Skip to content
AI, TECHNOLOGY & INNOVATIONTwo editions. One publication. / About us
News

The Journalist's New Colleague: What AI Can Do in a Newsroom, and What It Must Never Decide

AI can transcribe an hour of audio in minutes, translate a press release, and sift a thousand pages of data before deadline. It cannot judge whether a claim is true, protect a source, or look a reader in the eye. For journalists, that distinction is the whole job.

The Journalist's New Colleague: What AI Can Do in a Newsroom, and What It Must Never Decide

The change is already in the newsroom

This is no longer a debate about the future. In July 2014, the Associated Press began automating its corporate earnings reports. Reporters had been writing about 300 of them by hand each quarter; within a year the automated system was producing more than 3,000, and later reports put the figure above 4,400, with fewer errors than the manual versions. The most important detail is what did not happen: nobody lost their job. The time saved, around 20 percent of business reporters' time, went into investigative, explanatory and breaking news work. Managing editor Lou Ferrara said at the time that he could not have journalists spending their time on data processing; he needed them reporting. That sentence still summarises the whole subject. A global survey by JournalismAI at the London School of Economics found that more than 75 percent of newsrooms across 46 countries use AI somewhere in gathering, production or distribution. A Reuters Institute survey of 1,004 UK journalists in 2024 found that 56 percent use AI professionally at least once a week, and only 16 percent have never used it for journalistic work. The question is no longer whether AI enters the newsroom. It is already there. The question is which jobs it gets, and which it never should.

What the machine is genuinely good at

The machine is best at work that demands labour but not judgment. Transcribing an hour-long interview, translating a long press release, finding one name or figure in a thousand-page government document, monitoring social media for a developing story, summarising a lengthy report, suggesting headlines, searching an archive. In the UK survey, the most common uses were transcription and captioning (49 percent at least monthly), translation (33 percent) and grammar checking and copy-editing (30 percent). A Thomson Reuters Foundation survey of 221 journalists across 76 countries found 55.3 percent using AI to improve content and 51.8 percent for translation. This is work where the machine saves a journalist's time so the journalist can do actual journalism: meeting people, asking questions, demanding documents, checking claims.

Where it fails, and why that matters

The machine's greatest weakness is that it can be wrong with complete confidence. AI systems invent quotes, figures and citations that never existed. The Reuters Institute's 2025 report notes well-publicised cases in which generative AI made factual errors and made up or misrepresented quotes and citations. Apple News withdrew its AI-generated summaries of news alerts after inaccuracies. One wrong name, one wrong figure or one invented quote in print breaks reader trust, and trust is something that takes years to build and one story to destroy. The machine does not know that it is wrong. It has no doubt. That is why its output can never be a final verdict. It is a first draft, never the last word.

The trust gap

Audiences understand this distinction. According to the Reuters Institute's Digital News Report 2025, only 12 percent of people are comfortable with news produced entirely by AI, against 62 percent who are comfortable with news made entirely by a human journalist: a 50-point gap. In a survey of nearly 100,000 people across 48 markets, 58 percent said they find it hard to tell what is true from what is false in online news. On net scores, the public believes AI will make news cheaper (+29) and more up to date (+16), but also less transparent (-8), less accurate (-8) and less trustworthy (-18). Interestingly, audiences do want some AI uses: summarisation (27 percent), translation (24 percent) and better story recommendations (21 percent). The public is not against the machine. It only wants a human to remain in the process. Any newsroom that understands that will keep its trust.

The Pakistan test

In Pakistan, the question matters even more. According to the World Bank, Urdu makes up just 0.03 percent of Common Crawl web addresses, which means AI systems have seen very little Urdu and regional-language content. The result is more translation errors, missing local context, and mistakes in names and places. Small newsrooms have no separate verification desk, and competing with misinformation and fake videos spreading on WhatsApp is already difficult. In this environment, the journalist's verification role has not shrunk; it has grown. Pakistani journalists can use AI for translation, transcription and monitoring, but every claim, every figure and every quote must be checked against a primary source. A story published on AI's word alone will lose its credibility at the first error.

A working code for the AI newsroom

Every newsroom therefore needs a simple code. First: AI output is a first draft, never a final verdict. Every name, figure and quote is checked against a primary source. Second: never paste a source's identity, an unpublished document or sensitive material into a public AI tool, because protecting a source is a journalist's first duty. Third: where AI use matters, be transparent. The Associated Press carried a note on its automated stories saying they were generated automatically. Fourth: the byline stays human. Whoever's name is on the story is responsible for it. Fifth: take labour from the machine, never judgment. Is this true? Should we publish? Does this put someone at risk? Those decisions belong to a human being.

The question that actually matters

In the end, the question is not whether AI will replace journalists. It is who the reader will trust. Readers do not trust machines. They trust the journalist who does the hard work: reaching the source, reading the document, checking the claim, admitting an error and correcting it. AI will keep getting faster and cheaper, but it will never have judgment, courage or accountability. The newsroom that treats AI as a colleague, tireless and useful but never in charge, will survive. The one that lets it decide will lose the one thing that cannot be bought back: trust.

Sources & attribution

Reuters Institute Digital News Report 2025 and UK journalists AI adoption survey (2024); JournalismAI (LSE) "Generating Change"; Thomson Reuters Foundation Insights; Associated Press automation case via Poynter and Tow Center; World Bank MENAAP AI update 2026.

View original source
About the author

Watan Zaib Mehsood

Watan Zaib Mehsood is a journalist from South Waziristan who reports on the issues of the former FATA region and Khyber Pakhtunkhwa. He is the founder of two digital publications: The Journal Dev (thejournaldev.com), a bilingual site covering artificial intelligence, technology and innovation in English and Urdu, and PrimeUrdu (primeurdu.com), an Urdu news website. He also writes about how emerging technology is reshaping work, education and everyday life in Pakistan, and builds digital learning tools, including Medricks, an app for medical students.