Personalization Discovers the Desire; It Does Not Create It.
The argument in favor of using AI in advertising is growing stronger. Yet the assumption on which it is based, that personalization can create purchasing intent, might be completely the wrong one.
A study involving 324 young adults found that, across 20 or more behavioral, cultural, and technological indicators, whether someone had already wanted to buy something was the main factor in determining whether AI-powered advertisements would lead them to purchase. In contrast, the other factors were merely secondary noise.
A basic assumption underlying almost all AI advertising pitches is that personalization, when both accurate and widely applied, can create desire that didn't exist before. The reasoning is appealing: if the appropriate message reaches the right person at the right time, the advertisement need not interrupt; it will be relevant. And it is claimed that relevance leads to a purchase: a person who originally hadn't considered buying a product becomes a buyer after being presented with an offer that is both well-timed and closely aligned with the situation.
This idea forms a major part of the current justification for investing in AI marketing. It explains the focus on lookalike audiences, intent-signal targeting, behavioral prediction models, and personalization engines. If AI can create desire, or at least trigger latent desire, in ways no conventional advertisement could, the value of such an investment reaches a ceiling well above what could be achieved through traditional media efficiency.
The premise is likely incorrect. Or, at the very least, it is wrong in the way that matters most for money allocation.
Researchers who looked at AI-powered advertising among young adults in India developed a particularly thorough model. Based on survey data from 324 participants aged between 18 and 26, the study examined more than 20 behavioral, cultural, and technology-related factors that predict purchase intention—including AI perception, cultural influence, collectivist values, societal attitudes towards AI, access to technology, cognitive reactions, confusion about AI systems, and others. To do this, they used a random forest classifier, checked its results with SHAP and LIME analyses to confirm which variables were responsible for the predictions, and then examined the actual factors that influenced purchase intent regarding AI advertising.
The intention to buy beforehand was the most important factor. It wasn't just one of several factors, but the main predictor, the one the model always referred to first and the variable that had the greatest predictive power when the analysis was examined. Although there were twenty or so other indicators, including the technological and cultural factors most commonly used in AI advertising strategies, they contributed only a little. The model reached a moderate level of overall accuracy, about 66 percent, a figure that is worth mentioning: even at its best, the strongest signal the model had to work with was whether or not the person had already intended to buy an item in that category. What AI advertisements did most effectively was target people who were already inclined to make a purchase and then give them a final push.
The study has certain limitations in scope. The sample in question comprises young adults in India, a cultural setting characterized by specific attitudes toward AI novelty and strong collectivist social norms. The methodology records intended actions rather than actual transaction behavior. It is not possible to draw definite conclusions from a single study. Yet the direction of the result aligns with what behavioral economics has long asserted about desire and decision-making, and it has direct relevance to how AI advertising strategy is currently being formulated.
The twenty-something indicators covering aspects such as access to technology, cultural attitutes, and AI novelty made only a minor contribution. The main factor the model used was whether the person already intended to buy something.
Amplifier vs. Generator
The key point of this study is the distinction between amplification and generation.
An amplifier takes a signal that is already present and makes it stronger, clearer and more efficiently received. A generator, on the other hand, produces a signal from nothing.
The majority of what AI advertising can do, according to this research and in light of the wider behavioral literature it draws on, consists of amplification. It identifies people who already have or are actively developing a desire to buy, and reduces the gap between that desire and a purchase. This is truly valuable and is, in a strict sense, different from creating a desire to buy in people who originally had none.
The fact that there is a category error is important since it alters the point at which leverage can be applied. When AI advertising is mainly acting as an amplifier, the key strategic question becomes not 'how can we achieve greater personalization?' but 'how can we create a more genuine desire for the category and the brand, so that there is more signal for the amplifier to act upon?' Regardless of how advanced the personalization engine is, it cannot do better than the underlying desire that is available to it; if that desire is absent, the targeting process will identify people who are not yet ready to make a purchase and then show them highly relevant advertisements for things they do not currently want.
This has the same effect on the value of brand-building activities. There has always been a tension in marketing budgets between brand and performance: brand builds awareness and sentiment over time, while performance drives immediate conversion. Performance spending has increased because it is measurable, traceable to specific sources and directly linked to revenue results. AI has accelerated this growth by making performance channels more efficient. However, if the performance channels mainly amplify existing intent rather than create new desire, then failing to fund the earlier-stage work that is responsible for generating that desire results in a fundamental issue: the personalization engine becomes ever more efficient at targeting a group of people whose intent is no longer being renewed. The system thus works itself into a ceiling of demand.
What This Changes
Nearly all AI advertising briefs are drawn up to justify the money spent on personalization. They pay close attention to audience segmentation, message optimization, delivery timing, and conversion efficiency. These do constitute real advantages, and there is genuine value in enhancing them. Yet they are levers which act upon a desire that already exists. The brief seldom inquires into the extent of desire the brand is currently creating through its content, positioning, presence in culture, word of mouth, and the experiences that existing customers have accumulated—since that question is more difficult to answer and assign to a particular campaign.
The useful summary covers both aspects.
What is the present level of purchase intent in the category and what proportion of that intent do we hold? Is that level of intent increasing or decreasing?
What actions are we taking to create new desire rather than just fighting for intent that is already there?
When AI-powered personalization is used, does it speed up conversion among those already in the consideration process, a group for whom it works well, or does it instead try to induce intent in people who aren't yet ready? In this group, research indicates it works less reliably.
There is a geographical and demographic limitation that is important to note. The cultural views of young adults in India, their access to technology, and the norms related to collective decision-making do not translate directly to other contexts. It would be wrong to generalize the findings regarding what does or does not influence this group without proper qualification. However, the result that previous purchase intention dominated a predictive model consisting of twenty-one variables, outperforming factors such as technology access, cultural attitudes, the novelty of AI, and societal perceptions, is not as culturally specific as a finding concerning, for example, collectivist values. The idea that intent predicts intent is a widely supported behavioral pattern in the literature.
Strengthen the desire that you have already created and continue to do so. AI can identify the people who are taking action, even though it is not the reason that first prompts them to act.
Some ideas are worth discussing in the context of your organization.
