Running a technical SEO audit is not hard in the sense people expect. Any crawler will hand you four hundred issues in twenty minutes. The difficulty is that perhaps six of them matter, the tool cannot tell you which six, and an audit that does not rank its findings is a document that gets read once and filed.
So this is an order of operations rather than a checklist.
First: can it be indexed at all?
Everything else is downstream of this, and it is routinely skipped in favour of more interesting work.
Start in Search Console's index coverage report rather than in a crawler, because it tells you what Google actually did rather than what a tool thinks should happen. The number to reconcile is simple: how many pages do you believe you have, and how many are indexed? A large gap in either direction is the finding, and it outranks everything below.
Too few indexed means something is blocking — a robots rule, a noindex left over from a staging environment, canonical tags pointing somewhere else, or pages so thin they are being ignored. Too many almost always means the site is generating URLs by itself: filters, parameters, search results, pagination.
Fixing this category first is what makes everything afterwards worth doing. Publishing content onto a site that is competing with itself is a poor use of money.
Second: is what gets indexed the thing you meant?
Crawl the site and compare it with reality. Three specific questions, in this order.
Does every page declare one canonical address, and is it the address you would choose? Self-referencing canonicals on pages that should be canonical; a canonical pointing at the main version on the duplicates. The failure to look for is a page canonicalising to something that no longer exists.
Do redirects go somewhere in one hop? Chains accumulate quietly over years of small changes, and a chain three or four hops long is a page bleeding value at every step. A redirect to a redirect to a 404 is common on any site older than a few years and nobody notices, because a browser follows it silently.
Does the page render without JavaScript? Not as a purity test — as a question about whether your content exists in the initial response. Fetch a page with rendering disabled and read what comes back. If the product description, the copy or the internal links are absent, you are relying on a rendering step that is slower and less reliable than the first pass.
Third: internal links, which is where the leverage is
This is the part tools are worst at and where the most value usually sits.
Find the pages nothing links to. A page reachable only from the sitemap is a page you have told Google exists and told it nothing about. On most sites there is a cluster of these, usually pages built for a campaign and never wired into the navigation.
Then look at the reverse: which pages receive the most internal links, and are they the pages that matter commercially? On a great many sites the answer is the contact page and the privacy policy, because they are in the footer, while the pages the business depends on receive three links each.
And check the anchors. A page linked forty times from forty identical anchors is a pattern; a page linked with varied, descriptive anchors reads as a site written for readers. This is cheap to fix and it is the thing an audit most often fails to mention at all.
Fourth: speed, measured honestly
Use the field data in Search Console, not a lab score. The lab tool measures a machine in a data centre; the field data measures your customers on their phones. They disagree regularly and only one of them has money attached.
When they disagree, trust the field data and go looking for why. Usually it is a third-party script, an image weight problem, or a font loading strategy that blocks the first render — and usually it is the same one or two causes across the whole site rather than a per-page problem.
Fifth: structured data, last and least
It is last because it rarely moves anything on its own and because the failure mode is asymmetric: correct markup is a modest gain, and markup describing things that are not true is a manual penalty. Check that what you claim is real. Reviews you do not have, prices that do not match the page, availability that is stale.
What to hand over at the end
Not four hundred issues. Six, in order, each with what it costs to fix and what you expect it to do. If a finding has no expected outcome attached, it is an observation rather than a recommendation, and the person receiving it will treat it accordingly.
The hardest discipline is discarding real problems that do not matter enough to action. An audit that lists everything is protecting its author rather than helping its reader, and it is the reason so many of them are never implemented. Ours comes back as a short written list in priority order; the ongoing version of this work is described on the SEO page.
