Audience Signals: Suggestions, Not Targeting

An audience signal is a set of audiences you attach to a Performance Max asset group — customer lists, website visitors, custom segments, interest and demographic groups — as a starting point for the campaign's machine learning. The word doing the work is signal. It is not targeting. Google uses it to find likely converters faster in the early days of a campaign, and then deliberately reaches beyond it.

Most PMax frustration with "audiences being ignored" is this definition being missed. The advertiser keeps editing the signal to "fix" reach; reach keeps ignoring the fix; both sides are behaving exactly as designed.

How audience signals work

When an asset group starts serving, the system has no conversion history of its own. The signal fills that gap: it tells the algorithm where converters have been found before, so early exploration starts warm instead of cold. As real conversions accumulate, the campaign's own data increasingly outweighs the hint you provided — which is why signals matter most in the first weeks and fade in influence afterwards.

What Google promises is exactly that and no more: signals inform learning, they do not cap delivery. Search themes play a companion role on the query side, suggesting search territory the way signals suggest people. Neither is a boundary.

Strength ordering is consistent in practice. First-party data leads — customer lists of people who actually paid, converters, high-intent visitors — because it encodes something Google cannot infer: who your real buyers are. Custom segments built from genuine buyer searches come next. Broad interest categories trail, since the algorithm can already see interest-level behaviour without your help.

Updating signals is a rare event, deliberately. The signal is scaffolding for early learning; once a campaign converts steadily, editing the signal mostly adds noise without adding control. Change it when the business genuinely changes — a new market, a new customer list worth seeding — not as a weekly tuning ritual.

What an agent fixes first: the misreading

The first thing an AI agent fixes here is the misreading itself. Ask what an audience signal actually controls and you get the honest answer above — which prevents the classic loop of editing signals to steer a campaign that is not steered from there. When you want reach changed, the agent can point at the levers that do bind: locations, brand lists, budgets, asset-group structure.

The second is signal construction. An agent with account access can read which of your campaigns and segments actually produce converters and propose signal composition from evidence — buyers, not lookalike guesses. Google's insights reporting shows which audience segments index highly among your converters; pulling that regularly and comparing it against what the signal assumed is a genuinely useful ritual, and a natural part of a weekly PMax review.

The discipline an agent should enforce is restraint: signals set at launch from your best data, reviewed occasionally against who PMax really reached, and otherwise left alone. Steady hands beat busy ones here.

Signals live inside asset groups — defined in asset group — and both sit within the larger practice of managing Performance Max with AI.