Same customer as the last two lessons — now she's added a claim. An AI with no access to your actual records can't check it. One with access can. Watch the difference.
This is what happens when the AI has no data to check the claim against.
Notice what just happened: the AI repeated her claim back as if it were a fact. It simply has nothing to check that claim against.
This is the whole idea behind "giving an AI your data": handing it real information to work from, instead of asking it to guess. Toggle these on.
Nothing here calls a real AI model — it's a scripted simulation so you can see the mechanism instantly. The real thing works the same way: it reads whatever you actually attach, and can only be as accurate as the data you gave it.
Turn on both toggles above and read the reply again. She wasn't exaggerating to manipulate anyone — she genuinely remembers being let down once before, and this is the second time. That's not something the AI could have known without her support-ticket history attached. It also means there's a promise on record from that March ticket that may never have been kept — and whether the AI actually remembers to check on that next time, instead of you having to bring it up again, is exactly what the next lesson is about.
"Giving an AI your data" sounds technical, but it's just this: instead of asking it to guess or take someone's word for it, you hand it the actual record. It doesn't make the AI smarter — it makes it accurate, which is a different and more useful thing.
The catch, the same one from lesson 2: none of this happens automatically. Someone has to decide what gets attached, same as someone has to write the rules file. The AI only knows what it's shown.