YouTube Hook Testing on the First Five Seconds

Where the spend actually dies: In the skippable in-stream accounts I have taken over since 2023, between 68 and 82 percent of impressions ended before second five. Everything else in the account — bids, audience lists, bidding strategy — moves the cost of a held view less than a six-point change in that one number. The hook is the media plan, and it is the only part of a video ad that can be tested without lying to yourself about what caused the result. How I build the test: One variable per cell, and the variable is the first five seconds. Everything from 0:05 onward is the same edit, the same grade, the same audio bed, the same end card. If the body changes too, the test returns a winner nobody can act on. Cells run as separate campaigns with matched audience definitions and matched budgets, or as a Google Ads video experiment on a 50/50 split where the account structure allows it. Never as four ads rotating inside one ad group: delivery starts concentrating on its own early signal within hours, and the read is gone before anyone opens the report. Nothing is read before a cell has 30,000 impressions and seven complete days, whichever lands later. Seven days because frequency and reach inside a single audience shift across a week, and a Tuesday-only read has flattered a losing hook more than once. What I read, in this order: Twenty-five percent quartile rate is the primary signal, pulled through the Google Ads API rather than the interface, because it is the earliest point at which the viewer has made a decision and the sample is large enough to move inside a week. View rate second. Conversions only where a cell has cleared twenty-five of them, and never as the tiebreaker below that. For owned uploads I also take the absolute retention curve from YouTube Studio and read the first eight seconds frame by frame. The drop is almost never where the client thinks it is. What I will not do: I do not score creative against a checklist and hand back a grade; that is an opinion with a rubric stapled to it. I do not generate variant sets and let the auction sort them, because a loss nobody can explain teaches nothing and repeats next quarter. I do not read a hook test on brand lift, since survey samples at this budget do not reach significance and the platform will say so if you ask it properly. And I do not change two things at once because the deadline moved. What you get: The hook set ranked with each retention curve, and a recut brief written to frame timings: what is on screen from 0.0 to 0.8 seconds, the first spoken word and where it lands, the on-screen text character count, whether the logo appears before second three and why in your case it should not. Plus the killed hooks with the reason each one died, which is the half most teams find useful three campaigns later. Not included: Production. I write the recut brief, your editor or your agency cuts it. No shooting, no scripting from a blank page, no voice work, no footage sourcing. The media buy itself sits only in the retainer. Who should not book this: Anyone under roughly forty thousand euros a month in YouTube media, because the cells starve and the read is noise. Anyone whose creative sits behind an approval cycle longer than two weeks, because the second round never ships. And anyone who has already decided which hook wins and wants the test to confirm it.