If Chat GPT can't figure out the actual topic of a highly cited paper in the QJE, and just makes up a different, wrong, description each time, not sure this is going to work out well for helping summarize any literature.
Here's the paper and abstract I asked about.