracts a particular kind of over-engineering. An audit tool returns four hundred issues, the report gets sent over, and three weeks disappear into fixing things that were never costing anything. Meanwhile the product page that matters most is shipping an eleven hundred kilobyte hero image and nobody has mentioned it.

We run these audits too. The difference is in what we do with the list. Here is roughly how we sort it.

Tier one: things that stop pages existing

These are rare and they are emergencies. A page that cannot be found or cannot be indexed earns nothing, no matter how good it is.

• A stray noindex left over from a staging environment. We have found this on live sites three times. It is always a surprise and it is always expensive.
• A robots.txt disallow that blocks a whole section. Usually added for a reason that stopped applying a year ago.
• Canonical tags pointing somewhere unintended — every product canonicalising to the category, for instance, which quietly removes all of them from the index.
• Pages reachable only through a form or a script, with no crawlable link anywhere. Filtered listings do this constantly.

Everything in this tier is checked first, on every site, before anything else is discussed. It takes about an hour and it is the only part of a technical audit that can justify dropping other work.

Tier two: performance, which is mostly images and fonts

Performance work has a reputation for being fiddly and marginal. In our experience it is neither, because the same three problems account for most of it.

Images that were never resized

The most common single fault on the sites we inherit. A photograph exported straight from a camera or a design tool, displayed at a third of its dimensions, in a format from the nineties. On a desktop with a good connection nobody notices. On a phone on mobile data, it is the difference between a page that loads and a page someone leaves.

The fix is dull and enormously effective: serve images at the size they are actually displayed, in a modern format, and let anything below the fold load lazily. On one storefront we worked on, the listing page was carrying several megabytes of product photography that rendered at a few hundred pixels wide. Correcting that alone changed the page more than every meta description on the site combined.

Fonts that block the first paint

Four weights of two families, loaded from a third party, with no display swap, is a surprisingly common setup. The page is ready and the text is invisible while a font file crosses the Atlantic. Cut the weights you do not use, and let the text render in a fallback while the real font arrives.

Layout that moves while you read it

An image without declared dimensions pushes the content down when it arrives. A cookie banner does the same thing a moment later. This is not only a ranking signal; it is the thing that makes someone tap the wrong link and leave. Reserve the space. It costs two attributes.

Tier three: structure, which is slower but compounds

Once pages exist and load, the question becomes whether the site makes sense as a whole.

• Internal linking. The pages you most want to rank should be the easiest to reach. On most sites the opposite is true: the money pages sit four clicks deep behind a mega menu nobody scrolls, while the privacy policy is linked from every footer.
• One clear URL per thing. Duplicate paths to the same content — with and without a trailing slash, with tracking parameters, under two categories — split the signals that should be pointing at one page.
• Heading structure that describes the page rather than decorating it. This matters more than it used to, because it is part of how a machine decides what a passage is about.
• Structured data where it describes something real. Products, articles, local business details, frequently asked questions that are genuinely on the page. Marking up things that are not there is a short-lived trick.

The list we routinely ignore

Audit tools flag these. For most sites under a few thousand pages, they are noise.

• Crawl budget. Unless you are running a very large catalogue, Google is not struggling to get around your site. This gets raised on forty-page brochure sites and it is never the reason for anything.
• Meta keywords. Still, somehow, in reports.
• Long URLs, absent any other problem. A descriptive URL is nicer. It is not a ranking factor worth a migration.
• Chasing a perfect performance score. The gap between a good score and a perfect one is usually spent on diminishing returns while a genuinely slow page elsewhere goes untouched.
• Low text-to-code ratio, as a standalone metric. It measures almost nothing on a modern site.

A note on rebuilds

Sometimes the honest technical recommendation is that the platform is the problem — that the theme carries so much unused weight, or the plugin stack has grown so tangled, that fixing it costs more than replacing it. We say that when it is true, and we are careful about saying it, because a rebuild is where rankings most often get lost.

If you do rebuild, the non-negotiables are a complete URL map with redirects for every page that had traffic or links, a pre-migration crawl kept somewhere you can compare against, and a staging site that is genuinely blocked from indexing. Most migration disasters are one of those three being skipped under deadline.

What good looks like

A technical baseline is not a project that finishes. But it does reach a state where the remaining items are genuinely marginal, and at that point the honest thing for an agency to do is say so and move the budget to content, digital PR, or the local work that actually moves prominence. An agency still billing for technical SEO in month nine of a brochure site is billing for reports.