Marketers have sorted their channels into two piles, and the rule that sorts them is countability. A Q1 2026 Digiday+ Research survey of more than 100 marketing professionals found AI now running social media for 49% of them, while connected TV drew just 18%. That gap has little to do with technical difficulty; it tracks whether the outcome lands in a system the marketer can read. Where the result closes into a number, the machine gets the wheel. Where credit stays contested, a person keeps it.

The two piles line up almost exactly with how each channel handles attribution. Retail media and social lead because each closes its loop inside a system the platform controls: retail media books the sale beside the impression, social scores the conversion against its own pixel. That tight coupling is why 42% of marketers now let AI into retail media. Influencer marketing, at 25% adoption, can only credit a sale to a creator by inference, through promo codes and last-click guesswork. Connected TV is harder still, its impressions sealed inside streaming walled gardens that Modern Retail’s Q1 measurement research calls encrypted and reluctant to share audience data. The channels marketers trust to AI are the ones that hand back a clean receipt.

Even inside the channels it trusts, marketers keep AI on a short leash. Most point it at reading results rather than spending money: 73% of retail-media AI users put it on campaign analysis. Far fewer hand it the checkbook. Only 32% of AI users let the software buy social or retail inventory, the one act that commits real budget. “I want a person overseeing the bot,” Bayer’s Glenniss Richards said at Digiday’s Programmatic Marketing Summit. The wheel is surrendered by degrees, and the last degree, the spend, stays in human hands.


The clearest proof that counting is the constraint sits inside the laggard channels themselves. Among the minority who do use AI for influencer work, the top use is data analysis, named by 75%, ahead of content creation (63%) and outreach (56%). Connected TV repeats the pattern, with 69% of its AI users aiming the tool at measurement before anything else. Marketers are not reaching for AI to run these channels. They are reaching for it to count them, hoping the software can build the attribution the channel never supplied. The order is the whole argument: measure first, automate later, and never the reverse.

An optimizer is only as good as the metric it chases, and in influencer and streaming that metric is a proxy no one has learned to trust.

Marketers in streaming count what they can reach, which is attention rather than sales. Their most common measures of success are watch time and impressions, with 46% naming watch time as their main metric on YouTube. Those figures record whether an ad played, not whether it worked. An AI told to maximize watch time will maximize watch time, faithfully, whether or not one product moves. Handing it that objective does not close the attribution gap; it automates the optimizing of the wrong thing. Until the metric maps to a sale, the rational move is to keep a doubting human between the model and the money.

The strongest objection here has nothing to do with attribution; it is about authenticity. Influencer marketing, the argument runs, resists AI because its value is a human voice audiences believe, and synthetic creators break the spell; a World Federation of Advertisers study found 60% of brands have no plans to use virtual influencers at all, and 96% cite consumer trust as a concern. The mechanism is real, and for influencer marketing it is probably part of the story. It cannot explain connected TV, where no one asks that a pre-roll be authentically human. What links influencer and streaming is the absence of a clean count, which no appeal to authenticity can explain away. Authenticity accounts for one reluctant channel; attribution accounts for both.

This is not a story about channels marketers are abandoning. A separate Mediaocean H1 2026 outlook, as reported by PPC Land, found 63% of marketers planning to raise connected TV budgets, as high as any channel in the survey. Money is pouring into the place AI is least welcome, so the resistance is about control, not conviction. Marketers broadly cite the difficulty of connecting insights across siloed systems, at 41%, as a top barrier to AI adoption; cost of media and lack of budget rank alongside it in Modern Retail’s separate accounting of CTV-specific obstacles. Solve the plumbing that makes an outcome countable, and the autonomy will follow the budget that is already there.

If attribution is the ceiling, the race in AI marketing will be won by whoever closes the loop, not whoever trains the largest model. Retail media leads adoption for a structural reason rather than a clever one: Amazon and the big retailers own the shelf and the checkout, so the outcome resolves inside a single system. The remaining channels will be handed to autonomous AI in the order their measurement is solved, as clean rooms and unified data stacks turn proxies into sales. Vendors selling “AI for CTV” today are pitching capability into a market whose real shortage is a number the buyer can trust. Marketers are not asking whether AI can run an influencer campaign or a streaming buy; the software plainly can. They are asking whether they can prove what it did, and until they can, the human keeps one hand on the wheel.