The Content Audit Framework That Actually Catches Things
Most content audits return a spreadsheet of URLs sorted by traffic, a column for “keep / update / consolidate / kill”, and a project plan nobody runs. The reason the project plan does not run is that the audit asked the wrong questions in the wrong order. A real content audit catches four classes of problem: contamination, cannibalisation, render gaps, and citation gaps. Each class has a different diagnostic, a different fix, and a different cost. Conflating them produces the unworkable spreadsheet. This piece walks the four-layer framework, the audit pattern that catches each layer, and the numbers from a 648-page steel exporter where 2,081 contamination issues were surfaced before a single keep-or-kill decision was made.
The Four Layers, In Order
The framework runs in a strict sequence because each layer’s findings change the inputs to the next.
Layer one is contamination. Does the page accurately describe what the brand actually sells, in language that does not mislead the visitor or the retrieval crawler? Contamination findings include competitor brand names used as the client’s own product names, fabricated price guarantees or warranties, installation content on a supplier site that does not install, and category labels that no longer match the underlying product taxonomy. Until contamination is cleared, the rest of the audit is operating on a substrate the brand cannot defend in front of a customer.
Layer two is cannibalisation. Where two pages on the same property compete for the same query, both lose to a competitor with a single focused page. Cannibalisation findings include duplicate product pages from category sprawl, blog posts targeting the same head term, and location pages that fan out without differentiation. The fix is consolidation, redirect mapping, and canonical reassignment.
Layer three is render gap. Does the page expose its primary content to the retrieval crawler before JavaScript executes? On a client-side-rendered surface, the pre-JS HTML may carry almost no content, and the page will lose to a competitor whose HTML is server-rendered. Render-gap findings sit downstream of contamination and cannibalisation because there is no point fixing the render layer of a page that should be merged or rewritten.
Layer four is citation gap. Are the pages that should be cited by AI surfaces actually being cited? Findings include thin author bylines, missing schema, weak entity binding, and answer text buried behind tab UIs or accordions. Citation findings build on a clean substrate. Running this layer first against a contaminated, cannibalised, render-broken property produces noise.
Layer One: Contamination, In Practice
On the 648-page steel exporter we audited, the contamination scan returned 2,081 issues across the property. Before the report was useful, the issues had to be triaged. 727 false positives in the DIY-installation category were removed (the brand mentioned installation generically without claiming to install). 40 comparison-context false positives were removed (genuine competitor mentions inside comparison pages). 60 fabricated false positives were removed. What remained, after triage, was a tractable list of real contamination findings, 80 plus in number, that the brand could either remove, rewrite, or annotate.
The substantive findings were instructive. Competitor brand names appearing as if they were the client’s own product lines. Price-match guarantees on archived promotional pages that had never been retired. A 74.7-percent-converting product category (Corodek Roof Sheeting) hidden behind a 21.1-percent-converting category (Insulated Panels) that was losing traffic. The contamination scan was not just a hygiene exercise. It surfaced the commercial structure of the property in a way the brand’s own team had stopped seeing.
Layer Two: Cannibalisation Has Two Shapes
The first shape is template-driven. A category page, a sub-category page, and a tag archive all targeting the same head term. The Search Console queries on each show overlap, and the page rank cycles between them on different crawls. The fix is structural: pick one canonical page, redirect or noindex the others, and confirm the underlying CMS does not regenerate the duplicates.
The second shape is editorial. Two blog posts targeting the same query, written months apart, where the newer post was supposed to refresh the older but the older was never redirected. We have surfaced this pattern on every long-running content programme we have audited. The fix is content consolidation: the better of the two posts is kept, the weaker is folded in via a content merge, and a 301 redirect points the weaker URL to the survivor.
On the wealth-management RFP engagement, the cannibalisation audit identified that the original IA blueprint had ~38 percent of pages out of scope (targeting LOB subdomains that the engagement did not cover), and within the in-scope content, multiple pages competed for the same head terms. The replacement Business Loan focus surfaced because the existing Home Loan focus was both out of scope and partially cannibalised by a separate LOB subdomain. The audit changed the brief.
