What Language Malaysian Facebook Ads Use (2026)

45.6% of running Malaysian ads carry an English label, 27.1% Malay and 22.7% Chinese. The first number is lower than most people expect and the third is higher. The practical consequence is that English is a plurality rather than a default. More than half of Malaysian advertising is not in English, and it splits into two audiences that need different creative rather than a translation. An advertiser who runs English only is not covering the market with a lingua franca, they are competing in the most crowded of three lanes. Two smaller numbers are worth sitting with. Tamil accounts for 56 ads out of more than two hundred thousand, which is close to absent for a language spoken by a substantial community. And 1.9% of ads carry a language that is not a market language here at all, mostly Spanish, Turkish, Korean and Thai. Those are almost certainly Malaysian-registered advertisers selling into other countries, which is a real and separate observation rather than a data error, and it is why those labels are mapped deliberately rather than dropped.

The national split hides most of what is useful. Transport ads run 46.7% Malay against 36.4% English. Fitness inverts it at 16.4% Malay and 57.7% English. Chinese moves less at the extremes but sits above a quarter of ads in several categories. Read that as audience rather than as language preference. Categories that sell into a broad national market skew Malay; categories that sell to an urban, higher-income or younger buyer skew English; categories with a strong Chinese-Malaysian customer base carry a Chinese share far above the national average. None of that is surprising to anyone who sells here, and until now none of it had a number attached. What this does not license is a rule like "use Malay for services and English for software". The figures describe what advertisers currently do, and the archive holds no spend or results, so a category running mostly in one language is evidence about convention rather than about what converts.

The archive stores language as free text written by a model, and it holds 374 distinct values for what is really a handful of languages: Chinese arrives as Chinese, as zh, as zh-CN, as Traditional Chinese, as Chinese (Simplified) and as Mandarin. Producing a split at all means merging them, and merging is a judgement. So the judgement is published. The merge map is a file in the repository, every raw label ships in the CSV as its own row with the family it was folded into and the rule that did it, and the builder refuses to run if a label above a size threshold is not in the map deliberately. That last part is not decoration: it fired on the first run and named fourteen languages that were about to fall through silently. One rule worth stating because it is the easy mistake: a label naming two languages, like English/Malay or Mixed (English and Chinese), is counted as Multiple and never split between families. There is no way to know from a single label which language the ad led with, so splitting it would be inventing the answer.

A language study is only as good as its list of languages, and getting that list is harder than it sounds. The search index returns at most 100 distinct values for a field, alphabetically, so one query gives a list that looks complete with an arbitrary tail missing. Partitioning by vertical, then by format, then by year brings 73 partitions back under the cap, and 4 still truncate at the deepest split used. Rather than claim completeness we measured the gap. Counting every enumerated label exactly and subtracting the sum from the ads known to carry any label leaves a residual of 10 ads out of 221,588. The list is complete to within ten ads, and that residual is a published row rather than a footnote. The study also had a target closer to home. Our own skincare hub asserts, with nothing behind it, that in Malaysian skincare "English leads, Chinese is a close second, and Malay carries a meaningful share". Measured across beauty and personal care ads: English 40.1%, Chinese 31.3%, Malay 23.8%. The claim holds, including the ordering, which is a better outcome than we had any right to expect from a sentence written on instinct.

The limit that matters most is one label per ad. Malaysian ad copy code-switches constantly, and a caption that opens in Malay and closes with an English call to action carries a single label. So the Multiple family at 1.7% is a floor on multilingual creative and nothing like a measurement of it, and the single-language shares are correspondingly inflated. Nothing in the data can fix this; only re-reading the copy itself could. The label is also model-written rather than declared by the advertiser, so it is an inference from the creative. Read a gap of twenty points as real and a gap of two as noise. And one thing this page is explicitly not about: it describes ad creative, not websites. It says nothing about which language a Malaysian business should publish its site in, and it is not an argument for or against a Malay-language version of anything.

Every raw label, every family total, the per-vertical cut and the coverage residual are in one CSV under CC BY 4.0, free to reuse including commercially with attribution to AdPlay.ai. Because the merge map ships inside the file, you can rebuild the families a different way from the same rows if you disagree with ours, which is the point of publishing a judgement rather than a conclusion.