AI and the Future of JournalismHow Newsrooms Work and Where Readers Go

Newsrooms were using automation before generative AI

The relationship between journalism and automation did not begin with ChatGPT. In 2014, the Associated Press began using automation to produce corporate earnings stories from structured financial data. AP later expanded automated coverage to selected sports previews and recaps. These systems addressed a practical problem: a newsroom has limited time, but financial results and sports scores arrive in large volumes and often follow repeatable formats. AP’s account of its newsroom automation places those early earnings stories nearly a decade before generative AI became a mainstream newsroom discussion.

Automation can also help journalists decide where to look. For its Doctors & Sex Abuse investigation, The Atlanta Journal-Constitution collected more than 100,000 disciplinary documents from US regulators. The team used a machine-learning program to rank cases by their likelihood of involving sexual misconduct. Journalists then read the documents in more than 6,000 cases and carried out further reporting, including interviews and records research, as the AJC’s methodology explains.

That distinction matters. The program helped the team navigate a collection too large to examine in the same way by hand. It did not determine the final story or replace the reporting needed to establish what had happened.

The Panama Papers offer another useful lesson, although the technology is often described too broadly. The original investigation relied on tools that helped journalists organise, search and connect information across 11.5 million leaked documents. In 2019, the International Consortium of Investigative Journalists explored how machine learning might help uncover additional stories in the material. Its account of that research makes clear that searchable databases and collaborative reporting were already central to the original investigation.

Today’s generative AI can produce fluent text, images, audio and code. That creates different possibilities from a tool that sorts documents or fills a template. It also creates a different problem: an answer can sound convincing even when the underlying fact, source or quotation is wrong.

How major news agencies use AI today

International news agencies work across languages, time zones and formats. They also supply reporting to other news organisations, financial clients and businesses. Their AI tools therefore need to support speed and scale while fitting into an editorial process that clients depend on.

Reuters: faster workflows, with editorial review

Reuters has described several tools designed around specific newsroom tasks. Its translation support tool, CheckMATE, helps translators identify potential issues in machine-translated copy. Its FactGenie tool assists with alerts and text annotation.

In a July 2026 account of its newsroom tools, Thomson Reuters said these systems produce starting points for journalists, who edit and verify the results. The company also described involving journalists in tool development, including when an alert suggestion missed a news judgment or a translation needed the right tone.

That last detail is important. A technically accurate translation may still lose meaning if it mishandles political context, an idiom or the way a source intended to speak. The task is not merely to turn one set of words into another; someone has to assess whether the result says the right thing.

AFP: translation, transcription and source protection

AFP has developed its own AI tools, including AFP Translate and AFP Scribe, which it identified in a 2026 announcement about its editorial innovation leadership. Its 2025 editorial standards set out why an apparently routine task such as translation still needs close review. Names, acronyms, places, idioms and cultural sensitivities can all be mishandled. The standards require a journalist fluent in the language to check translations and warn against uploading sensitive interview recordings to transcription applications when doing so could put sources at risk.

The source-protection point can be overlooked when AI is discussed mainly as a productivity tool. An interview may contain information that a reporter is obliged to keep confidential. Before uploading it to any service, the newsroom needs to understand where the data goes, who can access it and whether its sources have been adequately protected.

Associated Press: AI assistance has limits

The Associated Press updated its newsroom AI standards in July 2026. Its permitted uses include assistance with early research, document summaries, transcription, translation and suggestions for headlines or story summaries. AP says journalists review and edit the output before publication.

The agency continues to prohibit generative AI from creating, altering or enhancing news photographs. That boundary reflects the role of a news image: readers reasonably expect it to document an event, not add details that were never there.

These policies do not mean every newsroom will draw the same line. They do show why a useful AI policy needs to be more specific than “a human is in the loop.” It should explain which tasks are permitted, what must be checked, what must be disclosed and who has authority to publish.

How Asian media are using AI

Asian news organisations face many of the same questions as their counterparts elsewhere, but they also work across markets where language, platform habits and audience expectations can differ substantially. Examples from Hong Kong and Singapore show how publishers are applying AI to editorial work and reader-facing products.

South China Morning Post: assistance across editorial work

The Hong Kong-based South China Morning Post says in its published policies and standards that it uses AI to assist journalists with work including research, data collection and assessment, translation, illustrations and artwork. The publication also states that its journalism will not be solely AI-driven and that humans remain responsible for editorial and creative decisions. That is a stated editorial policy; assessing how well it works in practice would require looking at individual workflows, published disclosures and corrections.

