Recommendation–Reality Gap research

How AI advice
performs in practice.

Reports about AI recommendations for restaurants, travel, products, and services. Each report identifies the category, the recommended entity, and the model when the source names it.

Explore the findings

Real-world choices, not coding or writing capability tests. Calibration remains experimental.

The dataset

Every entry links
to its original source.

—Submitted outcome reports
—Model labels
—Topics explored

Reports describe individual experiences, not a benchmark of model reliability.

Contribute to the research

What happened when you tried it?

Share a source and the outcome. Help us document what AI advice gets right and wrong.