Coca-Cola’s AI Ads: What Went Wrong and What It Proves

Coca-Cola’s AI-generated advertising has become one of the most discussed experiments in modern marketing, and not entirely for the right reasons. The brand used generative AI to recreate its iconic 1995 “Holidays Are Coming” Christmas ad, and the reaction was swift, polarised, and genuinely instructive for anyone thinking seriously about where this technology fits in a commercial context.

What the Coca-Cola experiment actually proves is not that AI advertising is broken, but that the bar for emotionally resonant creative is higher than most AI advocates are willing to admit, and that brand equity built over decades is not something you can replicate with a prompt.

What Coca-Cola Actually Did With AI

In late 2024, Coca-Cola released an AI-generated version of its classic “Holidays Are Coming” television ad, the one featuring the illuminated trucks rolling through snowy towns that has been a fixture of British and European Christmas advertising since the mid-nineties. The remake was produced using a combination of generative AI tools and was positioned as a demonstration of what the technology could do for one of the world’s most recognisable brands.

The response was not what the brand’s marketing team would have hoped for. Critics pointed to uncanny valley effects in the imagery, subtle distortions in the human figures, and an overall quality that felt slightly off in ways that were hard to articulate but immediately noticeable. The trucks looked like the trucks. The snow looked like snow. But something in the composition felt assembled rather than crafted, and audiences noticed.

What followed was a fairly predictable cycle of media coverage, with some outlets declaring AI advertising dead on arrival and others defending the experiment as a necessary step in understanding the technology’s limits. Both positions missed the more interesting question, which is what this tells us about how brands should be thinking about AI in their creative processes.

If you are thinking seriously about how generative AI fits into your broader marketing operation, the AI Marketing hub at The Marketing Juice covers the full picture, from workflow integration to the honest limitations most vendors will not put in their sales decks.

Why the Backlash Was Predictable

I have been in rooms where creative decisions get made under significant commercial pressure, and the pattern I have seen repeatedly is that the brief gets compressed when time or budget is tight. What gets cut is usually the thing that makes the work feel human: the extended craft phase, the iterative feedback loops, the willingness to throw out a direction that is technically fine but emotionally wrong.

Generative AI, in its current form, is very good at producing work that is technically fine. It can hit the visual references, approximate the tone, and deliver an output that passes a surface-level check. What it cannot do is feel the discomfort of getting something wrong and push through it to something better. That discomfort is where a lot of great creative work comes from.

The original 1995 Coca-Cola Christmas ad worked because it was genuinely cinematic for its time, because the production values were exceptional, and because it arrived at a cultural moment when that kind of warmth felt earned rather than manufactured. Recreating it with AI does not recreate those conditions. It recreates the visual surface while stripping out the context that made it meaningful.

This is not a criticism of AI as a technology. It is a criticism of how the technology was applied. Using a generative AI tool to remake a beloved piece of brand heritage is one of the highest-risk applications imaginable, and it is worth asking why that decision was made rather than whether the tool itself was up to the task.

The Real Problem Is Where AI Was Applied in the Process

When I was building out the performance marketing operation at iProspect, growing the team from around 20 people to over 100, one of the consistent challenges was getting clients to understand that the tools themselves were not the strategy. A well-configured paid search campaign is not a substitute for knowing what you are trying to achieve and who you are trying to reach. The technology executes. Humans decide what to execute.

The same principle applies to AI in advertising. The Coca-Cola situation is a case study in applying a powerful tool at the wrong point in the process. Generative AI was used to produce the final output, the thing that millions of people would see and judge, rather than to accelerate the thinking, research, and iteration that happens before a single frame is shot.

If you use AI to explore fifty different creative directions in the time it would previously have taken to explore five, you are using the technology intelligently. If you use it to generate the finished ad for a flagship campaign built on decades of brand equity, you are taking a risk that the technology’s current capabilities do not justify.

The Semrush data on generative AI adoption in marketing shows that the majority of marketers using these tools are applying them to content production and efficiency tasks rather than to high-stakes creative output. That is probably the right instinct, even if it is not always a conscious strategic decision.

What AI Is Actually Good at in Advertising

There is a version of this conversation that is genuinely useful, and it does not involve flagship Christmas campaigns. Generative AI creates real value in advertising in several specific areas, and being clear about those areas is more productive than either dismissing the technology or overselling it.

