Half the "technical audits" that land on our desk for a second opinion are 400-row crawler exports where a missing alt attribute sits next to the fact that an entire catalogue is blocked from indexing. The document is formally correct and practically useless, because it never says what to fix first. What follows is the order we run on client projects: everything that removes pages from search first, and the things worth a couple of percent last.

The order that saves weeks

Technical problems form a hierarchy, and a fault at the top invalidates every fix below it. Rewriting title tags is pointless if the page returns a 404 to the crawler, and shaving tenths of a second off LCP is pointless if half the catalogue is canonicalised into the homepage.

1. 1

   Level 1 — can the crawler see the site at all

   robots.txt, the robots meta tag, the X-Robots-Tag header, response codes, availability from different IP ranges. One stray directive here costs more than every other finding combined.

2. 2

   Level 2 — has the site multiplied itself

   Canonicals, filter parameters, www variants, trailing slashes, pagination, language versions. Duplicates do not "lower rankings" — they force Google to choose for you, and it picks the wrong page.

3. 3

   Level 3 — does the structure lead anywhere

   Redirect chains, broken internal links, orphan pages, click depth, and the sitemap measured against the site as it actually exists.

4. 4

   Level 4 — rendering and speed

   JavaScript rendering, lazily injected content, Core Web Vitals, the mobile experience. It matters, but fixing it before levels 1–3 is money on the wrong problem.

Indexation: where sites disappear most quietly

The standard disaster runs like this. Developers stand up staging with indexing disabled, then ship that config to production. The site works, customers browse, forms submit — and for two months nobody notices organic traffic decaying. That is why every audit starts here.

# Common mistake: trying to "hide" pages from the index via robots.txt
User-agent: *
Disallow: /catalog/

# This only blocks crawling. Pages that are linked from anywhere will stay
# in the index with no description — and Google never sees your noindex,
# because it is not allowed to fetch the page.

# Correct: allow crawling and set noindex in the meta tag
User-agent: *
Allow: /
Sitemap: https://example.com/sitemap.xml

That distinction deserves its own paragraph, because experienced teams still trip over it. Disallow controls crawling. noindex controls the index. Blocking a page that is already indexed is the fastest way to leave it in there permanently as an empty snippet.

Duplicates, canonicals and language versions

Duplication is rarely deliberate. It emerges from filters, sort orders, tracking parameters, print views and translated copies. The symptom is always the same: search shows a different page from the one you have been building links to.

• A canonical is a hint, not an instruction. If the pages differ meaningfully, Google ignores it and indexes both; if they are genuinely identical, it is honoured.
• Hreflang must be reciprocal: if page A points to B, B is obliged to point back. One-directional tags are discarded entirely — the most common fault on multilingual sites.
• Canonical and hreflang do not conflict: canonicals work inside one language version, hreflang between versions. A canonical from the German page to the English one breaks both at once.
• Tracking parameters create no duplicates as long as pages carry a self-referencing canonical. Five lines in a template close an entire class of problems.

Redirects and URL structure

Redirect chains accumulate over years: an HTTPS migration, then a catalogue restructure, then someone removes trailing slashes. The result is one external link travelling four hops, with the crawler losing interest around the third.

1. 1Find every chain longer than one hop and rewrite the rules so the old URL lands on the final destination directly. One rule instead of three also removes 300–600 ms of time to first byte.
2. 2Check that your 301s have not quietly become 302s. A temporary redirect left in place for a year means lost link signals, because equity transfers less reliably through it.
3. 3Remove internal links pointing at redirects. Linking to old URLs from inside your own site burns crawl budget daily for no reason.
4. 4Settle on one URL form: lowercase, with or without a trailing slash — applied identically everywhere, including sitemap, canonicals and internal links.

JavaScript rendering and speed

Google renders JavaScript, but does it in a second pass with no guaranteed timing. On a client-rendered site the typical gap between crawl and content appearing in the index runs from several days to several weeks. For a news project that is fatal; for a corporate site it is tolerable. How to read the specific indexing statuses and what to do about each is covered in our piece on indexing problems.

• Test with URL Inspection in Search Console, not your browser — it shows the HTML the renderer received. Any difference from your dev tools is the diagnosis.
• Content injected after a click or scroll does not exist for search. A product description behind a tab must still be in the HTML, even if visually collapsed.
• Internal links have to be real <a href> elements. Navigation built on onClick with no href is a dead end — the crawler simply stops there.
• Server rendering or static generation eliminates this whole category. That is an argument about your stack, not micro-optimisation; the trade-offs are in our piece on website loading speed.

Speed is checked last, not because it is unimportant but because it is the one category where fixes are measurable yet gradual. Thresholds and the order of work are covered in the Core Web Vitals guide; everything that should be closed before launch is in the SEO checklist for 2026.

Frequently asked questions about technical SEO audits

In short

• Audit top-down: indexation, duplicates, structure, rendering. Lower-level fixes do nothing while an upper level is broken.
• Disallow in robots.txt blocks crawling, not indexing. Removing a page from search requires noindex plus permission to crawl it.
• Hreflang only works reciprocally. One-directional tags are ignored and language versions end up competing with each other.
• Redirect chains and internal links pointing at old URLs burn crawl budget daily — and both are an hour of work to fix.
• A good audit ends with ten prioritised tasks, not a crawler export.