How audience signals work
In Performance Max, you do not choose a fixed audience the way you would in a manual campaign. Instead you provide audience signals, your customer lists, custom segments built from search and browsing intent, and other data, and the algorithm treats them as a strong hint about who is likely to convert. It begins its search there, then expands to people the signal resembles. The signal does not cap who sees the ad, it seeds where the machine starts, which is a crucial distinction.
Why signal is not the same as target
This is the most misunderstood part of Performance Max. Advertisers assume an audience signal limits delivery to that audience, then panic when the campaign reaches far beyond it. That expansion is by design. A strong signal, built from real converters and high-intent custom segments, gives the algorithm a good place to start and speeds up learning. A weak or generic signal makes the system explore blindly for longer, which is slower and more expensive. The signal shapes the search, it does not fence it.
Building a strong signal
The best signals come from first-party data, your customer match lists and lists of high-value converters, layered with custom segments describing the intent and behaviours of your ideal buyer. Weak signals lean on broad, generic interests that describe almost anyone. Because the signal seeds learning, the effort invested in a precise, converter-based signal pays back as faster ramp and better efficiency. This is another place where owning and activating first-party data directly improves automated performance.
How iClick uses audience signals
iClick builds Performance Max audience signals from real converters and first-party lists rather than generic interests, because the signal decides how quickly and how well the campaign learns. Customer match lists and high-intent custom segments seed the system with a strong starting point, and the expansion beyond the signal is expected rather than fought. The recurring lesson is that a precise, data-backed signal shortens the expensive learning phase that a vague one drags out.
