A feed audit answers one question: which prompts can this feed not support? Run it as a scored checklist rather than an opinion piece, so it can be repeated next quarter and compared. Five dimensions cover most of it, coverage, specificity, consistency, freshness and disambiguation, each scored out of twenty for a total out of a hundred.
Coverage measures the percentage of records carrying each attribute your clusters reference. Score it per attribute, not as an average: a feed with 98% coverage overall but 40% on price is a feed that loses every budget prompt, and the average hides that completely.
Specificity measures whether values are checkable. Count the fields that hold marketing adjectives where a number or enumeration belongs. Every one of those is a prompt family you cannot satisfy, and converting them is usually the highest-yield repair available.
Consistency compares the feed against the site and documentation on a sample of records. Any divergence in price, availability or a headline spec is a confidence leak. Freshness measures update cadence and whether change is visible and dated. Disambiguation checks variants, bundles and identifiers.
Repair in a fixed order: fill missing high-impact attributes first, convert adjectives to values second, resolve site-feed contradictions third, fix variants fourth, and improve cadence last. That order is deliberate, contradictions are worth fixing before cadence, because a fresh feed that disagrees with the site is worse than a slightly stale one that does not.
Re-run the audit a month after the repairs and put both scores in the client report alongside citation share. The correlation between the two, shown over two quarters, is the most durable argument for continued evidence investment that an agency can produce.