The AI Video Campaign That Handed the Render to the Audience

An AI video campaign handed the render to the street

The most effective AI video campaign of 2026 so far did not come out of a studio. Brewlander, a Singapore craft brewery, printed AI text prompts on billboards, bar coasters and social posts and invited drinkers to render their own cinematic beer commercials. The work, AI Irresponsibly by BLKJ Havas, took the first Grand Prix in the programme's 23 editions at the Effie Singapore Awards on 10 September, and the David and Goliath Award for outsized impact on a fraction of the budget alongside it.

That gala handed out 50 trophies across 27 categories, and the top one went to a campaign whose footage was mostly made by strangers. The organisers measured a 32% increase in spontaneous brand awareness and 15% year-on-year sales growth, and their own description of the work is the informative part: it was built to drive awareness and engagement without compromising the brand's focus on product quality. The award did not recognise a film. It recognised a mechanism that produces films in volume and lets the brand keep the product as the one thing it fully controls.

Brewlander's founder John Wei and BLKJ Havas chief client officer Vivien Foo told the Effie Singapore Learning Festival that the result came from giving consumers creative control and treating participation itself as a competitive advantage. The rest of the winners list repeats the logic. Grab's Awwbandon Cart gave abandoned shopping carts emotional personalities and lifted conversion on the first message from 9% to 31.5%. McDonald's GriddleCore A.M. Fest, grounded in ethnographic fieldwork rather than trend data, added 10% to launch impact and 17% to average campaign sales. FairPrice reframed a supply-chain advantage to move purchase intent. Very little of this is a craft story, and all of it is a behaviour story.

The prompt became the media buy

Put a prompt on a coaster and the media buy stops being a slot. It becomes an instruction, and the brand's production line becomes whoever is holding the coaster. The brief, the claim sheet and the brand-safety policy collapse into one short string that has to survive being copied, screenshotted, mis-typed and rendered by someone who has never read a brand book. The closest precedent is {{link}} in retail, where a brand's templates let partners produce their own cuts inside a fixed visual system.

The difference is who sits downstream. A retail partner signs a contract and can lose it; a drinker in a bar does not. So the usual production controls have nowhere to attach. A shot list describes shots that were never planned, a review gate sees one render in thousands, and a signed-off master has no relationship to what actually ships. What replaces them is pre-commitment. The prompt has to be written so that every plausible render is acceptable, because nothing later in the chain will catch the one that is not.

That inverts the direction most video teams write in. Instead of producing assets and then checking them, you specify a permitted range before anything exists, and you accept that the range is the creative. It is a smaller document than a treatment, and it is the only artefact in the campaign that a stranger will follow literally.

The closest precedent is template-led co-creation in retail, where a brand's templates let partners produce their own cuts inside a fixed visual system.

Awards grade behaviour while the tools grade throughput

The juries are not the only people scoring AI work this year, and the two scorecards disagree. IAB Europe's 2026 Impact of AI on Digital Advertising report, released on 24 September at DMEXCO, found that 86% of respondents use AI for marketing, that 72% increased general AI investment in 2026, and that operational efficiency is the most commonly selected measure of AI's effectiveness, cited by 69%. The same respondents rated AI's current performance in advertising operations at 2.76 out of 5.

Hold the two together and the problem is not adoption. The industry has standardised on a metric that reports whether work happened rather than whether it worked, and the score that metric returns is barely a pass. {{link}} closes that gap by putting the constraint in the brief, where it can be enforced, instead of in a review that inspects one cut in fifty.

Effie Worldwide's own framing is the cleaner half of the same argument. It is not marketing if it is not effective, the organisation says, and marketing is the business of changing minds, behaviours and outcomes. It has been running that position for more than 50 years, which is long enough for it to have seen several generations of tooling arrive and not change the question. Efficiency is what a tool reports about itself. Uplift is what a jury asks you to prove.

constraint-led production closes that gap by putting the constraint in the brief, where it can be enforced, instead of in a review that inspects one cut in fifty.

What a brand still controls when a stranger renders the cut

When the audience renders, the brand's controllable surface is small, and it is worth naming precisely. The first is the prompt itself, which now carries the claims, the product truth and the exclusions at once. The second is the product: the label, the pour, the pack, the silhouette, the colour. Those are the only frames a brand can guarantee appear in someone else's render, which is why the line about product quality in the award write-up is load-bearing rather than a pleasantry.

The third is the mechanic, and it is the one teams underrate. Where the prompt appears, how a render is submitted, what the brand does with the entries and whether they are published are media and product decisions, all of them testable before a single frame exists. The same logic showed up in {{link}}, where the standard moved from perfect to fast enough to be in the conversation, and the winner was the team that shipped inside the window rather than the team that shipped the best frame of the month.

Then watch where the Effie numbers actually point. A 32% rise in spontaneous brand awareness is a property of a mechanic rather than of an asset, and so is a first-message conversion lift from 9% to 31.5%. When you cannot enumerate the cuts, asset-level reporting has nothing to enumerate. The measurement has to be rebuilt around participation instead: how many people entered, what share of entries cleared the claim rules, and what changed afterwards.

The same logic showed up in AI parody ads, where the standard moved from perfect to fast enough to be in the conversation, and the winner was the team that shipped inside the window rather than the team that shipped the best frame of the month.

Disclosure does not transfer with the prompt

One duty stays where it was. A render produced by a member of the public is still an advertisement for the brand that supplied the prompt, and the disclosure and substantiation obligations sit with the advertiser, not with the person who typed it. IAB Europe's report captures how unsettled the rulebook still is underneath that: 78% of organisations have named a formal owner for AI governance, but only 48% have AI guidelines that are specific to advertising and marketing, and 38% are still assessing the impact of the EU AI Act.

{{link}} is the work of deciding what a generated frame has to prove before it can ship, and that work does not get easier when the person shipping it does not work for you. The same report names the reason the gap persists: a lack of expertise or training is the most commonly cited barrier to adoption, ahead of tooling, cost and integration. Nobody is short of a generator. Teams are short of a written rule that says what the generator may produce.

The practical form of that rule is pre-clearance rather than review. Define the range of depictions the prompt may produce and the claims it may imply, then treat anything outside the range as off-brand by default. It is shorter and duller than a brand book, and far more likely to be used, because it is short enough to travel attached to a prompt printed on a coaster.

the trust and QC gate is the work of deciding what a generated frame has to prove before it can ship, and that work does not get easier when the person shipping it does not work for you.

Where the mechanic breaks

Participation has a predictable failure mode, and the APAC Effie Awards, held on 25 September, supply the counterweight. Of 121 finalists, 67 took metal, and the Grand Effie went to Special Sydney and Uber Eats for Get Almost Almost Anything, which the judges described as a witty, highly adaptable and effective brand platform that repositioned the client, and which carried a second execution to Gold. The awards chairperson, Danielle Jin, set the bar plainly: in a world of increasing complexity and distraction, effective marketing still begins with understanding people.

An adaptable platform and an open prompt are not the same thing. A prompt any stranger can render is open by construction, which means distinctiveness has to be carried by the handful of things the brand can force into every render: a colour, a shape, a pack, a phrase, and a product truth that survives being interpreted by someone with no incentive to protect it. The Grand Prix campaign did not have to invent that constraint, because a brewery has one and a coaster has room for one.

So write it down before you print the prompt. One sentence naming what every render must show, one naming what no render may claim, and one naming what the brand will do with the entries. That is the whole control surface in 2026, and the campaigns that won this autumn are the ones that treated it as a deliverable rather than a disclaimer.