Custom Segments

Custom Segments work by letting you hand Google your own list of inputs - keywords or phrases someone might search, specific website URLs, or app package names - and Google assembles an audience of users whose behavior matches those signals, using the same underlying machine-learning matching that powers its prebuilt audience categories. This is functionally the successor to two now-retired options, custom affinity and custom intent audiences, which Google folded into a single Custom Segments tool so advertisers were not choosing between two overlapping setups.

It is available on Display, Video, and Demand Gen, built once in the shared audience manager and then reusable across campaigns, similar to how a remarketing list gets built once and attached wherever needed. There are three input types and they behave differently: keyword-based segments read like the closest analog to Search keyword targeting outside of Search itself, matching people who have searched those terms or visited sites about those topics recently; URL-based segments match people who visit sites similar to the URLs you list, useful when you know your competitors' or category's core sites but don't have a clean keyword list; app-based segments match users of similar apps, mainly relevant for app-install or app-engagement campaigns.

The setting it interacts with most is input purity, and this is where most accounts go wrong: mixing keywords, URLs, and unrelated concepts into a single segment blurs the signal Google builds against, producing an audience that matches none of the inputs particularly well. A segment built from ten tightly related keywords about one product category will typically outperform a segment built from thirty keywords spanning several loosely connected categories, because the matching model has a cleaner pattern to generalize from in the first case and a muddled one in the second.

It matters most when you have a clear, narrow audience concept that Google's prebuilt In-Market or Affinity categories don't cover precisely - a niche professional group, a specific competitor's customer base, a very particular hobby subculture - and you have enough real signal (actual search terms people use, actual URLs they visit) to describe it well. It matters least when a prebuilt in-market category already covers your concept closely, since building a redundant custom segment usually just splits an already-thin audience into two thinner ones competing for the same budget. It is also worth checking your Search campaign's own search terms report before building a keyword-based custom segment from scratch, since the real queries already converting on Search are often the cleanest possible input list for a new segment rather than a list assembled from guesswork.

In reporting, monitor a new custom segment's size estimate before launch and its delivery volume after, since narrowly built segments (especially URL-based ones with a short list) can come back too small to serve reliably, a common early failure mode. Once live, compare a custom segment's performance against your other audience layers in the same campaign's Audiences tab rather than judging it in isolation, and revisit the input list periodically - a segment built from six-month-old search terms can drift out of relevance the same way a stale keyword list does.