Layer Three: Render Gap, A Specific Diagnostic
The render-gap audit on an Angular 17 SPA fintech we worked with returned a pre-JS word count of approximately 1 word and a post-JS word count of approximately 1,200 words across the stock pages. Robots.txt and sitemap.xml were both serving Angular SPA HTML because of a Router intercept bug. Zero Open Graph tags and zero Twitter Cards across all 3,677 pages. The 50-page Playwright audit confirmed that the H1 was present on 50 of 50 pages post-JS, while a Screaming Frog crawl reported 748 missing H1s. The distinction was load-bearing: this was a render-gap issue, not a template fix.
The audit pattern that distinguishes the two is straightforward. A standard SEO crawler reports what is in the pre-JS HTML. A Playwright or rendering crawler reports what is there after JavaScript executes. Where the two disagree at scale, the property has a render gap, and the fix is server-side rendering or pre-rendering, not template edits.
Layer Four: Citation Gap, The 2026 Addition
The citation-gap audit is the newest layer in the framework. The diagnostic is an AI visibility test against a representative prompt panel. On the 25,000-page NBFC engagement, a 300-prompt panel across ChatGPT, Google AI Overview, and Google AI Mode returned mention rates of 8 percent, 15.6 percent, and 19 percent respectively. The variance across engines was the diagnostic. The lower number on ChatGPT reflected the trust prior penalty from on-site contradictions (the contamination layer). The higher number on AI Mode reflected the index hygiene that Google had built upstream. The same content was scoring differently on different engines, and the gap was attributable to property-level issues the earlier layers had already surfaced.
The Audit Output
Four-Layer Audit Sequence
| Layer | Diagnostic | Fix shape |
|---|---|---|
| 1. Contamination | Full-property scan for brand-mismatch, false claims, taxonomy errors | Editorial remove or rewrite |
| 2. Cannibalisation | Search Console query overlap by URL; template duplicate scan | Consolidate, 301, canonical reassignment |
| 3. Render gap | Pre-JS vs post-JS crawl delta on a sample of templates | SSR or pre-rendering implementation |
| 4. Citation gap | 300-prompt panel across ChatGPT, Claude, AI Overview, AI Mode, Perplexity | Author bind, schema, answer-block restructuring |
Run layer one first. The fixes in later layers depend on a clean substrate from the earlier ones.
What the Output Should Look Like
A useful content audit produces three artefacts. The first is a contamination list with triage status (real, false positive, requires brand review). The second is a consolidation map with source URL, target URL, redirect status, and CMS implementation note. The third is a per-template render and citation report, with the pre-JS, post-JS, and AI-citation observations on a representative sample of each template. The “keep / update / consolidate / kill” decision sits on top of these three, not in place of them.
On the steel-exporter engagement, the four artefacts together drove a 29-sheet Excel with sprint-phased remediation. On the NBFC engagement, the same shape produced a 16-sheet Drupal-and-Akamai-specific fix roadmap. The audit format scales with the property. The four layers do not.
Practitioner Takeaway
- Run layer one before opening the spreadsheet. A full contamination scan before any keep-or-kill decision. The shape of the property usually changes the inventory.
- Use Search Console query overlap by URL to find cannibalisation, not just title or H1 similarity. The query data tells the truth about what is competing.
- Audit pre-JS versus post-JS on a sample of every template. One render-gap finding can invalidate twenty page-level recommendations.
- Run the citation panel last. The variance across engines is the diagnostic, and the panel results are noisy on a property that has not yet cleared the earlier layers.
- Triage false positives before reporting. The 727 DIY-installation false positives on the steel exporter would have buried the substantive findings.
The full audit pattern, including the 8-stage Pydantic pipeline we ran on the steel exporter, sits inside the SEO engineering service. The citation-layer addition is documented in the AI visibility audit. Sector-specific applications appear in the manufacturing growth engineering write-up.
Frequently Asked Questions
How long does a four-layer audit take on a mid-sized property?
