With web search on, Claude gives you a market-size figure and three citations. Following them: the first page states the figure outright, the second is a blog quoting an unnamed report, and the third no longer loads. You are about to publish the number. What is the sound judgement?

The principle — citations make a claim checkable rather than checked, and three of them are not three pieces of evidence.

Why the key is correct — Anthropic's point about web search is that every response includes citations so you can verify sources yourself, which is an invitation to look rather than a guarantee. Looking is what separates these three. One page states the figure and is therefore evidence. A blog quoting an unnamed report is a claim about a source, not the source; without the report there is nothing to check. A link that no longer loads supports nothing at all, whatever it once said. So the figure currently rests on a single page, and before publishing you want to know whether that page is the origin or is itself repeating someone else.

Why the others are wrong — treating three references as a strong basis counts sources instead of reading them. Discarding the figure over one dead link throws away the one citation that did its job. And asking Claude again tests only whether the answer is stable, which it can be while remaining wrong.

Remember this — read every citation before you rely on one. The failure this guards against is not Claude inventing a source; it is a real, reachable, entirely genuine page that turns out not to say the thing it was offered for. That failure looks identical to success until someone follows the link, and if the number is going into something you publish, that someone should be you.

Sources — Anthropic's web search guidance and its article on Claude's knowledge cutoff.