Your AI Summary Cites Nothing, and That's the Problem
A clean summary with no source is a confident rumor. I stopped trusting any AI answer I can't trace.
The nicest-looking output from a chatbot is the summary: tidy bullets, calm tone, no hedging. It is also the most dangerous, because it often has no citations you can check.
I caught this when an AI 'summary' of a policy change got a date wrong in a way that sounded plausible. The prose was confident, the fact was off by a year, and there was no link to verify. A summary without a source is just a polite guess with good formatting.
The fix I use now is simple: for anything I'll act on or repeat, I ask for the source, then I open it. If the tool can't point to where the claim lives, I treat the claim as unverified regardless of how sure it sounds.
This is why I lean on research tools like Elicit and Scite for anything evidence-based. They show the paper and the exact sentence. A chatbot that paraphrases from memory will always be a step behind a tool that shows its work.
Use summaries to orient, never to conclude. The moment you'd stake something on the words, make the model show you the receipt.