Claude states a statistic you intend to put in a published article. What is the sound way to verify it?

The principle — verification means reaching the source, and the useful question is whether a path to one exists.

Why the key is correct — Anthropic describes web search as expanding Claude's knowledge with real-time data, and states that every response includes citations so you can easily verify sources yourself. That gives you a link to follow rather than an assurance to accept, which is the difference between checking a claim and being told it is fine.

Why the others are wrong — asking the same source how sure it is produces another generated statement, not independent support. Repeating the question in a new conversation tests consistency, and consistent recall of a wrong figure is entirely possible. And attributing the number to an assistant does not make it accurate; the responsibility for a published claim stays with the publisher.

Remember this — a citation you can open beats any assurance you can request.

Sources — Anthropic's web search documentation.