Production efficiency is the most obvious one. Creating multiple variants of a display ad for different audience segments, adapting creative assets for different formats and placements, generating background imagery for product shots, these are tasks where AI tools deliver genuine time and cost savings without the reputational exposure that comes with using them for hero content. Moz has written thoughtfully about generative AI imagery in a content context, and many of the same principles apply to advertising production.

Personalisation at scale is another area where the technology has real traction. If you are running a campaign across multiple markets, languages, or audience segments, AI tools can help you adapt messaging and creative elements in ways that would be prohibitively expensive to do manually. This is not about replacing creative thinking. It is about extending the reach of good creative thinking across more touchpoints.

Concept exploration and rapid iteration during the pre-production phase is where I think the technology has the most underappreciated potential. Being able to visualise a creative direction quickly, test it against a brief, and discard it without significant cost changes the economics of the creative development process. The ideas that survive that process are better for having been tested against more alternatives.

Copy testing and optimisation is a further area with genuine commercial value. AI-driven marketing automation has been handling email subject line optimisation and ad copy testing for some time, and the underlying capability is well-established. Applying it more broadly to advertising copy, including headline variants, call-to-action language, and body copy for different segments, is a logical extension of what already works.

The Brand Equity Problem No One Is Talking About

There is a dimension to the Coca-Cola story that I have not seen discussed particularly well, and it is the one I find most commercially interesting. When you use AI to recreate a piece of brand heritage, you are not just risking a bad ad. You are potentially eroding the cultural weight of the original.

I have spent time judging the Effie Awards, which are specifically focused on marketing effectiveness rather than creative craft. One of the consistent findings across effective campaigns is that brand equity compounds over time. The 1995 Coca-Cola ad has value in 2024 precisely because it has been running, in various forms, for nearly thirty years. It is embedded in cultural memory. People feel something when they see those trucks because they have felt it before, often repeatedly, often in contexts that carry emotional weight of their own.

An AI-generated version of that ad does not carry that weight. It references it, but referencing something and carrying it are different things. If the AI version becomes the primary version that people see, the cultural memory starts to attach to the AI version rather than the original. That is not a gain in efficiency. That is a dilution of an asset that took decades to build.

This is a strategic question, not a technology question. The technology did what it was asked to do. The question is whether anyone asked whether it should be asked to do that.

How Other Brands Are Getting AI Advertising Right

The Coca-Cola story has attracted a disproportionate amount of attention partly because of the brand’s scale and partly because the backlash was vocal. But there are brands using AI in advertising contexts in ways that are generating genuine commercial value without the controversy.

The pattern I see in the more successful applications is that AI is being used to solve a specific production or efficiency problem rather than to replicate or replace a creative vision. A retailer using AI to generate thousands of product image variants for different seasonal campaigns is solving a real problem. A travel brand using AI to personalise ad creative based on destination preferences and booking history is extending a creative strategy rather than substituting for one.

Early in my career, when I was at lastminute.com, I ran a paid search campaign for a music festival that generated six figures of revenue in roughly a day from what was, by any measure, a relatively simple campaign. The insight that made it work was not sophisticated. It was just a clear understanding of what someone searching for that festival actually wanted at that moment, and a creative approach that met them there. AI tools can help you find those moments faster and at greater scale, but they cannot supply the insight itself.

The brands getting this right are the ones treating AI as a capability multiplier rather than a creative replacement. They are using it to do more with the strategic and creative thinking they already have, not to avoid doing that thinking in the first place.

What the Creative Industry Gets Wrong About This Debate

The creative industry’s response to AI advertising has largely been defensive, and I understand why. If you have spent twenty years developing craft skills in film production, illustration, or copywriting, the prospect of a tool that can approximate those skills in seconds is genuinely threatening. But the defensive response often leads to arguments that do not hold up under scrutiny.

The argument that AI-generated creative is always inferior to human creative is already losing ground. There are AI-generated images, copy, and even video sequences that are indistinguishable from human-produced work in controlled comparisons. The Coca-Cola Christmas ad was not one of them, but that is a reflection of the specific application and the current state of the technology, not a permanent ceiling.

