Telling people more clearly that an advert is being shaped by AI can change how that advert affects purchasing decisions, but the effect is not as simple as “more transparency means less persuasion”. A peer-reviewed study of 609 young adults in China found that AI transparency changed the relationship between people’s cognitive assessment of an in-feed advert and their intention to buy. The Scientific Reports study also found that intrusive advertising damaged both emotional and rational attitudes towards the advert.
The work focuses on the kind of advert people encounter while scrolling through social commerce feeds. These systems can select products, creative material and timing using behavioural data. For the person holding the phone, the result may look like an ordinary post or recommendation even though algorithms have decided why that item appears at that moment.
The researchers separated feeling from thinking
The study modelled two kinds of response. Affective attitude covers the emotional reaction to the advert. Cognitive attitude covers the more deliberate judgement of whether the advert seems useful, credible or worth considering. Purchase and sharing intentions were then analysed in relation to those attitudes.
Entertainment, personalisation and incentives were positively related to affective attitude in the sample. Informativeness, personalisation, incentives and credibility were positively related to cognitive attitude. Intrusiveness moved the other way, showing a negative relationship with both emotional and cognitive attitudes.
That distinction matters because an advert can work through different routes. A personalised video might feel enjoyable even when it adds little factual information. A clear product comparison might persuade through usefulness rather than entertainment. AI can optimise both routes, but consumers may react differently when they understand that automation shaped what they saw.
Transparency changed the rational path to purchase
The study found that AI transparency significantly moderated the link between cognitive attitude and purchase intention. It did not show the same moderating effect on the link between affective attitude and purchase intention. In other words, disclosure appears to interact more with the thinking route than the feeling route in this dataset.
That is a more subtle result than the usual debate about whether AI labels make people trust content more or less. A disclosure can become additional information that the consumer uses when judging the advert. It does not necessarily neutralise the emotional appeal of the creative material.
The same disclosure can mean different things to different users
Transparency also sits inside a wider trust problem. LiveAIWire has reported that people can keep giving AI systems data even when they feel uneasy about it and that public trust changes when AI is involved in decisions. Simply adding the words “AI-powered” does not tell a user which data were used, whether a person reviewed the content or what part of the advert was generated or targeted by AI.
A useful disclosure therefore needs to be specific enough to answer a real question. Was AI used to create the image? Did an algorithm choose the product based on browsing behaviour? Was the price personalised? Is the recommendation sponsored? Those are different interventions with different implications for the consumer.
The sample was specific
The research used an online questionnaire completed by 609 Chinese young adults and analysed the responses with a statistical model. That gives the study a substantial sample for the design, but it does not prove that the same relationships apply to older consumers, other countries or different kinds of shopping platform. Attitudes and purchase intentions are also not the same as observing actual purchases over time.
The article was published as an early-access peer-reviewed version and carries a permanent DOI, although the publisher notes that final editorial changes may still occur before the version of record replaces it. The authors report no competing interests.
Intrusiveness may be the more immediate warning sign
For shoppers, one of the clearest results is not actually about AI labels. Intrusive advertising was negatively associated with both emotional and cognitive attitudes. That suggests personalisation has limits. An advert that feels too invasive can damage the experience even if it is highly relevant.
That finding connects with LiveAIWire’s reporting on personality-matched AI messages and higher click rates. Optimisation can improve a short-term metric while creating a longer-term trust cost. A system that becomes better at predicting what gets a click may also become better at crossing the line where relevance feels like surveillance.
Why this matters for AI advertising rules
Regulators and platforms increasingly talk about transparency as though it were a single switch. The study suggests designers should think more carefully. A label may help users interpret the rational basis of an advert, but it may not change the emotional pull of entertainment, incentives or attractive creative work. Disclosure therefore needs to sit alongside rules about data use, targeting and deceptive presentation.
For advertisers, there is also a business reason to avoid overreach. If intrusiveness harms both the feeling and thinking sides of the response, more data and more personalisation do not automatically produce a better advert. The optimum may involve leaving some distance between what the system can infer and what it chooses to reveal.
An AI label is useful only if it answers the right question
Consumers will see more algorithmically selected content as shopping becomes more conversational and automated. LiveAIWire has covered the rise of AI-assisted shopping and the same trend will make provenance harder to see. A recommendation can be generated by one model, ranked by another system and paid for by a merchant, all inside the same interface.
The study does not provide a universal disclosure formula. It does show that transparency can change one part of the persuasion process and that intrusiveness can damage several parts at once. The practical challenge is to tell people enough about the role of AI to make an informed judgement without turning every advert into unreadable legal text.
That makes transparency a design problem as much as a compliance problem. A useful disclosure should help a shopper understand why this product is in front of them, what data helped put it there and whether money changed the ranking. If it cannot answer those questions, simply declaring that AI was involved may be technically true but practically uninformative.
Transparency needs to answer a useful question
That direction is consistent with the OECD’s principle on transparency and explainability for AI, which calls for meaningful information that helps people understand when they are interacting with AI and, where appropriate, the factors behind an outcome. Advertising is a useful test case because a technically correct disclosure can still leave the most important commercial question unanswered. A shopper may care less that machine learning was used than whether their behaviour, location or purchase history affected the recommendation.
A meaningful disclosure could therefore be layered. The first level might simply identify that a recommendation is personalised or sponsored. A second level could explain the broad signals that influenced it, such as previous purchases or the current search. More detailed information could remain available for people who want it. That is different from forcing every user to read a dense explanation before seeing an advert, and it gives product teams room to test whether clarity can be improved without overwhelming the interface.
The distinction between the study’s cognitive and affective routes also matters. People can enjoy an advert while making a separate judgement about whether its claims are credible. Transparency appeared to change the latter relationship in this sample, which suggests disclosure may be particularly relevant when consumers are consciously evaluating whether to act. It should not be treated as a magic label that neutralises every persuasive technique.
There are limits to how far the result can travel. The participants were young adults in China, and advertising norms, privacy expectations and platform habits differ across countries and age groups. The experiment also captures a structured research setting rather than every kind of real-world advertising system. Even so, the central design question is widely applicable: if AI influences why a person sees a commercial message, can the person understand that influence well enough to make an informed choice?
The same principle applies when the system is wrong. A useful transparency design should give people some route to question or correct the assumptions behind a recommendation. If a shopper keeps seeing irrelevant products because an old purchase has been misread as a lasting preference, knowing that personalisation exists is only half the solution. The person also needs a practical way to change what the system thinks it knows.
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.