AB Angelo Baccelloni Versione italiana
Blog

When the algorithm is too good

2025 · AI & choice · Source: Journal of Interactive Marketing (2025) · doi.org/10.1177/10499968251358181

A highly specialized recommender feels ideal: it knows your taste and always shows "the right thing". My research shows a hidden cost: when the agent is too narrowly tuned to your past, it narrows options and reduces discovery.

For companies, the lesson is to design recommenders with some novelty and "unlearning" built in — not only relevance, but variety, so the customer isn't locked in a bubble.

Read next

Weak ties, strong influence

Do you trust a human or an AI more?

Reading the storm before it breaks

← All articles