ChatGPT Says 'I'd Buy This' — But Its Shopping Advice Keeps Changing

If you've ever asked a chatbot which laptop or phone to buy, you're part of a fast-growing group. Researchers collected 2,528 real shopping questions from consumers and fired them at ChatGPT, Google Gemini, and Google's AI-generated search summaries. They then studied 1,536 answers in detail. Their goal: find out whether the advice is consistent, honest, and the same for everyone.

The first surprise is tone. ChatGPT talks like a friend with a favorite: it used first-person language such as "I'd go with..." in 79% of its product recommendations. Gemini did this only 7% of the time, and Google's AI Overviews just 2%. That personal-sounding framing is persuasive, but it isn't a real opinion — and since OpenAI and Google are now building advertising into their AI, there's real money riding on which products get praised.

The second surprise is inconsistency. Ask the same question twice and you may get different products. The websites the chatbots point to barely overlap either: for the same query, ChatGPT and Gemini shared only 5.4% of their cited websites, and in 76.7% of comparisons they cited no website in common at all. The behind-the-scenes version of the AI that companies plug into other apps also cited different sources than the consumer app did.

So what does this mean for you? Treat AI shopping advice as a starting point, not a verdict. Ask the same question a few times and see if the answer holds. Click through to the sources it names, and be suspicious when a chatbot says "I" — it doesn't have preferences, but someone may be paying for yours.