Native Creative and Headline Testing in Czech and English

A native creative test that ends at three thousand impressions has told you nothing. At the click rate normal for Czech-language widget placement, around eight hundredths of a percent, three thousand impressions is two or three clicks, and the variance in that range is larger than any effect worth acting on. We test one variable at a time to a stopping rule agreed in writing before the wave launches, and we will not ship the variant that wins by pretending to be news. Image and headline are different instruments: The thumbnail moves click rate. The headline moves what the reader does after the click. Testing both at once is how a campaign ends up with an excellent click rate and no conversions, and how a team spends a quarter congratulating itself on the wrong number. A wave therefore holds one image against another at a fixed headline, then one headline against another at the winning image, and the results document says which of the two did the work. The Czech decay problem: The Czech publisher pool is small. A winning thumbnail in cs loses roughly a third of its click rate within ten to fourteen days at steady spend, because the same readers are seeing it for the fourth time. In a large English-language market the same asset holds for six weeks or more. Creative cadence here is a structural fact of the market rather than a matter of taste, and any plan that assumes one winner will carry a Czech quarter is already wrong on the day it is signed. What ships: Image and headline pairs as a master at 1200 by 800 plus the network crops, because the same picture cropped to MGID's 492 by 328 and to a four-by-three widget slot is two different pictures and a face survives only one of them. A naming convention your team can still read a year later. A test plan with the stopping rule and the sample size written down before launch. A results document that includes the losers and why they lost, which is the part that stops the next wave repeating this one. What we will not test: Fake interface elements, meaning close buttons that are not buttons, play triangles on still images, imitation system alerts. Before-and-after body imagery. Publisher logos we have no right to use. Health claims in headlines: for foods and supplements Regulation EC 1924/2006 restricts what may be said, and for devices Article 7 of the Medical Devices Regulation forbids claims that mislead about function. We have taken down variants that were winning, and we would do it again on your account. Not included: Buying. The accounts stay in your team's hands and we take reporting access only, which also means the bid strategy is not ours to defend. Video and animated formats, brand identity work, and the landing article. Languages beyond Czech and English, because we test copy we can read, and a translated headline judged by someone who does not speak the language is a guess with a chart underneath it. Who this is not for: Teams who want as many variants as possible each month. That is a volume order, every studio will take it, and the library arrives untested. Also not for anyone who needs the test to conclude by a board meeting date rather than at a sample size, since the only way to do that is to call a winner that is not one.