Custom Segment Architecture for YouTube and CTV Buys

Most video accounts we inherit are targeting between twenty and sixty audience segments. We normally hand back six, two of which are exclusions, and the account spends more per segment as a result. A custom segment is a set of intent signals — search terms people typed into Google, URLs, app identifiers — that Google resolves into a reachable audience. The part buyers get wrong is the URL field. A URL pasted there is read as a topic hint: Google looks at what the page is about and finds people interested in that subject. It does not target people who visited that page. We have opened accounts where a competitor's homepage had been sitting in a custom segment for two years under the belief that it was conquesting. How we assemble the term side: We start from query data you already own — Search Console, the search terms report, your own site search — rather than from a session with a keyword tool. A term goes in only if typing it implies a decision has already been made about the category and the choice is now between options. Category-level terms go into a separate segment we expect to lose, because we want the comparison in front of us. Ten to fifteen terms is where these stop improving; past that the signal blurs and Google's own expansion fills the gap on your behalf. Turkish term lists: Turkish is agglutinative. A term list translated from English breaks on the first suffix and the segment under-delivers quietly rather than failing loudly. We build Turkish lists from real query strings, watch the dotted and dotless i in case folding, and check the resulting reach inside the Turkish account rather than trusting an estimate produced against a global corpus. If you have been buying Turkey off a translated list, that is usually the first thing we find. Where affinity still earns its place: On a television screen there is no cookie and often no signed-in individual — there is a household. Tight intent segments on that device frequently return reach too thin to serve at any sensible frequency. Affinity and in-market segments hold up there because they are built on a longer behavioural window and survive the household blur. We use them on the TV device, keep the narrow custom work for mobile and desktop, and label which line in the plan is which. What we do not do: We do not stack expansion on top of a segment we have just narrowed; that reverses the work. We do not leave optimised targeting running while a segment is being evaluated, because it will spend outside the segment and you will not know what you tested. We do not deliver a taxonomy of forty segments for someone else to maintain. And we do not build audiences for token launches or exchanges where the platform's own policy prohibits the advertisement; the audience is not the problem in that case, the account is. Who this is not for: Accounts spending under about 30,000 euros a month on video. Below that the segments cannot reach a decision threshold in a sensible window and you are better served by broad targeting and stronger creative. Also not for teams who want the segment list without the exclusion list, since the exclusions are half of what makes it work. What you receive: A built and live segment set in Google Ads and, where relevant, Display and Video 360; the term and URL sources behind each one; a written kill list naming the segments we expect to fail and the threshold at which we would switch them off; and a rebuild date. Segments decay. We date ours so nobody inherits a two-year-old list and assumes somebody reviewed it.