Technical SEO audit:Start a projectwhat to fix first

Every crawler produces the same artefact: a spreadsheet with a few hundred rows, colour-coded by severity, that nobody acts on. The severity column is the problem. It ranks issues by how alarming they sound, not by what they cost you.

A useful audit answers one question per finding: how much traffic is at risk, and what does fixing it take? Two numbers, and the order of work falls out of them.

Fix first: things that stop pages existing

These are rare and they outrank everything else, because no other work matters on a page the crawler cannot reach or index.

Pages blocked from indexing by accident. A noindex left over from staging, a robots.txt rule that was meant to be narrower, a canonical pointing somewhere unintended. We have seen a single stray line remove a third of a site's pages from the index.

Pages the crawler cannot reach. Anything only linked from JavaScript that never runs server-side, pages orphaned when a navigation was redesigned, sections buried more than four clicks from the home page.

Redirect chains and loops. Each hop loses a little, and a loop loses everything.

Check these before anything else, and check them with the search engine's own coverage report rather than a third-party crawl, because that report is what the search engine believes.

Fix second: things that split your own signal

The most common cause of a site underperforming is that it competes with itself.

Several pages targeting one query. Three pages about the same service divide the links, the relevance and the clicks between them, and none of them wins. The fix is usually a merge and a redirect, not more content.

Near-duplicate pages generated by parameters or pagination. Filters, sort orders and session parameters can multiply one page into hundreds. Canonical tags handle most of it; the rest needs the parameters not to generate URLs at all.

Thin pages created for a keyword rather than a reader. These rarely rank and they dilute the site's overall quality signal. Deleting them is an improvement.

This layer is where structure work belongs: deciding which pages should exist, which should merge, and which should go. It is unglamorous and it usually moves more traffic than anything else on the list.

Fix third: speed, and only what is measured

Speed is a ranking factor and, more importantly, a conversion factor — but the lab score everyone reports is not what is counted.

What is counted is field data: what real devices on real networks actually experienced. A site can score 98 in a lab test on a desktop and fail Core Web Vitals in the field because most of its visitors are on mid-range phones over mobile networks.

Work in this order:

1. Largest Contentful Paint. Usually one oversized hero image, an uncompressed font, or a render-blocking script.
2. Cumulative Layout Shift. Almost always images and ads without reserved dimensions.
3. Interaction to Next Paint. Usually too much JavaScript doing work on the main thread.

Each of these has one or two typical causes. Find the cause, do not tune around it.

Fix last: the long tail

Missing alt attributes, imperfect heading order, structured data that could be richer, internal links that could be better distributed. These are real improvements and they belong in the backlog — not at the top of it.

The trap is that they are the easiest items to complete, so they get done first and the audit feels productive while the traffic does not move.

How to know it worked

Set the measurement before the work, not after. For each fix, write down what should change and by when: which pages should return to the index, which queries should consolidate onto one URL, which metric should cross which threshold.

Then report on qualified sessions and revenue rather than on the number of issues closed. An audit that closes 180 of 200 findings and moves no traffic was a tidying exercise.