SEO audit methodology
What an SEO audit should prove before you fix anything
A crawler can list errors. A decision-ready SEO audit has a harder job: show where each conclusion came from, why it matters to this business, what should happen next, and how the result will be verified.
Methodology note
This guide describes the EXPERT SEO evidence model. It does not claim that every site needs the same checks or that one audit score predicts rankings.
Start with the decision, not the score
A score can summarize a condition, but it cannot tell an engineer which template to change or tell an owner which constraint is blocking qualified demand. Begin with the decision the audit must support: restore indexation, clarify a service page, strengthen local relevance, consolidate overlapping content, or remove a rendering dependency. Then collect the evidence needed for that decision.
Preserve the source and collection time
A finding should name the provider, property, report, URL, query, market, device, and collection time when those details matter. A screenshot without scope ages quickly. A Search Console query export, crawl result, performance trace, Business Profile record, or backlink observation becomes useful when another person can reproduce the path and understand its limits.
Separate the finding from its impact
“Missing title” is an observation. The finding explains which indexable URL is affected and whether the title is absent, duplicated, misleading, or simply different from a tool’s preferred length. The impact then explains the plausible consequence: unclear page intent, weak search-result language, template duplication, or no material business risk. This separation prevents every warning from becoming an emergency.
Prioritize constraints, not error counts
Hundreds of repeated errors may come from one template. One blocked revenue page may matter more than thousands of harmless parameter URLs. Priority should combine business impact, affected demand, confidence, implementation effort, dependencies, and reversibility. The result is a sequence a team can actually execute rather than a spreadsheet sorted by red icons.
Define validation before implementation
A recommendation is incomplete until it states what will prove the work succeeded. Validation might require a fresh crawl, URL Inspection, a rendered HTML check, field performance data, corrected internal links, improved query impressions, or a clean structured-data test. Record the baseline first, then compare the same evidence after the release.
Ask what remains unknown
Responsible audits expose data gaps. If Search Console is not connected, say so. If a Business Profile API is pending, keep those metrics unavailable. If conversion tracking is absent, do not turn traffic changes into revenue claims. Unknown is a valid state; invented certainty is not.