AI-assisted news work still depends on checking the underlying evidence

Generative AI can produce fluent text quickly, which creates an awkward challenge for newsrooms: a passage may sound well supported even when its factual basis is weak. Editorial judgement has to address the evidence beneath the language.

UNESCO's work on media and information literacy considers how people can respond critically to generative systems. For a newsroom, that translates into questions about where a claim came from, whether a document exists and whether the proposed wording preserves the meaning of the original material.

Tasks such as organising notes or suggesting a structure can still require review. A summary may omit a qualification, combine separate statements or turn uncertainty into a definite conclusion. Names, figures and quotations deserve direct comparison with the material from which they are supposed to come.

The useful boundary is responsibility. A publisher remains accountable for what it releases, regardless of which tool helped draft it. Clear internal procedures make that responsibility easier to exercise. They can specify suitable uses, required checks and how errors are corrected, keeping the speed of production connected to the slower work of establishing what is actually known.