Our position on synthetic creative runs on evidence, not conviction. The four conditions we track, and what changes when they move.
The question every brand will face this year, if it has not already, is whether the people in its advertising need to be real. Synthetic creative, generated faces, generated settings, whole performances assembled by model, has moved from demo to production capability, the cost argument writes itself, and positions are hardening on both sides with a confidence the evidence does not yet support. One camp holds that synthetic is simply the future and resistance is sentimentality. The other holds that it is categorically off-limits, a brand-safety cliff. Both positions share the same weakness, they are convictions, and this is a question that should be run on evidence, reviewed as the evidence moves.
Stated as evidence rather than conviction, the current position is this: real people in real environments are a performance moat, and the moat is conditional, defined by four observable conditions, each of which is moving. The honest phrase is for now, and taking both halves of that phrase seriously is the entire discipline.
The mechanism is fluency. Feed audiences are the most creative-literate population in history, billions of hours of made-for-feed content have trained them to classify what they see pre-consciously, by style, in under two seconds, and the classification governs the swipe. Within that grammar, real people in real environments are the native dialect, the texture the feed's resident content is made of, and synthetic, today, still pattern-matches as off. Not always consciously, the viewer rarely thinks generated, they feel something adjacent to it, the uncanny residue, the stock-photo smell, and the thumb resolves the feeling before any message lands.
On a medium where attention is the gating resource and the auction prices it, that classification tax is expensive. Work that reads as not-quite-real pays it at second one, before idea, offer or brand get their hearing. And for categories built on demonstrable truth, skin, food, real outcomes on real bodies, the tax compounds into a credibility problem, a claim demonstrated on a face that does not exist is a claim that has quietly cancelled itself. Authenticity is not an aesthetic preference here, it is the proof mechanism, which is why the performance gap persists even as generation quality climbs.
Conditional positions are only as good as the conditions are explicit. These are the four that define this one, and none of them is static.
Disclosure rules. Platforms and regulators are converging on labelling for synthetic and AI-generated content, and the rules are tightening, not loosening. What a mandatory label does to feed behaviour, whether it functions as a neutral notice or an avoidance trigger, is being established in the data now, and it will not be the same in every market or format.
Algorithmic treatment. The platforms decide what their delivery systems do with labelled synthetic content, rank it neutrally, deprioritise it, segregate it. Their incentives cut both ways, they are invested in generation tools and in feed quality simultaneously, and their policies will move with user response. Delivery treatment is a larger performance variable than generation quality itself.
Brand legal policy. Likeness rights, talent union positions, IP exposure from training data, the question of who is accountable when a generated face resembles someone real. Large-advertiser legal teams currently range from cautious to closed, and their movement will set the practical floor for what major brands can deploy regardless of what performs.
And the performance gap itself. The only condition that ultimately matters, measured, not asserted. Attention, completion, brand effects, synthetic against real, like for like, format by format, tracked continuously, because generation quality improves monthly and audience fluency evolves with it, in whichever direction the exposure arms race actually runs.
Where the evidence already permits synthetic, use follows, backgrounds and environments, product visualisation, variation and localisation of real footage, the augmentation layer where the human in the frame remains real and the machine multiplies everything around them, which is where most of the production economics were hiding anyway. That is not a compromise position, it is what the current data supports, AI scaling production while the element audiences are fluent in stays authentic.
The strategic discipline this asks of a brand is unusual only in marketing, everywhere else it is just empiricism. Hold the position the evidence supports, instrument the conditions, and pre-commit to moving when they move. The weekly creative loop makes this nearly free, synthetic variants can be tested in bounded cells alongside real ones, the deltas observed in real delivery rather than debated in meetings, the policy updated by data instead of by whoever argued best. Media creates the opportunity to see your ads. Creative determines whether they work. Whether synthetic creative works is a question the feed will answer empirically for any brand willing to ask it properly.
What the discipline rules out are the two convictions. All-in adoption is currently a bet against the audience's measured response, paid for in attention tax across the whole portfolio. Categorical refusal is a bet that four moving conditions will never move, taken without instruments, and its cost arrives later, as unpreparedness, the day the conditions flip and the playbook does not exist. Both are comfortable. Neither is defensible to a board asking the simple question, what does the evidence say this quarter?
For now, the evidence says real people are the moat, the native dialect of the medium, the proof mechanism of the claim, the texture the feed rewards. For now is not a hedge, it is the position, held with instruments running and the exit mapped. Brands that operate this way will be the last to pay the synthetic tax while it exists and the first to stop paying the authenticity premium if it ever inverts, which is all a moat was ever for, buying advantage while the conditions hold, and knowing, before anyone else, when they stop holding.