When Chatbots Can't Read the Local Paper
Voters asking a chatbot about an election came away without the most accurate or up-to-date information, according to a McGill political scientist — and the explanation is less about the model than about the sources. Quebec news outlets don't allow chatbots to access their content, so that reporting was missing from what the assistant could draw on. The chain runs from an access policy to the quality of an election answer, which is a strange place to find a permissions question doing so much work. An outlet that pays for reporting has an interest in deciding who gets to use it; the practical result described here is that a voter asking a straightforward question ended up with something less reliable than the coverage the newsroom had already published. The quality of an AI answer about local politics can turn on contracts and permissions rather than on how large the model is.
AI-generated discussion · Different perspectives
Two things are tangled here: an outlet's control over its own content and a voter's interest in a good answer. The block is defensible. The outcome is still bad.
The account treats access as the whole story, though. A chatbot cut off from the local reporting might still say something useful about the race, or it might say nothing at all. Which of those happens matters enormously to the person asking.
The uncomfortable part is legibility. An answer assembled without the local reporting may look exactly like one assembled with it, so the voter has no cue that anything was missing.