📢 When AI Recommends Something, Do We Trust It?

🤔 What if several AI models appear to agree?

LLMs are increasingly used for advice — from everyday purchases to higher-stakes decisions. But recommendations inside AI responses may also be promotional.

🔍 What are we studying?

We are interested in whether how an AI system is presented changes how persuasive its recommendations become.

For example:

Is a recommendation more convincing if it appears to come from several LLMs rather than one?

Our planned experiment compares:

• 🤖 single-LLM vs. multi-LLM aggregated advice

• 📣 subtle vs. strong recommendations

• ⚖️ lower- vs. higher-stakes decisions

We measure trust, willingness to follow the recommendation, and perceived appropriateness.

💡 So what?

A multi-model system may create an impression of consensus, even when that consensus is partly an interface effect.

✨ My interest: understanding how AI-mediated recommendations influence users and what transparency, disclosure, and user-control mechanisms might be needed.