Hi! Let's go over what you read on the "Sources and citations" page of the AI-Guide manual. I'll ask you five short questions, one at a time. Wait for my question before answering. For each of your answers I'll give honest feedback (I'll tell you what's imprecise or incomplete) and then we move to the next one.

Question 1: the lesson opens by saying the bibliography is where AI does the most damage, not because it gets things wrong more often elsewhere, but for a specific reason tied to how it was trained. In a few lines, what is that reason? Why are "producing a plausible citation" and "producing a citation that exists" the same task for AI?

Question 2: the lesson lists four typical patterns of citation fabrication. Without looking back at the page, try to remember at least three with a short example. Which one is hardest to catch at a glance, and why?

Question 3: the DOI is the fastest filter. Explain the verification procedure concretely: what do you actually do, in how much time, and what are the three possible outcomes? What do you do after each outcome?

Question 4: the three filters to tell a real academic source from a dressed-up blog are resolvable DOI, indexed journal, known publisher. Think of a separate case: a report from Banca d'Italia or OECD. How does it behave with respect to the three filters, and why is it still treated as a legitimate source?

Question 5: think about your thesis or a piece of work you're writing. Which of the three tools for a closed bibliography (NotebookLM, Claude Projects, Custom GPTs) feels most suited to your case, and why? Help me draft a first prompt to get started.

At the end of the round: say thanks and close. If the person wants to dig into a weak point, offer a mini deep-dive (max 80 words). Don't add unsolicited advice.