UGC prompts in Arabic: why a Gulf dialect hook sells differently from the same line in Fusha

Nine words of English, stop scrolling if your skin dries out from the air conditioning, become three different sentences in a working set of Arabic UGC prompts. A UGC prompt is the brief a video model or a creator turns into a person talking to a phone camera, and in Arabic it has to name the dialect and spell out how each line sounds, or the actor ends up reading it like the news.

Picture the woman saying it, phone at arm's length in a kitchen. In Arabic her line has to be written three times before it is finished, because the woman in the Riyadh version, the Cairo version and the pan-Arab television version do not say the same sentence. They do not even use the same word for the air conditioner.

UGC stands for user-generated content, and a UGC ad is one that borrows the look of it: a person talking to a phone camera as if nobody paid them. The whole format rests on sounding like a neighbour. Fusha, the standard Arabic of news bulletins, school textbooks and subtitles, is a register nobody uses to talk to a friend. So the moment a UGC actor opens in Fusha, the viewer knows what is coming.

A UGC ad in Fusha announces itself as an advert in the first second, which is the one thing the format exists to avoid.

That is why, when I built the Arabic UGC Prompt Library, dialect became a layer of the prompt rather than a translation step at the end.

What actually changes when the same hook is written in Gulf, Egyptian and Fusha?

Here is the kitchen hook three ways. The transliteration is the line the prompt carries alongside the Arabic script, for reasons I come to below.

Three things stand out. First, the two dialect versions are not the Fusha sentence with the endings filed off. They are restructured. Spoken Arabic hooks someone with a short question, not with an imperative and an "if" clause, so the sentence changes shape before any word changes. Second, the everyday nouns move. An air conditioner is the most ordinary object in a Gulf or Egyptian household, and the two regions call it different things. A Riyadh viewer hearing "takyīf" places the speaker in Cairo within half a second. Third, the sounds move even where the letters do not. The same written qāf is a g in the Gulf line and a glottal stop in most Egyptian speech, which is why the script on its own is not enough.

Gulf itself is a family, not a single voice. An Emirati, a Saudi from Riyadh and a Kuwaiti share the register and differ on words. The library treats Gulf as one layer and leaves the last local touches to the creator or the client, because a prompt that tried to be Emirati for one order and Qassimi for the next would be wrong for both most of the time.

Why does the prompt carry a transliteration line as well as the Arabic?

Because the video model reads with an accent. Veo 3.1 now generates dialogue natively, so the actor's voice comes from the same prompt as the picture. In my tests with Veo, Kling and the avatar tools, Arabic script handed over on its own tends to be voiced the way a text-to-speech engine reads a news script: Fusha vowels, a q for every qāf, every ending pronounced. Feed it the Gulf line above in script only and the actor says "waqqif" like a newsreader, and the ad is dead on arrival for the audience it was written for. The transliteration pins the sound: "waggif", "shway", "il-mkayyif". It also lets a non-Arabic-speaking editor or client read along with the cut and see where each phrase lands.

The seven layers of a prompt in the library are frame, subject, timed action, dialect dialogue with transliteration, audio, camera and negatives. The dialect layer is the fourth. When a client wants the Cairo version of an ad that was written for Jeddah, the other six layers stay exactly as they were. The kitchen, the woman, the second she picks up the product, the ambient sound, the lens, the list of things the render must not do. Only the fourth layer is swapped. That is the whole dialect switch, and it is why the same prompt of 2,200-plus characters can serve two markets without being rewritten.

The negatives layer does quiet work for dialect too. Anything that makes the clip look produced belongs there, and a broadcast reading voice is one of those things.

What does a brand actually ask a UGC creator for?

The brief that lands in a creator's inbox is shorter than people expect. The market the product is sold in. The product and the one claim allowed about it. Who is speaking: a mother, a man leaving the gym, a student. The length and the platform. And then the sentence that decides everything: make her sound Saudi, or make it Khaleeji, or, more rarely, make it something everyone can understand.

Those three requests are three different jobs. A brand selling in one country wants the actor to sound like the buyer's cousin, and any slip toward Fusha reads as an import. A brand with pan-Arab reach and a television budget usually wants Fusha, or an Egyptian that the whole region has been trained to follow through decades of film and comedy. Dialect is a distribution decision made by the brand, and the creator's job is to deliver it without a trace of the other register showing through.

Which is also the quiet commercial point. A creator who can hand back the same brief in Gulf, Egyptian and Levantine is not selling one ad. They are selling three, to three markets, from one shoot plan. The library's 178 prompts across five registers exist because that is the shape of the work, not because five is a rounder number than one.

The dialect is not a coat of polish applied at the end. It is the first claim the ad makes, before the product is even named: this is who is speaking, and she is one of you. The viewer checks that claim against every vowel in the first sentence, and a hook that fails the check never gets to the second one.

Sources

• Gulf Arabic, phonology (/q/ became g) · Wikipedia
• Egyptian Arabic phonology (Classical /q/ became a glottal stop in Cairo) · Wikipedia
• Egyptian Arabic (bi- present prefix; why it is understood region-wide) · Wikipedia
• Veo 3.1 · Google DeepMind