On a 500-to-1,500 page property, four to six weeks for the full sequence, with the contamination and cannibalisation layers running in parallel during the first two weeks. Larger properties scale roughly linearly because the AI-citation panel and render-gap diagnostics are sample-based, not full-property.
Can we skip the contamination layer if we know the content?
The 80-plus substantive contamination findings on the steel exporter were on a property the brand team thought they knew. The scan is cheap relative to the cost of acting on a contaminated substrate. We have not yet run an audit where a careful contamination scan returned nothing useful.
What is the difference between cannibalisation and content overlap?
Cannibalisation requires that two pages compete on the same query in a way that materially splits clicks or rankings. Content overlap is a softer signal that two pages discuss similar topics without necessarily competing. The Search Console query overlap data distinguishes them.
When does the citation-gap layer warrant a full-property fix versus a sample fix?
For YMYL properties and any brand that depends on commercial visibility on AI surfaces, full property. For brands where AI surfaces are not yet a material referral source, a sample fix on the top 50 traffic and 20 commercial URLs gets most of the lift while the broader programme catches up.
Run the Four-Layer Audit
For brands with a content programme that has accumulated 500 or more URLs, our four-layer audit walks the property through the sequence above and returns a sprint-phased remediation plan with engineering and editorial specifics.
{
“@context”: “https://schema.org”,
“@graph”: [
{
“@type”: “Article”,
“headline”: “The Content Audit Framework That Actually Catches Things”,
“description”: “A four-layer audit framework: contamination, cannibalisation, render gap, citation gap. Each layer has its own diagnostic and fix shape, run in sequence on a substrate that gets cleaner with each pass.”,
“author”: {
“@type”: “Organization”,
“name”: “ScaleGrowth Digital Editorial”,
“url”: “https://scalegrowth.digital/about/”
},
“publisher”: {
“@type”: “Organization”,
“name”: “ScaleGrowth Digital”,
“logo”: {
“@type”: “ImageObject”,
“url”: “https://scalegrowth.digital/logo.png”
}
},
“mainEntityOfPage”: “https://scalegrowth.digital/content-audit-framework/”,
“datePublished”: “2026-09-14”,
“dateModified”: “2026-09-14”
},
{
“@type”: “FAQPage”,
“mainEntity”: [
{
“@type”: “Question”,
“name”: “How long does a four-layer audit take on a mid-sized property?”,
“acceptedAnswer”: {
“@type”: “Answer”,
“text”: “On a 500-to-1,500 page property, four to six weeks for the full sequence, with the contamination and cannibalisation layers running in parallel during the first two weeks. Larger properties scale roughly linearly because the AI-citation panel and render-gap diagnostics are sample-based, not full-property.”
}
},
{
“@type”: “Question”,
“name”: “Can we skip the contamination layer if we know the content?”,
“acceptedAnswer”: {
“@type”: “Answer”,
“text”: “The 80-plus substantive contamination findings on the steel exporter were on a property the brand team thought they knew. The scan is cheap relative to the cost of acting on a contaminated substrate. We have not yet run an audit where a careful contamination scan returned nothing useful.”
}
},
{
“@type”: “Question”,
“name”: “What is the difference between cannibalisation and content overlap?”,
“acceptedAnswer”: {
“@type”: “Answer”,
“text”: “Cannibalisation requires that two pages compete on the same query in a way that materially splits clicks or rankings. Content overlap is a softer signal that two pages discuss similar topics without necessarily competing. The Search Console query overlap data distinguishes them.”
}
},
{
“@type”: “Question”,
“name”: “When does the citation-gap layer warrant a full-property fix versus a sample fix?”,
“acceptedAnswer”: {
“@type”: “Answer”,
“text”: “For YMYL properties and any brand that depends on commercial visibility on AI surfaces, full property. For brands where AI surfaces are not yet a material referral source, a sample fix on the top 50 traffic and 20 commercial URLs gets most of the lift while the broader programme catches up.”
}
}
]
}
]
}