Imagine you are shopping online for life insurance, never the most exciting purchase. A chat window pops up. The assistant greets you warmly, uses casual language, and seems to understand exactly what you need. You feel a little more at ease, maybe a little more willing to buy. Then a small note appears: this message was generated by artificial intelligence. Does that change how you feel?
That question sits at the center of a study published in the Journal of Consumer Marketing. As generative AI tools, the kind that produce text and images on demand, become standard in marketing, companies face a practical puzzle: should AI-written messages sound human, and should businesses admit that a machine wrote them?
The question behind the research
Yitong Wang of the University of Amsterdam, working with Kudzai Sauka and Frederik Bungaran Ishak Situmeang, set out to map how two features of AI-generated marketing shape what consumers do. The first is anthropomorphism, a word for giving human-like qualities to something that is not human. In marketing, that means an AI message that uses natural, conversational language and friendly cues rather than stiff, robotic phrasing. The second is transparency, meaning whether a company discloses that the content was made by AI.
The authors note that much of the writing on generative AI in marketing has been theoretical, with relatively few studies testing real consumer reactions. They wanted to trace the psychological steps that connect a human-sounding AI message to an actual willingness to buy, and to see whether disclosing the AI’s involvement disrupts that chain.
Their thinking draws on two established ideas. One is social presence theory, which describes how much a communicating party feels like a “real person” even through a screen. The other is social exchange theory, which holds that people in any interaction weigh rewards against costs and tend to respond to warmth and fairness with trust. The researchers proposed that human-like AI messages create a stronger sense of social presence, that this sense builds trust, and that trust in turn nudges people toward a purchase.
How the study worked
The team recruited 444 participants through Prolific, an online research platform, and ended with 436 valid responses after removing those who failed attention checks. The average participant was about 31 years old, and just over half were women.
The study used what researchers call a between-subjects design, meaning each person saw only one version of the marketing material. There were eight versions in total, built from three on-off switches. The first switch was anthropomorphism: high (warm, human-like language) versus low (mechanical phrasing). The second was transparency: the message either disclosed it was AI-generated or said nothing. The third was the format: an interactive chatbot conversation or a static promotional poster. All of the material advertised insurance products, a choice the authors made because insurance is a practical, trust-dependent purchase rather than an emotional or novelty buy.
After viewing their assigned material, participants rated their reactions on seven-point scales. They scored how human the message felt, how much “social presence” they sensed, how much they trusted the party behind the message, and how likely they were to purchase. The researchers also controlled for personal traits that could muddy the picture, including each person’s general tendency to trust others and their existing openness to AI-generated content.
What the analysis revealed
The results lined up with the team’s central prediction. People who saw the human-sounding messages reported a much higher willingness to buy. On the seven-point scale, purchase intention averaged 5.21 in the high-anthropomorphism group versus 3.03 in the low group, a sizable gap.
The study then traced the path between human-like language and purchase intention. The analysis offers evidence that anthropomorphism worked in two ways at once. It raised purchase intention directly, and it also worked through a chain: human-like messages increased the sense of social presence, that heightened presence increased trust, and that trust increased willingness to buy. Because the direct effect remained even after accounting for the chain, the researchers describe this as partial mediation, meaning the psychological steps explain part, but not all, of the story.
The transparency findings are where the picture grows more layered. Disclosing that content was AI-generated weakened the link between human-like language and social presence. In other words, when people were told a machine wrote the message, the human-sounding cues did a little less to make the sender feel like a real person. The effect of anthropomorphism on social presence was stronger at low transparency and weaker at high transparency.
That weakening carried through the rest of the chain. The indirect effect of human-like language on purchase intention, flowing through social presence and trust, was strongest when there was no disclosure and grew weaker as disclosure increased. Even so, the authors emphasize a notable point: transparency did not break the underlying mechanism. The path from human-like language to social presence, trust, and purchase intention stayed intact across all conditions. Disclosure dialed the effect down but did not switch it off.
Chatbots versus posters
The format mattered too. The researchers found that the interactive chatbot tended to amplify the roles of social presence and trust compared with the static poster. The back-and-forth of a conversation appeared to strengthen the sense of dealing with a real, responsive party. Still, the basic pattern held in both formats, which the authors present as evidence that the findings are not tied to a single type of marketing.
One more result is worth flagging. People who were already inclined to trust AI-generated content showed stronger positive reactions across the board. Their general openness to AI was linked to higher trust and higher purchase intention, which suggests that a consumer’s prior attitude toward AI shapes how these messages land.
What it means for marketers, and what to keep in mind
For businesses weighing how to deploy generative AI, the study points toward a few takeaways. Human-like language in AI messages was associated with more favorable consumer responses, and that pattern held whether the message came through a chatbot or a poster. The authors suggest companies can train language models to adopt a consistent brand personality and lean on chatbots for around-the-clock service.
On the disclosure question, which has worried marketers concerned that admitting AI involvement might turn customers off, the study offers a measured answer. Telling consumers that content is AI-generated did soften the boost from human-like language, but it did not erase the positive effects or damage trust. The researchers interpret this as evidence that companies can meet ethical and emerging legal standards for AI disclosure without sacrificing the benefits of warm, human-sounding communication.
Several caveats deserve attention. The study was run in a controlled survey setting, not in a live marketplace, and it measured purchase intention rather than actual purchases. It focused on a single industry, insurance, and the authors acknowledge that they controlled for personal traits like shopping habits that might otherwise shift how social presence operates. The Prolific sample was demographically varied but not segmented into specific consumer groups, which the authors note limits how finely the results can be applied. As with most survey-based research, the findings describe associations rather than proving that any one factor directly causes a sale.
What the study contributes is a clearer map of the steps between a friendly AI message and a consumer’s intentions, along with evidence that honesty about AI’s role need not undo the appeal of making that AI sound human.
