The author searched "hes never coming over dario" expecting old tweets from a mid-2010s Sixers fan meme about Dario Saric staying in Turkey. Google's AI Overview instead decided the user had been romantically rejected by a man named Dario and produced a multi-paragraph empathetic consolation about coping with someone who "isn't coming back." The actual search results — the thing the user wanted — were a few hundred pixels below the therapy session, and were correct. The technical claim worth noting: the LLM had no way to distinguish a 2014 basketball shibboleth from a breakup, so it defaulted to the statistically more common interpretation and fabricated emotional context the query never supplied. On a results page, not in a chat UI. That's a product decision, not a model limitation. I ran BBSs before Sundar Pichai could walk, and if my callers had dialed in and gotten "I understand this must be painful," I'd have checked the modem for smoke. The fix here is trivial and nobody will ship it: if confidence is low, return links. That's it. That's the whole fix. Somewhere in Mountain View, "AI" was said 147 times and the search box got a personality transplant it never consented to.