The more defensible argument is about intent, context, and the conditions under which creative work acquires cultural meaning. A piece of advertising that carries genuine human creative risk, that represents a team of people making a call about what will resonate and why, carries something that a generated output does not. Whether audiences can detect that difference reliably is a separate question, but it matters for how brands think about what they are building over time.

For anyone thinking about AI’s role in SEO and content more broadly, Moz’s writing on AI content and E-E-A-T is worth reading. The same questions about authenticity, expertise, and trust that apply to content apply, in a different register, to advertising creative.

The Measurement Problem With AI Advertising

One thing that has not been discussed enough in the coverage of Coca-Cola’s AI experiment is the measurement question. How do you evaluate the commercial impact of a decision like this? The ad generated significant media coverage, which has a value of its own. The backlash generated further coverage. Whether that coverage was net positive or net negative for the brand is genuinely difficult to determine.

This is a problem I have encountered repeatedly across large-scale campaigns. The things that are easiest to measure are often not the things that matter most. Brand sentiment, cultural resonance, the long-term compounding effect of consistent emotional association, these are real commercial variables that are very hard to put a number on. The temptation is to focus on what you can measure and assume the rest takes care of itself.

AI advertising decisions need to be evaluated against the full commercial picture, not just the production cost savings. If using AI to produce a flagship campaign saves a significant amount in production costs but generates negative brand sentiment that takes years to recover from, the economics are not as attractive as the initial saving suggests. That calculation requires honest approximation rather than false precision, and it requires someone senior enough to make the call.

Ahrefs has explored AI tools in a marketing context with a similarly grounded perspective, and the consistent theme is that tool selection and application decisions need to be anchored in commercial outcomes rather than capability demonstrations.

What Senior Marketers Should Take From This

If you are a marketing director, CMO, or agency lead thinking about how to position AI within your advertising operation, the Coca-Cola story offers a few clear lessons that are worth sitting with.

The first is that the risk profile of AI advertising is not uniform. Using AI to generate fifty variants of a performance display ad for A/B testing carries very different reputational risk from using it to recreate a piece of brand heritage for a flagship seasonal campaign. Map your use cases against that risk profile before you commit to an approach.

The second is that the technology is moving faster than the industry’s frameworks for evaluating it. The conversation about AI advertising is still largely framed around whether the output is good enough, but the more important question is whether the process that produced it is sound. A good process with an imperfect tool will improve as the tool improves. A poor process with an excellent tool will continue to produce poor outcomes.

The third is that audience expectations are not fixed. The people who found the Coca-Cola AI ad unsettling are not necessarily the same people who will find AI-generated advertising unsettling in three years. Cultural acceptance of AI-produced creative is shifting, and the brands that figure out how to use the technology thoughtfully now will have a meaningful advantage when that acceptance reaches a tipping point.

When I first started in marketing, around the turn of the millennium, the MD at my agency refused to fund a new website build. So I taught myself to code and built it. The lesson I took from that was not that budget constraints are good, but that the people who learn to work with new tools before everyone else has access to them tend to develop a feel for what those tools can and cannot do that is genuinely valuable. The same dynamic is playing out with AI right now.

There is a broader body of thinking on AI in marketing worth exploring. The AI Marketing section of The Marketing Juice covers everything from practical workflow questions to the strategic considerations that most vendor content skips over entirely.

The Honest Assessment

Coca-Cola’s AI advertising experiment was not a failure of technology. It was a failure of application. The tools were used in a context where the margin for error was extremely narrow and the downside risk was significant, and the result was predictable in retrospect even if it was not predicted in advance.

That does not mean AI has no place in advertising. It means the place it currently occupies most effectively is not at the centre of flagship brand campaigns built on decades of emotional equity. It is in the production infrastructure, the testing frameworks, the personalisation layers, and the early-stage creative exploration that happens before a single frame goes anywhere near a broadcast slot.

The brands that will use AI well in advertising over the next several years are the ones that resist the temptation to use it for the most visible applications and focus instead on where it genuinely improves the quality or efficiency of their process. That is a less interesting story for a press release, but it is a significantly better strategy.

The application of AI to email and direct response is a good example of where the technology is quietly delivering value without the fanfare. The same discipline, applying AI where it solves a specific problem rather than where it makes a headline, is what the advertising context needs more of.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions