Running advertising in more than one language sounds like a translation task and is not. Three separate things go wrong, in three separate places — the image, the copy, and the campaign structure — and only one of them is visible in a report. Here is each, and how to structure a multilingual campaign that actually works.
Problem one: the script breaks on the image
Arabic joins and runs right to left. Devanagari, Tamil and Bengali have conjunct characters and vowel marks that attach above and below the line. Latin text does none of this.
When an AI image tool draws text onto a generated scene, the rendering pipeline was almost certainly built and tested on Latin script. Three failures follow, in order of frequency:
Characters unjoined
Letters that should combine are drawn separately. Technically the right characters, completely unreadable — the equivalent of English printed one letter per box.
Direction reversed
Right-to-left text laid out left to right. The sentence comes out backwards. Native readers see it instantly; nobody else does, which is exactly why it ships.
Font substitution
The display font has no glyphs for the script, so the system falls back silently to a default that looks nothing like the brand and often breaks the layout.
💡 Test before trusting any tool with this. Generate one image with a headline in the target script and have a native reader look at it — not a translator checking the words, a reader checking whether it looks like something a real company published. Those are different tests and only the second catches rendering faults.
The structural question to ask a vendor: how does text get onto the image? If the image model generates it as part of the picture, it will be wrong in any script — image models cannot spell in English either. If the tool generates the scene and then draws the headline as real text through a layout engine in a real font, script support becomes a font question rather than an impossible one.
Problem two: translated copy underperforms written copy
Even with perfect rendering. English ad writing leans on compression and wordplay that does not survive translation — "Book now, pay later" travels fine; "Stop guessing. Start selling." becomes two flat imperatives with none of the rhythm.
There is also a register problem that catches people out. In many languages the formal register reads as government or corporate communication rather than a business talking to a customer, and a literal translator will reliably choose it. The result is technically correct copy that feels distant.
The fix is to brief the model with the offer and the audience rather than the English sentence, and have it write in the target language from scratch. A tool offering only a "translate" button is doing the weaker of the two jobs.
Problem three: one language starves the other
The invisible one, and the reason people conclude a second language "does not work" when they never actually tested it.
Put two languages in one ad set and the platform optimises toward whichever creative gets early traction. That is frequently the majority language, simply because it reaches a broader initial slice. The second creative then starves before it has been fairly tested, and its numbers look terrible — because it barely ran.
One ad set per language
Each with its own budget and its own learning. You find out what the second language actually costs per result rather than assuming.
The cost of splitting
Platform minimums apply per ad set, and a conversion objective sits at roughly three times the impression floor. Two languages means two funded ad sets.
If you cannot afford both
Run one properly. Two starved ad sets teach you nothing and deliver worse than one funded one.
How platform language targeting actually behaves
Misunderstood often enough to be worth stating plainly. Meta's language targeting filters on the language a person has set on Facebook or Instagram — not the languages they speak.
💡 An enormous number of Arabic, Hindi, Tamil and Spanish speakers use their phone interface in English. Target their language and you exclude most of them. For most campaigns the better structure is to leave language targeting off and let the creative filter — an ad in a language is self-selecting, and you reach everyone while paying only for the ones it speaks to.
Match the destination to the language
A common and expensive mismatch: ad in one language, landing page in another. You bought the click and discarded it at the door.
If you cannot produce a translated page, that is a genuine argument for running those ads into a chat destination rather than a website — the conversation continues in the language the ad started in, with a person on your side of it. This is a large part of why click-to-chat advertising performs so much better in multilingual markets than in monolingual ones.
A structure that works
- 1One ad set per language, each funded above the platform minimum. Same objective, same geography.
- 2No language targeting. Let the creative do the filtering.
- 3Copy written in the target language from the offer, not translated from English.
- 4Creative tested by a native reader before launch, for rendering as much as wording.
- 5Destination matched — translated page, or chat.
- 6Cost per result compared per language. They will differ, and the difference tells you where the next unit of budget goes.
Written for the market, and rendered correctly
Adyft writes copy in the language of the market from your offer, and draws headlines as real text over the image so the script renders properly. 14-day free trial.
Start free — no card needed