AI capabilities are improving rapidly, and companies are adopting them almost as quickly. According to the World Federation of Advertisers, 63% of surveyed multinational brands were already using generative AI in their marketing in 2024. Yet the quality and effectiveness of marketing have not improved at anything resembling the same rate. More content is being produced, but where are the corresponding improvements in customer trust and revenue?
The standard explanation is that companies simply lack AI skills. There is some truth to that, but skill alone cannot explain the disconnect. In a study involving 758 BCG consultants, access to GPT-4 improved performance on tasks suited to the technology. On a complex problem outside its capabilities, however, AI users were 19 percentage points less likely to reach the correct conclusion–even though their incorrect recommendations sounded more coherent and persuasive. Additional prompting guidance did not solve the problem. What the consultants needed was not better instructions for the AI, but better judgment about what the AI should be doing in the first place.
The first question, then, is not, “How good is your company at using AI?” It is, “What is your company actually trying to accomplish with it?”
“Your scientists were so preoccupied with whether or not they could that they didn’t stop to think if they should.” — Ian Malcolm, Jurassic Park
With enough AI automation, your company could EASILY produce 10,000 blog posts per day. But should it? Who would possibly want to read them?
AI can dramatically reduce the cost of producing content, but it cannot eliminate tradeoffs. What a company gains in speed and efficiency, it risks losing in originality and judgment. There is no free lunch in marketing–nor in life. Every decision should be assumed to carry an associated cost, especially when it is to adopt a new, utopian technology.
The problem is that many businesses are still new to AI. They are afraid of falling behind, captivated by the pace of innovation and distracted by everything the technology makes possible. In that environment, companies can become so focused on what AI allows them to produce that they never stop to ask what those supposed improvements are costing them.
Research suggests that the objective matters. In a randomized study, participants rewarded for originality used the same AI more selectively, incorporated fewer of its suggestions verbatim and produced more diverse work. More experienced AI users, meanwhile, tended to rely more heavily on its suggestions and produce more homogeneous results. Better familiarity with the tool did not automatically produce better judgment about how to use it.
If poor AI marketing were merely a skill problem, better prompting would be the solution. It isn’t. Skill determines the quality of execution. The objective determines the direction–and the behaviors, decisions, and tradeoffs required to get there.
AI is increasingly being asked to create marketing strategies and evaluate the evidence underlying major decisions. That would be perfectly reasonable if AI were actually good at those tasks. It isn’t.
One of AI’s strangest characteristics is the unevenness of its intelligence. It can accurately explain a sophisticated statistical model or effectively summarize a dense technical paper—and then misread a date, invent a source or build an entire solution on an obviously false assumption. Those mistakes won’t announce themselves.
Fluency can conceal missing evidence and circular reasoning. Time pressure, overconfidence and sometimes plain human laziness make the problem worse by reducing the critical verification required to use these tools responsibly. Research from Microsoft and Carnegie Mellon found that greater confidence in AI was associated with less critical thinking among knowledge workers. At the same time, time pressure and low motivation made users less likely to check its work.
The takeaway here? AI deserves your attention, not your trust.
To be clear, this is not an anti-AI blog post. The problem is not AI-generated work itself. AI has a legitimate place in marketing. It simply is not the universal replacement for human intelligence that it is sometimes presented as.
Research shows that AI can improve individual performance, bridge gaps between areas of expertise and accelerate genuine innovation if kept in its place. In one field experiment involving 776 P&G professionals, individuals using AI produced work comparable in quality to that of two-person teams without AI while completing it faster. Other research has found that AI can improve the novelty and usefulness of an individual’s work, particularly for people who initially perform below the group average.
The strongest results appear when AI functions as a collaborator inside a human-directed system. Combined with verified evidence and human intelligence, AI can help teams explore vastly more alternatives, identify gaps in their reasoning and virtually eliminate repetitive production work. All good things. But this kind of relationship requires the right values and safeguards. Someone still has to decide which problem is worth solving, which evidence can be trusted, what the work should mean and whether the final result serves the audience. AI can contribute to each of those decisions boundedly, but it cannot be allowed to quietly assume responsibility for them.
Many people will follow the path of least resistance unless they have a compelling reason not to, and AI has made that path remarkably easy. Addressing the problem requires incentivizing the values the company wants its marketing to reflect.
If people are rewarded for producing more, they will use AI to produce more. If they are rewarded for accuracy and originality, they will be more selective about where AI belongs. They may still make mistakes as they learn, but those mistakes can help them develop better judgment and a healthier relationship with the technology.
That process begins with open, candid conversations about AI: what it does well, where it fails, which tradeoffs are acceptable and what the company is unwilling to sacrifice for efficiency. Those conversations allow shared values to become practical ground rules.
No set of rules will anticipate every new capability or failure. The goal is to build a culture capable of making sound decisions as the technology changes. That matters because, in the end, the companies that use AI best will not be defined by their output volume or adoption speed, but by the values they refuse to trade away in the name of greater efficiency.