The distinction between illustration and news photography is particularly relevant. An illustration may be created to explain an idea, while a photograph can serve as evidence of a real event. If AI is involved, readers need enough context to understand what they are looking at.

The Straits Times: bringing AI to readers

Some publishers are using AI behind the scenes. Others are making it part of the product a reader interacts with.

In 2025, The Straits Times introduced a career chatbot developed with SPH Media and support from Amazon Web Services. At launch, the publication said the chatbot drew on an archive of more than 5,000 articles about jobs, careers and working life.

The idea is different from asking a journalist to use AI to prepare an article. Here, the reader asks a question directly and receives an answer based on a publisher’s existing material. A useful follow-up question is whether the experience helps readers understand where the answer came from and gives them a reason to explore the original reporting.

The Reuters Institute’s 2026 Singapore market report also notes that SPH Media introduced an AI-powered newsfeed and short article summaries. It reports that CNA has developed an investigative unit using AI, data analysis and open-source intelligence tools in its work on disinformation.

These examples point to several uses of AI within the same market: helping journalists investigate, changing how stories are presented and letting audiences interact with an archive. They should not be treated as one activity simply because they share the label “AI.”

Is AI taking readers away from news publishers?

There is evidence that some people are using AI to get news and that AI-generated answers can reduce visits to websites. There is not yet a simple figure that tells us how much of every publisher’s readership has been lost to AI.

The Reuters Institute’s Digital News Report 2026 found that, across the markets it surveyed, 10% of respondents had used a standalone AI chatbot for news in the previous week, up from 7% in 2025. Only 1% said AI was their main source of news.

The institute found that usage was stronger among younger people and people already highly interested in news. For many respondents, a chatbot appeared to be an additional way to explore a story rather than their only source of news.

What people ask AI to do also varies. The Reuters Institute found that respondents in Taiwan and South Korea often used chatbots simply to get news, while making news easier to understand ranked highly in Japan. In Hong Kong, using AI to evaluate news sources was among the more prominent reported uses.

A reader may turn to a chatbot for a short explanation, a comparison of reports or a translation. Those are useful services. But if the answer draws on someone else’s reporting, the organisation that gathered the information may never see the reader.

AI search summaries can change the click

The effect is not limited to people opening a chatbot. AI-generated summaries can appear within a familiar search experience.

In an observational study of Google searches by 900 US adults, the Pew Research Center found that people clicked a traditional search result on 8% of visits where an AI summary appeared, compared with 15% of visits where no summary appeared. Users clicked a link in the AI summary itself on 1% of visits to pages with such a summary.

The study covered Google searches across topics, not only news. It also compared observed searches, which can differ in their subject and intent, so the figures alone cannot tell us that AI summaries caused every difference in clicking.

There is now experimental evidence as well. In a 2026 preprint involving 1,100 participants, researchers found that removing Google’s AI Overviews and AI Mode increased click-through rates to publishers. That supports the concern that AI search features can change where attention goes, although one experiment cannot measure the effect on every publisher or every kind of story.

The shift away from publisher websites is bigger than AI

AI entered a news environment that was already changing. Across the 48 markets in the Reuters Institute’s 2026 report, 54% of respondents said they used social media and video networks for news, compared with 51% who used news organisations’ websites and apps. People can use both, so those figures should not be read as two separate groups.

The institute also reported Chartbeat data showing that organic Google Search traffic to more than 2,500 news sites fell by 33% between November 2024 and November 2025. That is a substantial decline, but it does not establish that AI alone caused it. Changes in search results, platform distribution, reader habits, news events and publisher strategies can all affect traffic.

For a publisher, the practical concern is cumulative. A reader might encounter a story on social media, receive an AI summary in search and never develop a habit of visiting the organisation that reported it.

Who pays for the reporting behind an AI answer?

Reporting has costs that are easy to miss when the result arrives as a few sentences. A newsroom may spend days checking records, travelling to a location, interviewing sources, obtaining legal advice or correcting a developing story. A short answer can make the information easier to consume, but it does not remove the cost of finding and verifying it.

When fewer people visit a publisher, the effect depends on its business model. An advertising-supported site may lose opportunities to show ads. A subscription publisher may have fewer chances to introduce its journalism to potential subscribers. A smaller local newsroom may be especially exposed if it relies heavily on search to reach readers who do not already know its name.

