Malaysia's Longest-Running Facebook Ads (2026)

The population is every Malaysian ad in the AdPlay.ai archive that was still running the last time we checked that specific ad: 222,921 of them. The cohort is the 6,347 inside that set which had already been running for at least a year, 2.8% of the whole. For each creative label, the index is the label's share of the cohort divided by its share of the baseline, multiplied by 100. An index of 100 means a year-plus ad is exactly as likely to carry that label as any other running ad. Indexing rather than counting is the point. Both groups are drawn from the same archive with the same collection gaps and the same uneven refresh cadence, so most of what is wrong with one is wrong with the other in the same direction and divides out. What does not divide out is right-censoring: an ad is in the cohort because it has not stopped yet, not because it finished a successful run. So this page describes what survives. It does not describe what wins, and nothing here is evidence that a particular angle made anyone money. The archive holds creative metadata only, with no spend, no impressions and no return, so a page that claimed otherwise would be inventing it. One exclusion worth stating plainly. Creative angles are model-assigned free text, and asking the search index for every distinct value it holds returns a list capped at 100 entries in alphabetical order rather than a complete one, so a tail would go missing without anything reporting it. Every angle below is counted explicitly against the classifier's own twenty-label vocabulary instead. Labels outside that vocabulary are excluded rather than half-counted, and an ad can carry several labels, so the shares deliberately do not sum to 100.

Four angles are clearly over-represented among the ads that lasted: testimonial at an index of 157, before-and-after at 151, explainer at 147 and social proof at 143. Every one of them is a way of showing that the thing worked for somebody. Between them they are the closest this dataset comes to a pattern. The mirror image is the discount. It is the most used angle in Malaysian advertising, carried by 44.6% of everything running, and it falls to 36% among the ads that reached a year, an index of 81. That is not an argument against running offers, which plainly sell. It is an argument about shelf life: a promotion is written against a date and a price, so it stops being true, while a customer saying the product worked stays true indefinitely. If you want one ad you can leave running while you iterate on everything else, the evidence points at proof rather than at a percentage off.

Two labels dominate both groups and separate neither. Showcase sits at 70% of running ads and 68.3% of year-plus ads, an index of 98. Feature callout runs 40.5% against 40.2%, an index of 99. Showing the product and listing what it does is what a Malaysian ad is, so neither tells you anything about whether one will last. A few angles sit low on the index with too few ads behind them to lean on. Native content comes in at 74, and the quiz, meme, press and founder-story labels each carry fewer than 100 ads in the cohort, which is a number that moves on a single week of collection. They are all in the CSV with a headline column marking which rows are thin, because a figure you should not quote is still a figure someone should be able to check.

If format decided longevity, this is where it would show, and it does not. Video sits at an index of 101, which is to say exactly the national rate. Dynamic creative reaches 124, static images 83 and carousels 74. The spread from top to bottom is narrower than the gap between a single pair of creative angles, and there is a known contamination inside it: two different write paths disagree about whether a dynamic ad is filed as dynamic or as an image, so the split between those two is a range rather than a point. The placement rows say even less. An ad that lasted a year is eligible on Facebook at an index of 101, Instagram 96, Messenger 109 and Audience Network 112. Almost every Malaysian advertiser leaves placements on automatic, so this measures a default rather than a decision, and eligibility is not delivery in any case.

The dataset contains two rows that look like dramatic findings and are not, and it is worth saying why rather than quietly deleting them. Threads shows an index of 19 among year-plus ads and WhatsApp shows 0, against 41.7% and 25.1% of all running ads respectively. Meta added both placements to the delivery options during the period this cohort has been running. An ad created before a placement existed cannot be eligible for it unless someone reopens and edits the ad, so the low index is a measurement of Meta's own release schedule and of how rarely a long-running ad gets touched. Both rows stay in the CSV, marked as not for headline use. The general form of the trap is worth keeping: when a field's possible values change over time, any comparison between an old cohort and a new one will find a difference, and the difference will belong to the platform rather than to the thing you were studying.

Every figure on this page is a row in one CSV, published under CC BY 4.0, so you are free to use it including commercially with attribution to AdPlay.ai. Each row carries both groups, both denominators, the index, whether the label came from a model or from Meta, and whether we consider the row solid enough to lead with. If you are writing about Malaysian advertising, take a number you can source over a benchmark that circulates between agency blogs with no denominator attached to it.