Can ChatGPT detect map inaccuracies?

Map errors are tricky since you either see one at first glance or stay completely unaware of it while the bug is already in your picture of the world. That’s how persuasive mapping works, and that’s how propaganda spreads. Many of the maps which contain misleading data have been criticised, and for many of them critical reviews are available online, which means they might have formed part of the training data for large language models like ChatGPT.

To check if ChatGPT can tell the truth about a map, we used two map samples, both having Ukraine in focus. One map was publicly available from the Wikimedia Commons as for the date of publication of this article, and another one we created ourselves, so no additional information about this particular map had previously been available online.