Reading an AI audit: which findings actually matter

You commissioned an audit and received eighty findings in red, amber and green. Somewhere in there is the thing worth fixing first, and the colour coding will not tell you which it is.

Tool severity scores rank by technical category. A missing alt attribute and a blocked crawler can both appear as "critical" because both belong to categories the tool treats as important. Neither the tool nor the colour knows which of your pages carries revenue.

Three questions that reorder any list

Apply these to every finding and the ordering usually changes substantially.

Blocking issues always come first

A small number of findings make everything else irrelevant. A crawler blocked in robots.txt. Content that only appears after JavaScript execution. Pages excluded from the index. A noindex tag left on from a staging deploy.

These are not degradations, they are disconnections. Content behind them is invisible regardless of quality, and any content work commissioned before they are cleared is spend against a wall.

If an audit does not check these first, that tells you something about the audit.

Cross-reference against revenue, not page count

The most common prioritisation error is treating all pages as equivalent. A tool reporting "400 pages missing meta descriptions" sounds urgent. If 390 are archived blog posts from 2019, it is not.

Conversely, a single "warning"-level finding on the service page that produces a third of your enquiries is likely the most valuable item in the report.

Before prioritising, list your ten highest-value pages. Then read the findings with that list beside you. The ordering will change.

Findings that are usually overrated

• Large counts of trivial issues on low-value or archived pages.
• Meta keyword tags — obsolete, harmless, and reported as findings because they are easy to detect.
• Minor heading-order violations on pages with no commercial role.
• Image size warnings on pages that already load quickly.
• Duplicate title warnings on paginated archive URLs.

Findings that are usually underrated

• Entity contradictions across third-party listings — rarely detected by crawl tools at all.
• Commercial pages with no complete answer near the top.
• Statistics published without a source, on pages meant to establish authority.
• Absence of a recorded visibility baseline, which is a process finding rather than a technical one.
• Speed-to-lead failures, which sit entirely outside what a site crawler can see.
• Untested backups, because the downside is total rather than incremental.

The test of whether an audit is honest

Ask one question: what did it find that you cannot fix, or that the provider cannot help with?

An audit that surfaces only problems matching the seller's service list is a sales document with findings attached. A real audit turns up things outside its own scope — a pricing problem, a capacity constraint, a product gap, a process that belongs to another team.

The presence of those items is the strongest available signal that the diagnosis came before the recommendation, rather than the other way round.

What a good report contains

• A recorded baseline, dated, with transcripts attached.
• Findings ranked by commercial impact, with the reasoning shown.
• A citation for each observation so you can verify it independently.
• An explicit statement of what was not checked.
• A sequence with rough effort estimates, not an undifferentiated list.
• At least one item the provider cannot sell you a fix for.