AI in the Newsroom: Opportunity Must Come with Accountability
Artificial intelligence can help newsrooms work faster, reach wider audiences and strengthen investigative journalism. But editors were warned at the 4th Annual Editors’ Convention that without human oversight, clear policies and strong ethical safeguards, the same technology could undermine journalism’s credibility.
Artificial intelligence is rapidly finding its way into newsrooms, transcribing interviews, translating stories, testing headlines, analysing data and helping journalists distribute content across multiple platforms.
For news organisations operating with small teams and limited resources, these capabilities offer considerable promise. Tasks that previously demanded hours of manual work can now be completed in minutes.
But speed does not guarantee accuracy, and efficiency does not remove editorial responsibility.
That was the central message of an AI masterclass facilitated by Eng. Primera Muthoni, a Technology Business Systems Consultant specialising in user experience, product design, product management and automation, during the 4th Annual Editors’ Convention.
The session, “AI Models for Editors and Newsrooms,” introduced editors and senior journalists to the practical use of artificial intelligence while challenging newsrooms to establish safeguards before adopting the technology at scale.
“AI should not replace editorial judgement,” Muthoni emphasised during the masterclass.
Instead, she explained, the technology changes where journalists and editors devote their time. AI may handle some repetitive and time-consuming tasks, but journalists must spend more time verifying, editing and critically evaluating its output.
An opportunity for resource-constrained newsrooms
AI presents important possibilities for East African newsrooms, many of which are under pressure to produce more content with fewer journalists and limited financial resources.
The technology can support investigative research, data journalism, transcription, fact-checking, translation, audience distribution and the verification of images, videos and claims.
For radio stations and news organisations working with local languages, AI can help convert interviews and broadcasts into searchable text. It can also assist newsrooms to translate or adapt content for different language communities and distribute one story across radio, television, print, websites and social media.
This could allow smaller newsrooms to expand their reach without placing every new task on already overstretched journalists.
But Muthoni cautioned that these benefits depend on how the technology is introduced and managed. AI systems generate responses by identifying patterns in the data on which they were trained. They predict likely outputs; they do not understand truth in the way a human journalist evaluates evidence.
An answer may therefore sound convincing while containing inaccurate, biased or entirely fabricated information.
For editors, the challenge is not simply whether to adopt AI, but how to use it without surrendering the standards that distinguish professional journalism from unverified online content.
Treat every AI output as an “unverified tip”
One of the strongest recommendations from the masterclass was that newsrooms should treat AI-generated information as an “unverified tip.”
A tip may point a journalist towards a useful lead, but it cannot be published without independent verification. The same principle should apply to material generated by an AI tool.
If a system produces a name, date, quotation, statistic or historical claim, the journalist must check it against an authoritative source. AI should not be cited as proof of a claim, nor should its confident tone be mistaken for accuracy.
The responsibility remains with the newsroom.
Editors were encouraged to insist on human editorial approval before any AI-assisted material is published. Newsrooms were also cautioned against using AI to fabricate quotations or manipulate photographs and videos in ways that mislead audiences.
The technology may assist the production process, but it cannot carry professional or ethical accountability.
Better prompts, better—not necessarily truthful—answers
During the practical component of the masterclass, participants experimented with AI-assisted summarisation, headline testing, prompting and data extraction.
Muthoni introduced the PACE framework as a method for giving AI systems clearer instructions:
- Purpose: What should the tool accomplish?
- Audience: Who is the intended reader, viewer or listener?
- Constraints: What limits, standards or requirements must it observe?
- Example: What does a suitable output look like?
The framework can help journalists obtain more relevant and usable responses. For example, a reporter asking AI to summarise a document should identify the intended audience, specify the required length, instruct the system not to introduce information outside the document and provide an example of the preferred style.
However, even a well-designed prompt does not guarantee a factually correct response. Participants were therefore reminded to cross-check AI-generated material against the original documents and other reliable sources.
Prompting is a useful technical skill, but verification remains a journalistic obligation.
The danger of bias and lost context
AI systems reflect the material used to train them. If that data contains historical prejudices, geographical imbalances or limited information about African societies, the outputs may reproduce those weaknesses.
For Ugandan and East African newsrooms, local context is particularly important. An AI system may generate a technically coherent response that misunderstands local institutions, languages, political realities or cultural practices.
