What asset group signals do
A Performance Max asset group holds the creative for one product or service, and alongside that creative you supply signals: audience segments you think describe the buyer, and search themes describing the phrases you expect them to use. Segments can be your own customer lists and website visitors, or Google’s interest and in-market groupings.
Signals are a starting point, not a boundary. Performance Max uses them to decide where to look first, then keeps exploring beyond them if the conversion data suggests better demand elsewhere. This is the single most misunderstood thing about the campaign type: adding an audience does not restrict who sees the ads, in the way an audience signal is often assumed to. It shortens the learning period by pointing the model somewhere sensible on day one.
Why signals matter
Their value is mostly at the beginning. A campaign starting with no useful signal spends the first stretch of its budget discovering things you already knew, and in a smaller market that exploration is expensive because there is less demand to sample. A campaign starting from a customer list and a set of accurate search themes reaches sensible traffic sooner.
Signals also shape the structure of the account. Because each asset group has its own signals and its own creative, splitting asset groups by product line or margin lets you steer money towards what is worth selling. Signals are one of the few places where an advertiser’s own knowledge of the business — who buys, who never does, which phrasing customers use — enters an otherwise closed system.
Where signals go wrong
The commonest mistake is treating them as targeting and then complaining that the campaign reached the wrong people. Nothing was broken; the signal was a hint and the model looked further. If you genuinely need a boundary, use exclusions, location settings and brand controls, because those are the levers that actually restrict.
The second is stuffing in every audience available. A wide, contradictory pile of signals says nothing, and the model learns as little as it would from none. The third is a stale customer list uploaded once and never refreshed, which points the campaign at the customer you had years ago. The fourth is search themes written as keyword lists — long tails of near-duplicates rather than the handful of distinct ideas the field is designed for.
How to give better signals
Start with first-party data if you have it: recent buyers, high-value buyers, enquirers who converted. That list describes your actual customer better than any interest segment. Add a small number of search themes that cover genuinely different demand — the service name, the problem people describe, the buying phrase — rather than variations of the same words.
Keep asset groups separate where the buyer is different, so the signal in each one can be specific. Refresh customer lists on a regular schedule. Then read the results at the level that shows what happened: search terms, asset performance and channel reporting inside Performance Max, so you can tell whether the model followed your signal or found something better. If it consistently ignores a signal, that is information about the market, and the honest response is to change the signal rather than to keep insisting on it.