An SEO Audit That Produces a Fix List, Not a Score

An audit is only worth the time if somebody changes something because of it. The test of a good one is not how much it found — it is how quickly a developer can start on Monday morning.

Why scores fail

A composite score compresses a hundred unrelated facts into one number, which makes it feel like progress and impossible to act on. Worse, it invites optimising the metric: a site can raise its score by fixing forty trivial warnings while the one template that blocks half the catalogue from being indexed stays exactly as it is. Rank the findings by consequence instead, and let the number go.

Pass one — can it be crawled?

• robots.txt does not block anything you need fetched — including the CSS and JS the page needs to render.
• Important pages are reachable by links, not only from a sitemap.
• Redirect chains are collapsed to a single hop.
• Faceted or parameterised URLs are not generating an unbounded crawl space.
• Server responses are stable — no intermittent 5xx under crawl load.

Crawling problems come first because everything downstream is wasted if they are unresolved. A perfectly optimized page that is never fetched is not an SEO problem in any interesting sense; it is a plumbing problem.

Pass two — can it be indexed?

• Canonicals point where you intend, and the target is itself indexable.
• No page carries both a canonical to another URL and a claim to be the primary version.
• noindex is present exactly where it belongs and nowhere else — a noindex left on a template after a staging launch is one of the most expensive single characters in SEO.
• Near-duplicate pages are consolidated rather than left to compete with each other.
• Pagination, filters and sort orders do not each produce their own indexable near-copy.

Pass three — does the page say what it means?

This is where most audits get vague, because it is the part that cannot be fully automated. The useful questions are specific: does the page answer the query it is supposed to own, in its own words, near the top? Does the title describe the page rather than the site? Is there one H1, and does it match what the page is about? Do the internal links pointing at this page describe it, or do they all say "learn more"?

1. 01Take the query the page should ownIf you cannot name one, that is the finding. A page without an intended query is a page nobody planned.
2. 02Read the first screen onlyIf the query is not answered there, the fix is a rewrite of the opening, not more words at the bottom.
3. 03Check the competition for coverage, not lengthWhat sub-questions do the ranking pages all answer that this one does not? Length is a symptom, not a target.
4. 04Check the links inA page with no descriptive internal links is a page the rest of the site does not vouch for.

Pass four — is it fast enough, and stable?

Core Web Vitals — Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift — are worth fixing because a slow, jumpy page loses readers, and that is true independently of any ranking effect. Treat field data from real users as the source of truth and lab tools as a debugging aid; they answer different questions.

The output

Every finding should fit one row: the URL or template affected, what to change, why it matters, and roughly how much work it is. Sort by consequence, not by category — one row that unblocks an entire template belongs above thirty missing meta descriptions, however satisfying the thirty would be to tick off.

Then stop auditing and start watching

A one-off audit ages from the day it is delivered. The genuinely useful shift is to run the crawl-and-index checks continuously and treat new findings as they appear, so the next audit is a short list of regressions rather than another hundred pages.