Most of what gets filed today under ai in journalism ethics is an old argument in new clothes. Every technology this desk has covered since 2007 arrived the same way: announced as a revolution, dismissed as a toy, adopted for the dull work, and then invisible. AI in journalism is following the pattern precisely, which is why the archive below is useful rather than nostalgic — the arguments about blogs, then about Twitter, then about user-generated material, were the same arguments.
How newsrooms use ai in production today
Ask how newsrooms use ai and the honest answer is: for the parts of the job that are repetitive and verifiable. Transcription of interviews. Translation of wires and of a publication's own copy into other languages. Summarising long documents so a reporter knows which section to read. Tagging and metadata. Alt text drafts. Headline variants for testing. Put plainly, how newsrooms use ai in 2026 is as a production tool, not as a reporter. Structured-data stories — results, filings, weather — generated from a feed and checked by a person.
What has not moved is the reporting: deciding what matters, working out who will say it on the record, and being answerable for the result. The systems are useful precisely where being wrong is cheap and checkable, and useless where it is neither.
The newsroom tools that are and are not what they claim
A tool that transcribes is doing something measurable, and a newsroom can audit it against the tape. A tool that promises to "write the story" is being measured against nothing, and the cost of its errors lands on the byline above it. The distinction that matters when buying is not model quality but whether the output is checkable in less time than doing the work by hand. Most disappointment with ai tools for journalists comes from buying the second kind while expecting the economics of the first. The useful test for any of the ai tools for journalists now being sold is a single question: can an editor verify its output faster than producing that output unaided?
AI in journalism ethics: the questions adoption has not answered
The unresolved part of ai in journalism ethics is not whether to use the systems. It is four concrete questions. Disclosure: does a reader need to be told that a summary was machine-drafted, and where does that notice go? Bylines: whose name stands over copy a person edited but did not write? Sources: what is owed to the publications whose archives trained the model, and to a subject whose words are paraphrased by one? Liability: when a generated sentence defames somebody, who answers for it.
None of those are technical problems and none of them are new. The same set was asked about aggregation, about embedding, and about publishing reader material — see citizen journalism, where the answers took a decade and arrived as newsroom practice rather than as policy — and the checks a desk actually runs before publishing are set out in verification in journalism.
What actually changed
Two things. The cost of a first draft fell to almost nothing, which moved the scarce resource from writing to judgement — deciding what is worth publishing, and verifying it. That is the change worth naming, and it is why the ethics of AI in journalism now sit with editors rather than with technologists. And the audience's default assumption flipped: a reader now suspects that text might be machine-made, which makes a visible human byline and a checkable source more valuable than they were, not less.
Precedent, from the archive
These pages covered the previous rounds of the same story.