YouTube Hook Testing for Skippable In-Stream Buys

We test three hooks, occasionally four, and hold each cell open until it has served roughly 250,000 impressions to the target audience. Below that the five-point view-rate gap that makes everyone want to declare a winner sits inside the noise band, and enough of those gaps have reversed in week three that we stopped calling them early. The measurement constraint nobody mentions: YouTube gives a second-by-second retention curve for organic uploads. It does not give one for advertisements. On a paid video you get quartile completion — twenty-five, fifty, seventy-five, one hundred — plus the skip, which becomes available at the five-second mark on skippable in-stream. A hook test therefore has to be designed around the two signals that exist: who is still present when the skip button appears, and who reaches the first quartile. Anyone showing you a per-second read on ad creative is showing you organic data or a model of it. What varies and what is held: Only the first five seconds differ between cells. Same offer, same voice, same length, same end card, same audio bed, same landing page. When the offer changes as well, you have learned something about the offer and nothing about the hook, and that is the most common way this budget is spent without producing a result. We write and direct the hooks. The body of the film stays with you or your production company; we do not need to own it and we do not charge for it. What we judge on: Skip rate at five seconds first. First-quartile hold second. Cost per completed view third, and only as an input to budgeting. View-through rate on its own rewards hooks that hold the wrong people — a number on screen will keep almost anyone watching for eight seconds. Where the budget supports a Brand Lift study we run one alongside and size the cells to power it. Where it does not, we write that the recall question is unanswered rather than reporting a proxy in its place. Turkish-language hooks: A Turkish hook is not a dubbed English hook. Dubbed lines run long against the picture and the product name arrives after the viewer has already gone. We write in Turkish to picture and place the category and the brand inside the first three seconds, because for a viewer who skips at five that is the whole impression. For food and drink work the Ramadan calendar changes the hook and the hour together: the film that wins at four in the afternoon loses at nine in the evening, so the daypart is part of the cell design rather than something we blame the creative for afterwards. Energy work: Retail supply, solar installation and charging networks share one hook problem. The interesting fact is the bill; the client wants to open with the technology. We test both, and the bill has won often enough in the flights we have run that we now write it as the default and make the technology opening earn its place. What we refuse: Thirty variants. Volume testing on video produces cells too small to read and a report that resembles work. Testing on Shorts and reading the result across to in-stream, which is a different placement, a different posture and a different skip behaviour. Running an experiment with optimised targeting left on, which moves the audience underneath the test. And calling a winner before the threshold, whatever the meeting calendar says. Who this is not for: Teams with one film and no budget to shoot alternatives, because there is nothing to test. Teams who need an answer inside a week. And brands whose video budget cannot carry two cells to threshold without starving the rest of the plan. What you receive: Three to four hooks scripted, directed and delivered cut to your master; the experiment configured in Google Ads with the split at user level; the decision rule written down before launch; and a documented account of the hooks that lost, which is the part that transfers to the next film.