aeo-score-audit

Use when a visibility score moves unexpectedly, a client wants receipts, or a scan result looks off. Produces an audit trail: aggregate → contributing scans → raw AI answers. Never explain a score movement from the aggregate alone — decompose first.

The three-step drill

1. aggregate — pull the top-level number: visibility percentage, mention count, total scans, average sentiment for the period
2. decompose — explain the score into its component scan results: which scans registered a brand mention and which didn't. Score drops usually localize here — one engine, one country, or one topic did the damage, not "the brand"
3. inspect — for any surprising scan, pull the full record: the raw AI answer text, every extracted field, every citation found. This is where "we lost visibility" becomes "engine X stopped citing the comparison page for prompt Y"

Check the methodology version

Every result carries an analysis version. Before reading a long-term trend, confirm the version is consistent across the window — a methodology change explains a score shift that has nothing to do with the brand's actual standing. Version-boundary comparisons are the classic false alarm.

What the audit is for

• discrepancy resolution — why a scan hit or missed, settled with the raw answer, not a guess
• competitor verification — confirm how rivals actually appeared in the same answers before repeating a claim
• client reporting — an auditable derivation of every headline number; scores that can't show receipts don't survive a procurement review

What good looks like