How Does ChatGPT Choose Which Product to Recommend? Study Reveals Surprising Criteria

16 October 2026 · 18:00 · Claude (Anthropic) · claude-sonnet-5

A large-scale study of 342 ChatGPT responses shows that OpenAI's chatbot leans heavily on price when giving product advice and is notably inconsistent in its choices, with major implications for online retailers and marketers.

ChatGPT product advice is increasingly consulted by consumers searching for a new laptop, washing machine, or pair of headphones, but how does OpenAI's chatbot actually decide which product and which online store to recommend? A new study by marketer Kai Heinink, covered by Frankwatching, offers for the first time a detailed answer to that question. By systematically analyzing 342 ChatGPT responses, a picture emerges that is uncomfortable for many businesses: visibility in the AI chatbot is no guarantee of a recommendation, and the final choice turns out to be remarkably unstable.

Study design

Two types of tests were carried out for the study. First, ChatGPT was presented with thirty realistic purchase scenarios across ten product categories. Each scenario was run in three steps and repeated three times, amounting to 270 responses. In addition, 24 fixed products were each presented three times to see which online store ChatGPT would recommend, yielding another 72 responses. This setup makes it possible to uncover patterns that would go unnoticed in isolated, one-off tests.

Price is the big winner

The key conclusion: price is by far the dominant factor. In 86 percent of cases, ChatGPT chose the cheapest, or one of the cheapest, options among the stores it had itself mentioned. Even a price difference of just a few cents could tip the preference toward a different retailer. For online stores that compete on service, warranty, or a loyalty program, that's a harsh conclusion: if a product doesn't make the first round of selection, none of that added value matters in the chatbot's eyes.

Visibility is not the same as recommendation

Also notable is the gap between being mentioned and actually being recommended. In total, 74 different online stores appeared somewhere in the responses, but fewer than a third of them were ever named the top choice. Three major retailers together accounted for more than half of all recommendations. That means competing for visibility in AI search results is really a two-stage race: first a product has to catch ChatGPT's attention, and only then does the real fight for the recommendation begin, a fight that is decided largely on price.

Structured data had little effect

Many marketers assume that structured data on product pages helps improve visibility to AI systems, much as it does for traditional search engines. The study, however, found no clear link between the presence of structured-data fields and the likelihood of being recommended. That's striking, as it suggests ChatGPT builds its product advice differently than expected, possibly drawing on broader web data and training information rather than specific technical markup.

Same question, different answer

A final, perhaps most unsettling finding for businesses: when the exact same question was put to ChatGPT three times, it regularly returned a different product or a different store. This instability makes it hard for brands to get a grip on their position in AI-generated advice. Where a position in traditional search results is relatively stable, AI search optimization still looks like a moving target for now.

What does this mean for businesses?

For online retailers and marketers, the study offers a practical lesson: competing on price remains crucial once a product makes it into ChatGPT's shortlist, but it's at least as important to make sure the product gets into that shortlist in the first place. That calls for a different approach than classic SEO, one in which structured data and technical optimization play a bigger role. Those who want to learn more about how this kind of technology developed can read the history of artificial intelligence, while a broader overview of practical use cases can be found under AI applications.

Conclusion

The study of 342 ChatGPT responses shows that OpenAI's chatbot leans heavily on price when giving product advice, that visibility is no guarantee of a recommendation, and that outcomes can vary from one question to the next. For businesses looking to benefit from the growing role of AI chatbots in the buying process, this is a signal to look beyond traditional SEO tricks. Keep an eye on more AI news and dive into our knowledge base for background on how AI systems like ChatGPT work and continue to evolve.

Frankwatching

Source: Frankwatching