How to Run a Content Quality Audit After a Core Update

Why page-level audits stopped working

Content audits used to be page-level triage: find the underperformers, refresh them, watch the winners keep winning. The May 2026 core update broke that logic. As we documented in our analysis of the quality-tier mechanism, Google now assesses the publishing site as a whole and lets individual pages compete only within the visibility headroom that site-level band allows. The practical consequence for auditing: a page can be excellent and still fall, because the assessment holding it back was computed from hundreds of pages nobody has looked at in years. An audit that starts from your most important pages is therefore starting at the wrong end. The pages that decide your tier are, almost by definition, the ones without an owner — the archive, the tag sprawl, the blog posts shipped during a volume push in 2023. This audit inverts the traditional order: inventory everything first, find the sections dragging the aggregate, and only then decide where improvement effort lands.

The second thing that changed is the evidence bar. Post-update winners in our eight-market tracking share a trait beyond originality: their quality is legible — named authors, verifiable claims, first-hand data, sources. An audit that scores pages only on traffic misses this dimension entirely, which is why the scoring model below weighs evidence as heavily as performance.

Stage one: build the full inventory

Start with a complete list of indexable URLs from three sources cross-referenced: your XML sitemap, a full crawl, and the pages Google actually knows about from Search Console's indexing report. The gaps between the three lists are findings in themselves — crawlable-but-not-in-sitemap URLs usually mark forgotten sections, and indexed-but-not-crawlable ones mark technical debt. For each URL, pull twelve months of clicks and impressions, referring domains, publish and last-modified dates, and word count. None of this requires paid tooling: Search Console exports plus a crawler cover it. The deliverable is a spreadsheet where every row is a page and the columns let you sort by contribution — because the next stage is segmentation, and segments are where site-level problems hide.

Segment by directory and template before anything else. In the post-update audits we have run, the site average almost always concealed the real story: one or two sections in steep decline pulling an otherwise stable site across the tier boundary. A services site we audited in June showed a healthy-looking 12% overall decline that decomposed into stable service pages, growing case studies — and a 700-URL tag archive down 60% that made up over half the site's indexable footprint. The average was noise; the segment was the diagnosis.

Stage two: score every page on four dimensions

Scoring turns the inventory into decisions. We score each page one to five on four dimensions, and the discipline is scoring honestly rather than aspirationally. Originality: does the page contain anything that did not exist in the index before it — original data, first-hand testing, documented results, genuine editorial judgement? A five means a competitor could not produce this page in an afternoon; a one means they could generate it with a prompt. Evidence: are claims attributed, sourced and dated? Named authorship, reviewed-by attribution and primary citations score here — the legibility the update rewards. Effort signals: depth relative to the query, media that adds information rather than decoration, maintenance history visible in the content. Intent match: does the page answer what its target query actually asks, in the first screen, or does it answer an adjacent question the writer preferred? Pages scoring under ten of twenty are your subtraction candidates; over fifteen, your promotion candidates; the middle band is where rewrite budget goes.

Two calibration notes from running this at scale. First, structure quality correlates with retrieval far more than it used to — answer-first sections under question-shaped headings are what AI surfaces quote — so a well-structured three can outperform a rambling four. Second, metadata is part of intent match: a page whose title promises something the content does not deliver trains both users and quality systems to distrust the domain.

Stage three: the keep / improve / consolidate / remove decision

The proportion matters more than the individual calls. Sites that recovered fastest in our post-May tracking cut or consolidated 20–40% of their indexable pages in the first month. That number frightens teams — until you reframe what those pages were doing: each one was a vote against the domain in an aggregate assessment. Deleting a page that adds nothing is not losing content; it is removing evidence against yourself. The parallel task while pruning is fixing what remains: the recurring technical and on-page errors that read as neglect — broken internal links, orphaned pages, duplicate titles — get swept in the same pass, because effort signals are scored sitewide too.

A worked segment: reading one section honestly

Here is the scoring applied to a real-shaped segment from a June audit: a 140-post "industry news" blog category on a B2B site. Traffic contribution: 3% of organic clicks, concentrated in four posts. Originality scores: median two — the posts summarised announcements already covered by trade press, faster and better. Evidence: threes — competently sourced, but nothing first-hand. Intent match: twos — the queries these posts nominally targeted were owned by the primary sources. The honest verdict was that 136 of 140 posts were tier ballast, publishing effort spent manufacturing evidence of mediocrity. The action plan: keep the four earners, consolidate the twelve with usable passages into two evergreen explainers, remove the rest, and redirect the category's publishing cadence into the evergreen cluster that was actually earning. The uncomfortable part of every audit is this conversation — the section someone fought to build, measured and found to be net negative. Budget for that conversation; it is where audits stall.

The 90-day execution sequence

Days 1–20: inventory and scoring, exactly as above — resist starting fixes before the full map exists, because early fixes anchor on the wrong pages. Days 21–50: subtraction at scale — consolidations first (they need editorial care), then removals, then the technical sweep; update the sitemap and let the crawl digest the new, smaller footprint. Days 51–90: reinvestment — rewrite the improve-band pages in priority order (weakest segments first, not highest-value pages first, because the tier is computed from the weak end), and make expertise legible on everything that survives: authorship, review attribution, sourcing, dates. Throughout: hold the publishing bar at the new standard, because a quality tier is a moving average of everything you ship, and the fastest way to waste the audit is to resume volume publishing in month four. Expect measurable movement at the next core update cycle — that is the honest timeline — with leading indicators (crawl frequency, impression recovery in cleaned segments) visible earlier.

Frequently asked questions