Native Headline and Thumbnail Testing for Italian Supply

What a call actually rests on: A thumbnail can be judged on click-through rate at roughly three thousand impressions per variant. A headline cannot. We call headlines on post-click behaviour at around two hundred sessions per variant, and in our test logs the two verdicts disagree about a third of the time. Treating click-through rate as the answer is the most expensive habit in this channel: the headline that wins the click is frequently the one promising something the article does not deliver, and the widget algorithms will scale it happily until the account's conversion rate falls over. We do this as a creative cell working alongside whoever buys the media. You do not have to move the buying to us, and we would rather you did not move it mid-quarter. What gets produced: Masters at 1200 by 800, cropped per network rather than stretched: 1000 by 600 for Taboola, 492 by 328 for MGID, and the tighter Revcontent and AdNow ratios. Every image is reviewed at 250 pixels wide on a phone, because that is how it will actually be seen. One subject, one focal point, no product shot with text baked in, no logo below the size at which it is legible. Faces survive that test. Interiors and dashboards usually do not, unless shot close. Headlines come in sets of six against four images, run as eight selected combinations across two rounds rather than as a full grid. The grid is a way of spending a month learning what an experienced writer knows in an afternoon. Italian register: Tu for energy retail and used vehicles. Lei for property above the mid-market, for its adjacencies in private banking, and for business energy. Regional phrasing gets removed rather than added, because a Milanese turn of phrase costs you the south. Numbers stay in the headline where the number is the reason to click: a price band, a subsidy figure, a completion date. Claims your legal team cannot substantiate do not go in at all, and we will not write a headline shaped like a news story. That keeps the work on the right side of IAP article 7 on identifiability and, in practice, on the right side of the widget's own quality scoring. Fatigue: A winning native creative in these sectors loses roughly a quarter to two fifths of its click-through rate by day ten at the frequency these networks buy at. So the cadence is fixed rather than reactive: a refresh set every fortnight, with the previous winner kept running as control. Nothing is retired because of one bad day, and nothing is extended because a stakeholder likes it. What arrives each cycle: The creative set in every required crop, the headline matrix with the reasoning, a naming convention wired into your reporting so the test log survives a change of buyer, and a written verdict on each test, including the tests that produced nothing, which is about half of them. Not included: No video production, no CGI or 3D vehicle work, no property photography. We work from your asset library or a stock licence you hold. We do not write the destination article under this engagement either. That sits inside the buying services, and a testing programme aimed at a page we did not write only tells you which headline the page fails to honour. Not for you if: Creative is approved by committee on a monthly cycle. A fortnightly refresh cannot survive a four-week approval chain, and we have stopped pretending it can.