The AI-Assisted Ad That Went Viral — What the Data Says About Audience Reception

Every few months, an AI-assisted ad escapes the industry bubble and lands on the phones of people who have never heard of a diffusion model. It is shared in family WhatsApp groups, dissected on LinkedIn and turned into a meme by lunchtime. Agencies celebrate the reach, brand teams quietly check the view counts, and a familiar conclusion takes hold: audiences are ready for machine-made creative.

The data tells a more complicated story. Virality is a measure of attention, not of affection, and attention can be earned by delight, curiosity or outright mockery. The more useful question for a marketer is what happens after the share button is pressed. Did viewers remember the brand? Did they feel what the creative team intended? And did their opinion change the moment they learned how the ad was made?

Two kinds of viral

It helps to separate two phenomena that tend to be bundled together. The first is virality of the work itself, where people respond to the story, the humour or the craft. The second is virality of the controversy, where the conversation is about the technology and whether a brand should be using it at all. Both generate impressions. Only one tends to generate preference.

Recent research suggests that audiences are quite capable of holding both reactions at once, and that the split often depends less on the ad than on what viewers are told about it. That is where the evidence gets interesting, and where several independent studies begin to point in the same direction.

When the audience did not know

Coca-Cola’s AI-assisted holiday film, which reimagined its much-loved 1995 spot, remains the most instructive recent case. Parts of the creative and design community criticised it loudly on LinkedIn and X, arguing that a brand with Coca-Cola’s resources should have commissioned human artists. Yet when System1 tested the full-length film with ordinary viewers who were not told how it was made, it scored 5.9 out of a possible 6 on its star rating, a result the research firm classed as exceptional.

Negative emotion was almost absent. Only about 3 per cent of respondents registered contempt, the sole negative response recorded, and a few of those comments referred to the ad looking machine-generated. System1’s reading was that viewers were absorbed by the festive feeling rather than the production method. It also pointed to a cautionary counterexample: Google’s Olympics spot, which led with AI as the headline feature, drew heavy criticism and was eventually withdrawn.

The lesson is not that AI should be hidden. It is that audiences respond to what an ad makes them feel, and the technology becomes the story only when the brand or the commentariat makes it the story.

The disclosure penalty

A study by researchers Anindya Ghose, Hyesoo Lee, Vilma Todri and Panagiotis Adamopoulos, published on SSRN in November 2025, offers some of the cleanest evidence yet on performance. Comparing human expert ads, human work modified by generative AI and fully machine-generated visuals, the team found that the fully generated ads outperformed the human-made ones, lifting click-through by as much as 19 per cent. The hybrid category, where AI touched up human work, produced no significant improvement, a finding that should make any agency pause before assuming that a little AI is a safe middle path.

Then came the twist. When the same ads were labelled as AI-made, performance fell. Trade coverage of an IAB framework that cites the NYU work reported a 31.5 per cent drop in click-through once labels were applied. In other words, the creative did not change, but the audience’s willingness to act on it did.

For marketers, this sets up an uncomfortable tension. Transparency is rapidly becoming an expectation, and in some markets a regulatory direction of travel. Yet disclosure carries a measurable cost, at least for now. Brands that treat labelling as an afterthought may find that the penalty arrives precisely when they can least afford it.

Buyers are guessing wrong about viewers

Perhaps the most striking data point concerns the industry’s own assumptions. FreeWheel, Comcast’s ad tech arm, found that 48 per cent of media buyers expected viewers to find AI-generated creative off-putting. Among viewers who had actually seen such an ad, only about 10 per cent said they did.

It would be easy to read that as proof that buyers are out of touch, but the studies together suggest something subtler. Viewers pass a verdict on the ad in front of them, and an ad that entertains or moves them tends to be judged on those terms. Buyers, by contrast, are reacting to the abstract idea of AI in advertising, which carries all the baggage of the wider debate about jobs, authenticity and trust. The two are not measuring the same thing, and treating them as if they were is how brands end up either over-cautious or over-confident.

The generational fault line

Tolerance is far from uniform. January 2026 research from the IAB and Sonata Insights, titled The AI Ad Gap Widens, found that Gen Z respondents were nearly twice as likely as millennials to view AI-generated ads negatively, at 39 per cent against 20 per cent. The survey covered 505 US consumers aged 16 to 43, with 150 each in the Gen Z and millennial groups, alongside 104 advertising executives, so it is a useful directional signal rather than a definitive measure.

