A hint is a public claim. When an engine quotes your promise it becomes a statement made to a buyer, carrying the same regulatory and reputational weight as ad copy. Agencies that treat the hint library as a scratch pad eventually ship a claim that legal would never have approved, and discover it inside a competitor's screenshot.
Put four controls in place. Ownership: one named person per client library. Review: a fixed cadence, monthly at minimum, where every active hint is re-read against current product truth. Approval: any hint containing a number, a comparison or a compliance claim goes through the client's approver before it ships. Retirement: hints tied to a campaign, price or season have an explicit end date.
Retirement is the control that actually fails in practice. Most exposure in mature programmes comes from stale hints that keep asserting a price, an integration or a guarantee that quietly stopped being true. An expiry field on every hint, with a report of what expires next month, prevents almost all of it.
Version the library in a place the client can see. A shared, dated history of every hint, text, owner, approval status, expiry, turns an opaque optimisation into an auditable programme. It also protects the agency: when a claim is questioned, you can show exactly what was live, when, and who approved it.
Standardise the language across the library. Two hints that describe the same constraint in different words will be read as two different constraints, and inconsistency lowers confidence exactly as contradictions in the feed do. A short controlled vocabulary for recurring constraints, seat bands, regions, deployment types, is worth the hour it takes to write.
Finally, rehearse the takedown. Know how quickly you can pull every hint containing a given claim across every ad group and client, and test it once. When a product change lands unexpectedly, that rehearsal is the difference between an afternoon of work and a month of stale claims running in public.