AI does not enter a newsroom only when it writes a sentence. It can transcribe an interview, organize documents, suggest questions, translate a draft, create subtitles, find a passage in an archive or adapt content for readers with different accessibility needs. These activities are often invisible, but they affect the time available for rarer work: speaking with sources, understanding context, checking a contradiction and deciding what not to publish yet.
OpenAI describes some uses of AI in news organizations as support for original reporting, content accessibility and relationships with readers. It is a plausible direction, but not a single model. Every newsroom has to decide where the tool can increase capacity and where it risks imitating the tone of journalism without possessing its methods. Fluent text is not verification; a fast summary is not a source; an accurate translation does not automatically understand the sensitivities of a place.
The most useful rule is simple: AI can help find and organize material, but it cannot be the source of a news story. The source remains a document, an identifiable person, a verifiable data point or an observation gathered under clear criteria. When a system proposes a claim, the journalist needs to be able to trace it back to the passage that supports it, check its date and context and distinguish a fact from a possible interpretation. This discipline does not slow the newsroom down; it prevents speed from becoming a form of error.
The deeper issue begins with the division of roles. For low-risk tasks, such as a first transcription or indexing internal files, AI can return time to the staff. For high-risk passages—accusations, sensitive data, health, conflict, attributed quotations or reconstructions of events—explicit and documented human review is necessary. Adding an editor at the end of the chain is not enough if that editor cannot access the sources, cannot correct the output or does not have enough time to do so.
There is also an accessibility question. Translations, summaries, audio descriptions and subtitles can genuinely broaden the audience for an investigation. But because they change the form of a piece of content, they need to preserve references and clearly distinguish titles, data, quotations and commentary. An inaccurate subtitle can alter the meaning of a sentence; an overconfident summary can erase the uncertainty that makes a report honest. Editorial quality is also measured by what a system does not flatten.
The relationship with readers requires the same care. Site assistants, answers to frequently asked questions and personalized pathways can help people find an article again or understand a long dossier. But they need to disclose their limits, avoid inventing references and link back to the original article when an answer depends on a verifiable passage. AI should not become a neutral voice speaking in place of the newsroom; it should make it easier to encounter the newsroom’s work.
A good protocol can be short but rigorous: a list of approved tools, rules for confidential data, permitted tasks, quotation checks, named editorial responsibility and a correction log. Provenance and credits also matter, especially for images and video. Standards such as C2PA can help preserve information about transformations, but trust comes from the whole: method, bylines, sources and the possibility of correction.
AUGMENTED NEWS does not mean automated news. AI can free time, improve accessibility and surface connections; journalism remains the work of making reality verifiable, attributed and open to discussion.