How to Use AI as a Journalist, Without the Retraction
How to use AI as a journalist: fast drafts from material you already gathered, and why finding the story and verifying it both stay yours.
A journalist’s work splits into two different kinds of task filed under one job title: producing a first draft from information already gathered — a results summary, a transcript turned into a quote-led piece, a recap of a routine event — and the part that makes it journalism rather than content: finding the story nobody handed you, getting a source to actually talk, and verifying a claim before it runs. A general AI tool is genuinely fast at the first. It has no access to the second, and no stake in the byline underneath it. Will AI replace journalists covers that risk-to-the-role question directly. This one is about what to actually hand over day to day, and what never to.
Where AI genuinely helps
- A first-pass transcript cleanup, turning a recorded interview into a readable, quote-led draft you then check against the recording.
- A results or earnings recap built from a company’s own release or a box score, where the facts are already fixed and the job is presenting them clearly.
- A short rewrite of a piece for a different format or audience, once the reporting and the facts are already settled.
- A first-pass summary of a long public document — a court filing, a report, a transcript — to decide quickly what in it is actually newsworthy.
- Drafting routine questions for an upcoming interview from a subject’s public record, as a starting list you then revise with what only you know about the story.
Where it will quietly get you a correction
The failure that matters most in this specific job is a fabricated detail reaching print under a real byline. A BCG and Harvard Business School study of 758 consultants found genuine performance gains on tasks inside a model’s current capability and a measurable drop on tasks that look similar but sit just outside it. Summarising a document you handed it sits inside that frontier. Judging whether the document itself is misleading, or whether a quoted source is telling the truth, sits well outside it.
A model can turn a press release into a clean write-up in a minute. It cannot tell you the press release is misleading, and it has nothing at stake if the story is wrong.
Hallucination in large language models is well documented: a tool can produce a confident, specific-sounding statistic, quote or fact with nothing behind it, and it reads exactly like a verified one until someone checks. In most office work a mistake like that is an awkward correction email. In journalism it is a published retraction, a correction notice, and a direct hit to the one asset the job runs on. Accuracy of AI: what a major study found covers the research behind why a confident sentence and a true one are not the same thing — worth reading once, because the check it describes has to happen before a word reaches a reader here, not after a correction is already owed.
A worked example
Weak: "Write up this press release as a news story."
Better: "Here is a company’s press release and the transcript of the related call. Draft a 300-word write-up using only facts stated in these two documents. Do not add a statistic, a comparison or a quote that is not in either document, and mark clearly anywhere you had to paraphrase rather than quote directly." The second version restricts the draft to material you actually supplied and separates paraphrase from direct quotation, which matters because a paraphrase can drift from what a source actually said without anyone noticing until it is checked word for word against the transcript.
The source-protection line
A source’s identity, an off-the-record detail or an unpublished story’s working angle should not go into a general consumer tool with no agreement covering how it is retained. Deciding in advance what a draft may and may not see is the core idea behind the NIST AI Risk Management Framework — a line to set before any note or transcript goes near a tool, not a cleanup after it already has. Treat an AI tool the way you would treat an intern with no confidentiality agreement: useful for a cleanly public document, never for a source who trusted you with their name.
Checking before anything publishes
Build one fixed checkpoint before a draft moves toward an editor, not after it is already filed. Checking an AI answer when you are not the expert covers that discipline in general terms. How to use AI as an editor and how to use AI as a proofreader cover the next two checks a piece usually goes through after yours — worth knowing what each pass does and does not catch, since a mechanical edit for length is not the same thing as a fact-check against your own notes.
Getting started this week
- Pick one recurring, document-grounded task first — the transcript cleanup, or the recap built from a release — because the output is instantly checkable against material you already have.
- Read every AI-assisted draft once for whether it says anything new that was not in your source material, before it goes anywhere near an editor.
- Never paste an unpublished source’s name or an off-the-record detail into a general consumer tool, no matter how routine the request feels.
- Keep a written rule for what never gets drafted by a tool at your outlet — anything naming a victim, a legal allegation, or a claim resting on a single unnamed source.
The wider pattern this fits
This sits alongside how to use AI as a PR specialist, which covers the same drafting-plus-checking habit for material written to be pitched rather than published directly, and will AI replace content writers, which finds the identical gap between fast drafting and a judgement call a model has no access to. Will AI replace translators finds the same split in a different kind of language work, where the judgement call is whether the meaning survived the crossing into another language rather than whether the facts survived the retelling.
Use AI to get fast at the drafting layer — recaps, transcript cleanups, first passes from material you already gathered. Do not let it near a source’s identity, and do not let a fluent paraphrase substitute for a word-by-word check against your own notes. Challenger, Gray & Christmas tracked 54,836 US job cuts attributed to AI in 2025 alone, concentrated in roles built around producing standard documents — a pattern worth knowing, and a reason to be the reporter who is fast with the drafting and careful with everything that reaches print.
That checking habit is also the one the labour market is already paying for: PwC measured a 62% wage premium for workers with demonstrable AI skills, up from 57% the year before. Coursium teaches that layer directly — getting fast with the drafting tools, and building the habit of checking a result before it runs under your name. Stay ahead of AI by learning the tools on your phone.