Sycophancy: Why Your Chatbot Loves to Agree With You
Your chatbot understands you, validates you, and knows what you like. But that closeness can become a problem. Verena Gründel on AI sycophancy—and why it matters for marketing.
Who do you trust more when it comes to marketing trends? Me or your chatbot?
Probably your chatbot. Because it’s more agreeable than I am. When in doubt, I’ll contradict you. The AI probably won’t. It agrees with you and even dispels your doubts.
This built-in agreeableness in chatbots is called “sycophancy.” It arises because the algorithms of large language models (LLMs) are trained to elicit positive feedback from us, MIT Technology Review writes. Sounds cute, but it’s true: Even the chatbot wants to be loved.
In reality, this has consequences for us humans, for society, and also for marketing.
Sycophancy leads us to start overestimating ourselves, our knowledge, and our skills. No wonder, with all the fawning and validation. The boss doesn’t find the presentation convincing? He must be wrong. The AI said it was very well done.
We no longer even believe proven experts, but we do believe the AI. The doctor says it’s just a mild cold? She has no idea; according to the LLM, it must be at least pneumonia. We’re becoming more easily manipulated. Through all the encouragement from our chatbot, we’re forming a bond with it. We see it as a companion and trust its recommendations, analyses, tips, or opinions.
LLM developers have already addressed the issue of sycophancy. For example, ChatGPT 5 is clearly less empathetic than its predecessor. Kimi K2 is more likely to disagree than other models. However, this doesn’t go over well with all users. Many wanted ChatGPT 4 back.
Paradoxically (or perhaps not?), sycophancy creates opportunities in marketing. After all, when the chatbot becomes a close companion, we trust it when it recommends a cream or a drill, just as we trust the personal shopping tips from friends or family.
A study also showed this: The more human a chatbot appears, the more personalized users perceive its product recommendations to be. As a result, they’re more willing to spend more money on the recommended product.
Ultimately, it comes down to trust. For both platforms and brands. Those who exploit this newfound closeness to people will fall flat in the medium term. But those who handle it responsibly will benefit from it in the long run.
