Personalised recommendations, nudges and social cues can make online shopping feel frictionless, but a study of 847 female online shoppers in India found that some AI-driven persuasive design features were associated with more addictive shopping behaviour and, through that pathway, more impulse-compulsive buying. The research was published in the International Journal of Consumer Studies in September 2026.
The finding does not mean an AI recommendation causes somebody to become a compulsive shopper. The study is observational and analyses associations reported by a specific group of consumers. Its value is more practical: it separates different design features instead of treating every use of AI in shopping as the same thing, and it finds that some forms of persuasion appear more strongly connected with problematic buying than others.
Not every AI shopping feature behaved the same way
The researchers grouped AI-driven persuasive design into features including personalisation, social influence cues, nudging, hedonic elements and conversational tools. Social influence and nudging showed the strongest positive associations with addictive shopping behaviour in the analysis. Those behaviours were then associated with impulse-compulsive buying.
Conversational AI stood out because it moved in the opposite direction. In this sample, conversational AI was negatively associated with both addictive shopping behaviour and impulse-compulsive buying. The authors suggest that conversation may sometimes introduce friction or reflection rather than simply accelerating the purchase. A shopper who has to ask questions and receive an answer may pause in a way that somebody scrolling through personalised recommendations does not.
The vulnerable users showed stronger associations
The relationships were stronger among consumers with higher psychological vulnerability and greater platform engagement. Digital financial literacy also appeared to help in a conditional way: it reduced the association between persuasive design and addictive shopping among more psychologically vulnerable consumers, but the same protective pattern was not seen across all participants.
That matters because online shops do not present persuasive features one at a time. A user may receive personalised product rankings, countdown offers, social proof, recommendations based on previous behaviour and a conversational assistant in the same session. The combined experience can feel helpful while also being designed to keep attention and move the shopper towards a transaction.
AI can optimise the moment when people are easiest to persuade
The wider concern is not that computers invented salesmanship. Shops have always used display, scarcity, discounts and social proof. AI makes those techniques more adaptive. A system can learn which product, message or timing works best for a particular user. LiveAIWire has already covered how AI is becoming part of mainstream shopping and how AI product recommendations can change what people see and choose.
Personalisation can genuinely reduce search effort. If a shopper wants size 10 running shoes under £100, filtering thousands of products is useful. The ethical line becomes harder when optimisation targets emotional vulnerability, urgency or repeated engagement rather than the shopper’s stated need. A recommendation system that knows somebody buys impulsively late at night could theoretically use that information either to protect the customer or to maximise conversion.
Another study points to the same purchase funnel from a different angle
A separate 2026 study on AI-driven personalised advertising and impulse buying also reported that consumers’ perceptions of AI-enabled personalisation, promotions and trust were linked with advertising effectiveness and impulse buying. The two studies use different samples and methods, so they should not be treated as one combined experiment. Together, however, they show why the design of AI commerce systems deserves attention beyond simple accuracy.
An AI shopping system can be technically excellent at predicting what a person might click while still producing an experience that is not in that person’s long-term interest. The performance metric matters. Optimising purchase probability is different from optimising customer satisfaction after a month, return rates, affordability or regret.
Financial literacy helped only in some conditions
The finding on digital financial literacy is particularly useful because it resists an easy solution. Education was not a universal shield. The protective association appeared among psychologically vulnerable consumers under particular conditions. That suggests consumer education can help, but platform design still matters. It would be unrealistic to place the entire burden on shoppers to recognise every personalised nudge while the system testing those nudges can adapt continuously.
There is also a connection with dynamic pricing and automated sales systems. LiveAIWire reported on pricing bots and the risk of automated price increases. Persuasion and pricing are separate mechanisms, but they can meet in the same interface. A personalised system may decide what to show, when to show it and eventually what offer to present.
The study has important limits
The sample consisted of 847 female online shoppers in India. The findings cannot automatically be generalised to men, other countries or different age and income groups. The analysis also relies on measured associations rather than randomly assigning people to months of exposure and observing whether compulsive buying develops. Psychological vulnerability and heavy platform use may influence both exposure to persuasive systems and buying behaviour in ways that are difficult to separate completely.
The strongest conclusion is therefore about design signals, not diagnosis. Social cues and nudging were the features most strongly associated with addictive shopping in the model, while conversational AI showed a different pattern. That is useful information for researchers, regulators and companies deciding which parts of an AI shopping journey deserve closer testing.
A better shopping assistant might sometimes slow you down
The most interesting design implication is almost counterintuitive. An AI assistant that always removes friction may not always serve the user. For expensive, repeated or emotionally driven purchases, a useful system might ask a clarifying question, show total monthly spending, compare alternatives or make it easy to delay the order. In other words, intelligence could be used to create beneficial friction as well as eliminate it.
That would change the success metric from how effectively AI gets a user to buy into how effectively it helps the user make a decision they are still comfortable with later. The study does not prove that this approach will work, but its finding that conversational AI was negatively associated with compulsive buying gives researchers a reason to test it directly.
The persuasive design can be changed
The findings are also a reminder that compulsive behaviour is not an unavoidable side effect of adding AI to shopping. Retail systems are designed around objectives. If the only objective is to maximise immediate engagement or conversion, personalisation and nudging can be tuned towards ever more persuasive prompts. A different design could optimise for repeat satisfaction, fewer regretted purchases or a lower rate of rapid-fire transactions. The technology does not choose the commercial target on its own.
Practical safeguards can therefore sit inside the shopping experience rather than being left entirely to individual self-control. A retailer could make sponsored recommendations easier to identify, avoid scarcity messages that are not genuinely time-limited, provide clearer spending summaries or introduce friction after an unusual burst of purchases. Those choices would need testing because a feature that helps one group may annoy another, but they show why the paper is relevant to product design as well as consumer psychology.
There is also an important causal limitation. The study measures relationships among reported experiences and behaviours within a particular sample. It cannot establish that a specific AI nudge caused an individual to become a compulsive shopper. People who already spend more time shopping online may encounter more recommendation systems, and other personal or social factors can influence both exposure and behaviour. The associations are still useful for identifying where to investigate, but they should not be turned into a diagnosis.
For shoppers, the simplest defence is awareness of the mechanism. A recommendation may feel personal because the system has learned what tends to keep attention, not because it has discovered what is genuinely best for the buyer. Pausing long enough to ask whether the product was already wanted, whether the urgency is real and whether the recommendation is paid for can interrupt some of that momentum. AI can make persuasion smoother, but the decision remains a human one.
For retailers, that creates a measurable choice. Teams can track not only whether an AI recommendation produces a sale, but whether the customer later returns the product, complains, cancels or disengages. Those longer-term signals can reveal when short-term persuasion is undermining trust. A system that sells slightly less today but produces fewer regretted purchases may be better for both the customer and the business over time.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.