Editors must therefore ask whose knowledge is represented, whose perspective is absent and whether the output unfairly stereotypes particular individuals or communities.
Translation also requires human review. A tool may translate words correctly while losing the cultural meaning, tone or context of what a source intended to communicate.
Muthoni urged journalists to pay particular attention to “accuracy, attribution, bias and disclosure” when using AI. These principles should guide decisions from the initial gathering of information to publication.
Protecting sources and confidential information
The convenience of AI also creates serious privacy and source-protection risks.
Journalists may be tempted to paste interview transcripts, confidential documents, unpublished investigations or off-the-record information into publicly available AI systems. Yet once sensitive material is entered into an external platform, the newsroom may lose control over how it is stored, processed or used.
The masterclass warned journalists against entering sensitive or off-the-record information into systems that have not been approved by their organisations.
For investigative journalists in particular, poor AI practices could expose sources, compromise ongoing investigations and place individuals at risk.
Newsrooms must therefore decide which tools may be used, what information can be entered into them and which categories of material must remain entirely outside public AI platforms.
Where resources permit, media organisations could explore private or self-hosted models for sensitive work and develop in-house systems connected securely to their archives.
Deepfakes demand stronger verification
AI is not only changing how journalists produce content. It is also making it easier to generate or manipulate photographs, audio and video.
During elections, conflicts and other fast-moving events, fabricated content can spread before newsrooms have time to establish its origin. A convincing deepfake attributed to a public official could influence public opinion, provoke hostility or damage reputations.
The masterclass introduced participants to several approaches for verifying suspicious digital material, including metadata checks, reverse-image searches, digital forensic tools, provenance and watermarking standards.
However, no single detection tool provides complete certainty.
Effective verification requires a combination of technological tools, knowledge of the subject, examination of the original source and human expertise. Newsrooms may also need dedicated verification teams or clearly assigned personnel responsible for investigating questionable digital content.
Adopt the guardrails before scaling the technology
Muthoni advocated a “guardrails-first” approach to newsroom AI adoption.
Under this approach, a media organisation establishes its editorial boundaries before integrating AI widely into its operations. The safeguards should include human approval before publication, independent factchecking, protection of sensitive information, respect for privacy and consent, and clear disclosure of AI-generated or substantially altered public-facing content.
Newsrooms must also determine when audiences deserve to know that AI was involved in producing a story, image, video or audio product.
Disclosure should not become a substitute for accuracy. Labelling misleading material as AI-generated does not make it ethically acceptable. Instead, transparency should complement the newsroom’s existing obligation to publish accurate and fair information.
A 90-day roadmap for newsrooms
Rather than adopting numerous tools at once, the masterclass proposed a gradual 90-day roadmap.
Newsrooms should begin by identifying two or three difficult or time-consuming tasks that AI could help address. These might include transcribing interviews, searching archives, translating routine content or extracting information from large documents.
The organisation can then run a controlled pilot with clear measures of success and specific editorial safeguards. Editors should assess whether the tool saves time, maintains accuracy, protects confidential information and genuinely improves the quality of journalism.
Only successful applications should be expanded.
The final stage should involve formalising a newsroom AI policy that defines approved tools, prohibited uses, disclosure requirements, verification responsibilities and the person ultimately accountable for publication.
This gradual approach allows newsrooms to learn from experimentation without exposing the entire organisation to unnecessary risk.
Accountability cannot be automated
Artificial intelligence will continue to influence how journalism is gathered, produced and distributed. Ignoring it may leave newsrooms unable to respond to changing audience habits and new forms of digital competition.
Uncritical adoption, however, could be equally damaging.
AI can generate a summary, but it cannot take responsibility for omitted context. It can suggest a headline, but it cannot answer for the harm caused by an inaccurate claim. It can produce an image, but it cannot defend the newsroom when audiences are misled.
Those responsibilities remain with journalists and editors.
The opportunity presented by AI is therefore not to remove human beings from journalism. It is to free journalists from some routine tasks so that they can devote more attention to investigation, verification, context and public-interest reporting.
The message from the masterclass was clear: newsrooms should experiment, innovate and build new skills—but always with guardrails in place.
AI may assist journalism, but editorial judgement, ethical responsibility and public accountability must remain human.
