How to Audit a Corporate Website in the Right Order

Start By Proving The Data Is Real

Verify the analytics before you look at anything else, because every finding downstream gets weighted by traffic and conversion figures that may be wrong. An audit that prioritises the wrong pages because tracking was broken is worse than no audit.

The checks are mostly mechanical: the measurement tag on every template and firing once rather than twice; internal and agency traffic excluded; staging hostnames excluded or in a separate property; each form submission generating exactly one key event; and data retention reviewed, because the default is two months and the standard-property maximum is fourteen, which affects Explorations and funnel reports rather than standard aggregated reports.

Two consent checks matter for low-traffic sites. Confirm how consent mode is configured, and note that GA4's behavioural modelling requires at least 1,000 events a day with analytics storage denied for at least seven days, plus at least 1,000 daily users with it granted for seven of the previous twenty-eight. Most B2B properties never reach those thresholds, so they absorb the full data loss from consent refusal with none of the modelled recovery.

Crawl And Indexation: What Google Actually Holds

Compare three views of the site: what a crawler finds, what the sitemap declares, and what Search Console reports as indexed. The gaps between them are the findings. Tool class: a desktop crawler such as Screaming Frog or Sitebulb, the Search Console Page Indexing report, URL Inspection, and server logs where available.

What to look for: parameter and faceted URLs multiplying into near-duplicates; staging or preview subdomains that got indexed; canonical tags pointing at the homepage from templates that should self-canonicalise; sitemaps listing redirected or missing URLs; pages carrying a noindex directive nobody intended; robots.txt rules blocking CSS or JavaScript the page needs to render; and redirect chains, where Google documents following up to ten hops but does not document what happens beyond that.

A finding here should read like a measurement, not an adjective: 1,240 URLs crawled, 890 indexable, Search Console reporting 2,100 discovered of which 700 sit in "Crawled - currently not indexed," almost all of those being tag archives and pagination. That tells a developer what to do. Check one thing by hand regardless: that every page you sell from is in the index.

Architecture And Internal Linking

Architecture problems on corporate sites are rarely exotic. They are the accumulated result of every department getting the page it asked for.

Measure four things. Click depth from the homepage to each commercially important page, since anything four or more clicks deep is being told by your own structure that it does not matter. Orphan pages, meaning URLs in the sitemap or analytics that no internal link points to. Internal links into each money page from somewhere other than the global navigation, because navigation links appear everywhere and therefore distinguish nothing. And the anchor text actually in use, which on most corporate sites is "learn more" several hundred times.

The pattern worth naming: a resources section grown to several hundred pages, absorbing most of the site's internal links and organic entrances, and linking onward to almost nothing commercial. The content works and then leads nowhere.

A finding here looks like: six of the eleven solution pages sit four clicks from the homepage and receive no internal links outside the mega-menu. The remedy is cheap, which is what makes architecture findings rank highly.

Core Web Vitals, Measured Correctly

There are three Core Web Vitals, and a great deal of published advice still lists a retired fourth. The current set is Largest Contentful Paint, documented as good within 2.5 seconds of when the page first starts loading; Interaction to Next Paint, at 200 milliseconds or less; and Cumulative Layout Shift, at 0.1 or less. Interaction to Next Paint became a stable Core Web Vital on 12 March 2024, and First Input Delay support ended on 9 September 2024. Time to First Byte, First Contentful Paint and Total Blocking Time are supporting diagnostics, not Core Web Vitals.

Two facts change how the audit runs. Assessment happens at the 75th percentile of page loads, segmented separately across mobile and desktop, so a site can pass on desktop, fail on mobile, and have the aggregate hide it. And Interaction to Next Paint cannot be measured in a laboratory: Google states that tools loading pages without a user cannot measure it. A Lighthouse Performance score of 100 therefore does not mean a page passes.

Field data also requires sufficient real users, so many B2B pages have only origin-level data or none. Tool class: Search Console and field data first, Lighthouse only for diagnosis.

Content Inventory And Page Purpose

Build one row per URL and resist making it elaborate. The columns that matter: URL, template, declared purpose, primary query or job, date of last substantive edit, internal links in, organic entrances over twelve months, conversions over twelve months, and a decision.

Declared purpose should come from a short closed list, because open-ended descriptions defeat the exercise. Four or five values suffice: attract, explain, convert, support, and obligatory. That last category is real and useful — legal pages, careers, investor material — and marking it stops someone optimising a privacy policy.

The decision column is the point of the whole inventory: keep, improve, consolidate into another page, redirect, or remove. Every row gets one. An inventory without decisions is a spreadsheet.

What this typically reveals, uncomfortably: of 412 blog posts, 61 produced any organic entrance in twelve months and nine produced any form submission. That distribution is normal rather than scandalous, and the right response is usually consolidation rather than deletion, merging several thin posts into a page that answers the subject properly with the old URLs redirected to it.

Conversion Paths, Forms, And Tracking Verification

Walk the path a buyer walks, once per product, on a phone, as somebody who has never seen the site. Note where you have to guess what the company does, where a claim appears without evidence, where the price question goes unanswered entirely, and where a page has done its job and offers no next step.

Then examine the forms as instruments rather than as design. Count the fields, identify which are required, and check whether a validation failure discards what was typed. Check whether the thank-you page is indexable, since if it triggers a conversion it should be excluded from search. Check the consent wording against what the privacy policy says, because those two drift apart. Check spam handling, because a form drowning in bot submissions gets ignored by sales, which turns a technical problem into a revenue problem.

Then the step no tool performs: submit a genuine test lead and follow it. Confirm it reaches the CRM with a populated, correct source value, that the routing rule assigns it to someone who still works there, and that a human replies.

Ranking Findings By Business Impact, Not Severity Labels

Automated tools assign severity by how far a page deviates from a rule, which correlates with volume rather than with money. Re-rank everything by business impact before anyone sees the document.

The ranking rule is simple enough to apply in a meeting: multiply the revenue relevance of the affected pages by the share of buying journeys touching them by your confidence the fix works, then divide by effort to ship.

Why The Hundred-Item Automated Report Fails, And What To Ship

Automated audit platforms count deviations from a rule set across thousands of URLs faster than a person can. That is useful, and it is not an audit. Three failures are structural rather than fixable by choosing a better tool.

• They cannot know which pages earn money, so they weight by volume. Twelve hundred missing alt attributes outrank one solution page carrying an unintended noindex.
• They produce a single site health score, a weighted average of items with no common unit, expressed as a percentage to imply precision it does not have.
• They cannot test whether a form works, whether the lead reaches a human, or whether the page answers the question the buyer arrived with.

The deliverable that changes anything is short: one page listing the ten findings that matter, each with evidence, a named owner, a target date, and the outcome you expect. Behind it, an appendix holding everything else. And a stated cadence — technical crawl monthly, the conversion and forms walk quarterly, the content inventory annually. The audit is not finished when the document is delivered. It is finished when the fixes are verified in production.

Frequently Asked Questions