When Google's AI search summary gets facts wrong, does the law treat that differently than a company chatbot that lies to a customer?

This explores whether AI Overviews in search get different legal treatment from customer-facing chatbots. The short answer from the collection is that the two start from different defenses and end at the same conclusion. Neither the summary nor the chatbot is treated as a neutral middleman. Whatever it says counts as the company speaking.

The chatbot cases are about a company and its own tool. In the Air Canada case, the airline argued its chatbot was a separate entity responsible for its own words. The tribunal rejected that and applied ordinary tool-liability reasoning: if you deploy it, you own what it outputs Can a company escape chatbot liability by calling it separate?. Germany's OLG Hamm went further. A clinic's chatbot made false claims about doctors' credentials, and the court said it made no difference that the bot was programmed correctly and trained on accurate data. The false statements were attributed directly to the clinic Can companies escape chatbot liability through careful training?. So in chatbot doctrine, the question is simply whose tool it is. Being careful doesn't get you off the hook.

The AI Overview question is different, because Google has a defense chatbot operators don't: search engines have traditionally been treated as indexes that point to other people's content, not as authors of it. The Munich Regional Court's key move was to say that an AI-generated summary is not an index. It is Google's own content, which makes Google a direct infringer when the summary is false, and Google is expected to check what its algorithm produces Can courts hold Google liable for AI Overview false claims?. In other words, the fight in search is over whether the output is speech at all, or just a pointer to someone else's speech. Once a court decides that writing a summary is authorship, the result looks just like the chatbot cases.

Philosophy helps explain why courts keep landing here. One argument in the collection holds that AI systems can't really make claims, because a genuine assertion requires someone who can be held to account for it. Fine-tuned models can't be sanctioned, and they don't even work as stand-ins for the people who deploy them Can AI systems genuinely assert claims about the world?. If the system itself can't answer for what it says, someone has to, and courts in both settings are assigning that responsibility to the company. Sacasas's worry that handing language over to machines lets people dodge responsibility for their words Does AI language generation undermine human judgment and responsibility? is, in effect, the move these courts are blocking.

One caveat: the collection covers only three cases, all from Germany and Canada. Nothing here addresses US platform-immunity law (Section 230), which is where the search-versus-author distinction matters most. There is also nothing on what happens when assistants take actions rather than just answer questions, which DeepMind's ethics work flags as a different kind of problem What makes ethics of AI assistants fundamentally different from chatbots?. That is probably where the next round of liability questions will come from.