People often struggle to do what they ideally want because of a conflict between their actual and ideal preferences. By focusing on maximizing engagement, recommendation algorithms appear to be exacerbating this struggle. However, this need not be the case. Here we show that tailoring recommendation algorithms to ideal (vs. actual) preferences would provide meaningful benefits to both users and companies. To examine this, we built algorithmic recommendation systems that generated real-time, personalized recommendations tailored to either a person’s actual or ideal preferences. Then, in a high-powered, pre-registered experiment (n = 6488), we measured the effects of these recommendation algorithms. We found that targeting ideal rather than actual preferences resulted in somewhat fewer clicks, but it also increased the extent to which people felt better off and that their time was well spent. Moreover, of note to companies, targeting ideal preferences increased users’ willingness to pay for the service, the extent to which they felt the company had their best interest at heart, and their likelihood of using the service again. Our results suggest that users and companies would be better off if recommendation algorithms learned what each person was striving for and nudged individuals toward their own unique ideals.