Why Your AI Summary Needs a Source or It Is Just a Guess

I trust a claim only as far as its citation goes.

A summary without a citation is a confident paraphrase of a model's priors, which is another way of saying a guess.

The caveat nobody mentions: Perplexity and scite changed how I work: every claim comes with a link, and scite tells me if the paper was supported or contradicted by later work.

What I would verify first: For anything I might quote, I now require a source in the output. No source, no publish. It is a boring rule that has saved me twice.

Cite it if you must, but read the method.