That matters because younger audiences are exactly the ones brands most want to reach through social and short-form video. They are also the cohort most fluent in AI tools, which may make them quicker to spot artefacts and more sceptical of polish that feels synthetic. Familiarity, it seems, does not automatically breed acceptance.

Reaction is not recall

Kantar’s work with Affectiva adds another layer. Using facial coding alongside survey data, the research found that AI-generated ads tend to provoke stronger reactions, but with lower net positivity. Synthetic visuals, particularly of people, can be jarring or distracting, and a smile on a viewer’s face may reflect amusement at an odd artefact rather than enjoyment of the story.

Kantar illustrated the risk with an unofficial Liquid Death ad. It generated a strong emotional response, yet its brand memorability landed in the bottom quarter of all ads the firm had tested, partly because the brand was introduced early and then barely integrated into the narrative. The viewers were engaged, but not necessarily with the brand.

This is the core warning in the data. An AI-assisted ad can go viral for reasons that have nothing to do with the product, and a spike in shares can disguise a weak return on brand building.

What India’s best-known example teaches

India has its own landmark case, and it predates the current wave of generative tools. For Diwali 2021, Cadbury Celebrations, with Ogilvy, Wavemaker and the technology firm Rephrase.ai, created the Shah Rukh Khan My Ad campaign. Machine learning recreated the actor’s face and voice so that he could promote individual neighbourhood stores by name. According to the agency case study, roughly 130,000 ads were produced for more than 2,000 businesses, drawing around 94 million views across YouTube and Facebook. The work later won India’s first Cannes Titanium Lion, and the agency side has credited the campaign with a 35 per cent lift in business, a figure that is worth treating as the creators’ own claim.

What made it land is as instructive as what it achieved. The technology served a clear human insight: small retailers were struggling for attention during a festival that matters enormously to the brand, and Cadbury gave them access to a celebrity they could never have afforded. Industry voices reflecting on the campaign years later have stressed that it was the combination of idea, star and technology that worked, not the AI on its own, and that few festive campaigns since have matched its cultural impact.

The campaign also benefited from novelty. It arrived when the idea of a digitally replicated celebrity was genuinely new, and audiences had not yet formed settled attitudes about synthetic media. That window has narrowed considerably.

What brands and agencies should take from the data

The first lesson is to measure the right outcome. Reach, shares and views are the easiest numbers to celebrate, but brand recall, emotional response and conversion tell a more reliable story. Testing with audiences who have not been told how the ad was made, and then with those who have, gives a far clearer picture of the disclosure risk than any dashboard of impressions.

The second is to lead with the idea. Across the cases above, the ads that held up were those with a clear emotional or cultural anchor, where the technology supported the story rather than becoming the story. Leading with the tool as a feature, as the Google example suggests, invites judgement on the tool.

The third is to plan for transparency rather than hope to avoid it. If labelling will reduce response, as the evidence suggests, the answer is stronger creative and clearer value for the viewer, not quieter disclosure. Brands that build trust before they need it are likely to absorb the penalty better than those scrambling to explain themselves after a backlash.

The fourth is to segment carefully. A campaign aimed at older or more mainstream audiences may meet little resistance, while one targeting Gen Z may need more visible human craft and a more considered tone. It is also worth remembering that much of this research draws on Western samples. Indian audiences, with their diversity of language, region and media habits, are likely to respond differently, and the local evidence base is still thin. Brands would do well to run their own tests rather than import conclusions wholesale.

Verdict: the audience is not the problem

If there is a single takeaway, it is that audiences are neither the hostile critics some buyers fear nor the uncritical fans some vendors imagine. They respond to ads as they always have, with their emotions first and their reasoning second, and they adjust their judgement when new information arrives. An AI-assisted ad that tells a good story can earn genuine goodwill. The same ad, once framed as a technological stunt or a cost-cutting exercise, can lose it.

The viral moment, in the end, is only the opening scene. What the data suggests is that the brands who benefit most from AI-assisted creative will be those that remember the oldest rule in advertising: people do not remember how an ad was made, they remember how it made them feel, and whether they knew whose ad it was.