Some organisations are exploring licensing as another source of revenue. In 2023, for example, AP and OpenAI announced an agreement under which OpenAI licensed part of AP’s text archive and AP gained access to technology and expertise.

Licensing can create a business relationship between a news organisation and an AI provider. Whether it adequately supports reporting depends on the terms, the value of the content, the reach of the agreement and what happens to the publisher’s wider audience. A large international agency and a small local publication may have very different negotiating positions.

There is also a question of attribution. A link can help a reader inspect the source, but a citation alone does not guarantee that the reader will click, subscribe or remember the organisation that did the work.

AI can support verification, but it cannot take responsibility for it

A newsroom may use AI to search a large collection, flag an inconsistency or identify material that deserves closer attention. Those are useful starting points. They are not the same as establishing that a claim is true.

An AI-generated summary can omit a qualification. A transcription can mishear a name. A chatbot can invent a quotation or present an outdated claim confidently. The risk is not only that the technology makes a mistake; it is that a plausible answer makes the mistake harder to notice.

An Associated Press case study of a local-news AI project at KSAT-TV reported that summaries were sometimes inaccurate and could invent quotations. The finding illustrates why the original recording or document remains important even when a tool appears to save time.

Verification also requires judgment about context. A statement may be accurately quoted but misleading without a date, an opposing response or an explanation of how a number was calculated. A journalist has to decide what evidence is sufficient and what a reader needs in order to understand the claim fairly. For news organisations adopting AI, useful questions include:

• Can the journalist trace the output back to the original source?
• Has a person checked names, figures, quotations and important context?
• Is confidential material being uploaded to an approved system?
• Does the audience need to know how AI was used?
• Who is responsible for correcting the work if it is wrong?

What happens to journalists’ jobs?

AI can change a journalist’s workload without immediately eliminating the role. It may reduce time spent transcribing interviews or searching routine documents, while creating new work in checking outputs, managing tools and explaining their use to audiences.

In a survey published in the Reuters Institute’s 2026 industry report, 67% of respondents said their organisations had not reduced roles as a result of AI, 16% reported a handful of staff or freelance cuts, and 9% said jobs had been added. The survey involved 280 selected media leaders across 51 countries and territories and was conducted in late 2025; it is not a representative count of all newsroom jobs.

There are also individual cases where AI forms part of a wider restructuring. In August 2026, US broadcaster E.W. Scripps reported the elimination of 268 jobs across the company as it pursued a transformation involving AI and automation. Those were company-wide cuts, not a count of journalists replaced by AI; the company also described broader changes to its operations and costs.

The employment pressure did not start with generative AI. In a WAN-IFRA and Nanyang Technological University survey of 42 media professionals in 12 Asian markets, 19 respondents said editorial headcount at their organisations had fallen during 2024, while 14 reported an increase and nine saw no change. The sample is small and does not establish that AI caused those changes.

Work can also become broader for journalists who remain. In Muck Rack’s 2026 survey of nearly 1,100 journalists, 62% said their responsibilities had expanded beyond their core role over the previous year. The survey did not establish that AI caused that expansion, but it highlights the importance of asking whether time saved will support deeper reporting or simply lead to more work.

What publishers and other content creators can do

News organisations cannot control every place where their reporting appears. They can still decide what makes their work valuable and how they build a relationship with the people who rely on it.

Original reporting is one part of that value. A publisher that produces interviews, local knowledge, specialist analysis or an investigation contributes something that cannot be recovered from a generic summary alone. That does not guarantee a visit, but it gives readers a reason to seek out the source.

The next step is to make that source useful once people arrive. Can a reader see how the information was gathered? Are important claims linked to evidence? Can they follow a developing story, understand the local context or find the full interview? Does the publisher offer a newsletter, app or other way to return without depending on an intermediary?

The same questions apply beyond journalism. Businesses that publish research, advice or expert commentary may also find their material summarised elsewhere. Measuring only rankings or AI citations gives an incomplete picture. It is also worth looking at qualified visits, returning readers, subscriptions and enquiries, depending on what the content is intended to achieve.

AI will continue to make information easier to retrieve and repackage. The harder work remains deciding what is reliable, finding what has not yet been reported and giving people enough reason to return to the source. Good journalists can use AI to make their work better, giving them more time to investigate, question and explain. AI can never replace the curiosity, judgment and responsibility that good journalism requires.