By Professor Stoyan Stoyanov
Marketing teams across the Gulf have spent two years solving a real problem: more content, in more languages, across more channels, at lower cost. Generative AI solved it. It also solved the same problem for every competitor at the same moment, through the same small set of models trained on heavily overlapping data. The result is not only abundance. It is convergence.
My research on what I term the liability of algorithmic standardisation examines this in firms competing across borders, but the logic transfers to brand communication. When thousands of organisations query the same models to sound credible to the same audiences, those models pull every output toward the statistical centre of what has already been written. Content becomes cheap. Difference becomes scarce. Most marketing organisations are still measuring the first while spending down the second.
The convergence you cannot see
This is hard to govern because it is invisible from the inside. A language model produces a new sequence of words for every prompt, so no two outputs are identical and every marketer experiences the result as bespoke. The convergence is distributional rather than literal. Two AI-drafted campaign briefs may share almost no exact phrasing while sitting on top of each other in meaning, cadence and structure. You cannot detect that by rereading your own work, only by reading it beside eleven competitors.
Your audience occupies exactly that position. A procurement lead reads thirty proposals in one sitting; a consumer scrolls forty product pages. Buyers hold the only comparative view in the system, and they are developing a reliable ear for the register.
That has a consequence marketers tend to misread. Polish used to carry information: a well-structured, idiomatically confident piece of communication was expensive to produce, which is why it worked as evidence of competence and investment. Generative AI has made that surface almost free at every level of underlying quality. The rational audience response is not hostility toward polished work but indifference to it. The cue stops carrying information, and trust migrates to whatever stays costly: named results, verifiable provenance, an argument someone could be wrong about, a reputation visibly on the line.
Coca-Cola and the price of the average
Coca-Cola’s holiday advertising is the public test case. In 2024 the company rebuilt “Holidays Are Coming” using generative AI and drew sustained criticism for uncanny figures and an oddly lifeless feel. In 2025 it went again with animals rather than people, and the execution was visibly better. The criticism did not subside, because the objection was never technical. Coca-Cola’s holiday asset is a shared memory, and reconstructing a memory with a system trained on everything resembling it yields something adjacent, competent and unmistakably second-hand.
Company executives have said the work tested strongly and have continued with it, and both positions can hold at once. Efficiency and recall are measured inside the campaign; the cost is paid at the level of brand distinctiveness, over years, where no campaign metric is designed to look. That asymmetry is the trap. The gain is legible on this quarter’s dashboard, the loss is legible nowhere.
The same pattern runs less visibly through performance marketing, where it is closer to complete. Across whole e-commerce categories, product descriptions, ad headlines and subject lines have settled into a common rhythm: the same three-benefit structure, the same hedged claims, the same opening verb. No one authored that convergence and no individual asset looks wrong. The category simply stops sounding like anyone in particular.
Why Gulf brands are more exposed, not less
The regional context sharpens the risk. MENA AI marketing spend is growing at more than thirty per cent annually, Arabic content generation is among the fastest-growing use cases, and Gulf sovereign funds committed roughly sixty-six billion dollars to AI-related sectors as 2026 opened. Adoption is not the problem. Undifferentiated adoption is.
Two features make Gulf brands more exposed rather than less. First, the dominant models are anchored in Anglophone corporate register, so Arabic output frequently carries English structural DNA: translated fluency rather than written fluency. The further a market’s idiom sits from the model’s statistical centre, the more detectable the seam becomes locally, and the faster discounting sets in. Second, timing. Much of the region’s brand building is happening now, across giga-projects, new tourism properties and newly corporatised national entities. Brands built during a convergence period inherit its voice, and voice is the hardest asset to retrofit.
Governing for difference
The answer is not to restrict AI use but to govern it with distinctiveness as an explicit constraint. Four moves are available to any CMO this quarter.
Measure sameness. Run a distinctiveness audit: take your last fifty outbound assets and a matched set from your three closest competitors, and compute semantic similarity across them. If your own team cannot attribute them correctly with the logos removed, neither can your buyer. Report that score alongside reach and cost per acquisition.
Separate prompt-layer from model-layer work. Tone-of-voice instructions and brand guideline injections relocate your output within the shared distribution; they do not exit it, and any prompt technique that works is imitated within a quarter. Exiting requires assets competitors do not hold: fine-tuning or retrieval built on your own archive of what has actually converted, your own customer language, your own data. That is a differentiation investment, not an efficiency one.
Spend the saving on provenance. The efficiency gain is real and should be redirected rather than banked, into the things that stay expensive: named client outcomes, original regional research, an executive prepared to say something contestable, footage that is obviously unedited, local idiom no model produces unprompted.
Route by discretion. Where a decision turns on price, availability or rating, default AI output is efficient and sameness costs little. Where someone is deciding whether to trust you, distinctiveness is the whole proposition. Most organisations run one AI policy across both, which optimises the second channel for the economics of the first.
The scarce asset
The efficiency argument for generative AI in marketing is settled. The distinctiveness argument has barely begun. The scarce resource is no longer content, and arguably no longer reach or attention either. It is difference, and it is depreciating faster than most brand teams are replacing it. No brand was ever chosen because it sounded like all the others.
(Stoyan Stoyanov is a Professor of Management and Director of Research, School of Social Sciences, Heriot-Watt University Dubai)
