How Are Newsrooms and Media Houses Using AI?

TL;DR:

• Most newsroom AI work happens behind the scenes: transcription, tagging, translation drafts, copy checks, summaries and coding. Audiences are least comfortable with stories written entirely by AI.
• Big newsrooms (AP, the BBC, the Guardian) and Kenyan ones (Nation Media Group, TUKO) follow the same rule: AI assists, and a human editor checks and signs off.
• In Kenya, the Code of Conduct for Media Practice 2025 is in force. Section 27 asks for disclosure of AI use in editorial material, human review before publication, and consent before cloning anyone's voice or likeness.
• The big risks are invented facts, wrong AI summaries, deepfakes and leftover prompts in print. Each has happened to a real newsroom.
• Start with a short AI policy, a list of approved tasks, a habit of checking every output, and training for every desk.

Who this guide is for

This guide is for editors, reporters, sub-editors, producers, digital desks and media managers in Kenya, and for journalism students and lecturers. It covers what newsrooms are actually doing with AI, based on published reports and policies. For training, see our page on AI training for media and PR teams in Kenya.

The short answer: most AI work is behind the scenes

Most newsrooms use AI for the repetitive work around journalism. Reporting, sourcing and judgement are still done by people.

The Reuters Institute surveyed 280 senior news leaders in 51 countries and territories in late 2025. Its Journalism, Media and Technology Trends and Predictions 2026 report (January 2026) found:

• The vast majority (97%) of publisher respondents rated back-end automation "important" for 2026. That means tasks such as transcription, copy-editing help and automated metadata (tags and labels).
• Newsgathering uses (82%) were the second most important area, and faster coding and product development (81%) was also gaining ground.
• When leaders judged their current AI work, 13% called it "transformational", 44% "promising" and 42% "limited".

Working journalists say the same. A Thomson Reuters Foundation survey of over 200 journalists in more than 70 countries across the Global South, reported in March 2025, found that 81.7% use AI tools in their work and 49.4% use them daily. The main uses were drafting and editing, transcription, fact-checking and research.

It also found two warning signs: only 13% said their newsroom had a formal AI policy, and nearly 58% of AI users were self-taught.

What newsrooms around the world are doing

These examples come from published policies and reports.

The pattern is simple. Big brands such as AP and the BBC guard their accuracy, so AI stays in support roles. Smaller newsrooms experiment more: the Reuters Institute notes that CITE's AI presenters also leave individual journalists less exposed, which matters where reporters face threats. And every policy keeps a human responsible.

What Kenyan newsrooms are doing

Since mid-2025, Kenyan media houses have moved from talk to written rules.

Nation Media Group's AI policy

In March 2026, Nation Media Group (NMG) announced a policy to guide AI use in its journalism. According to NMG's own report, the policy:

• Sets 10 core principles, including accountability, fairness, data protection, human-in-the-loop oversight and transparency.
• Allows AI for limited content generation, translation, content curation, audience analysis, routine tasks and targeted advertising.
• Promises to tell audiences when AI is used to generate content, collect their data or personalise their experience.
• Keeps human editorial oversight at every stage, and commits to staff training in AI literacy, fact-checking and ethics.

In the policy's foreword, NMG's Editor-in-Chief, Joe Ageyo, wrote that the question is no longer whether to engage with AI, but how to do so responsibly. The policy works alongside NMG's existing Editorial Policy and does not override it.

TUKO's "checks, but never writes" assistant

TUKO's editor-in-chief, Jacob Otieno Onyango, described his newsroom's approach on INMA's Content Strategies blog, as reported by GXpress in August 2026. TUKO uses an in-house assistant called EditorialEye, built on the newsroom's own editorial guides, SEO standards and ethics rules. In his words, it "checks, flags, and recommends, but it never writes". A human editor makes every final decision, and copy editors audit how it is used.

This is a model worth copying. AI as a second pair of eyes is far safer than AI as a writer.

The Media Council of Kenya's code

The Code of Conduct for Media Practice 2025 was gazetted on 14 May 2025 and came into force on 29 July 2025, according to news reports of its launch. Section 27, in Part VI of the code, sets out obligations on the use of AI and other technologies. In the text published by the Media Council, journalists, media practitioners and media enterprises must:

• Make sure AI used in content creation is fair and unbiased, does not spread harmful stereotypes, does not harm accuracy, and respects intellectual property.
• Tell the audience whenever AI has been used to create, modify or edit images, videos or other editorial material.
• Make sure AI does not generate or spread misinformation or disinformation.
• Have AI-generated or technology-assisted content reviewed by human editors and approved before publication.
• Not use AI to clone the likeness, voice or persona of real people without their informed consent, unless it is justified in the public interest and complies with the law.

At a February 2026 training in Kajiado, a Media Council manager told journalists that publishing AI-generated content without disclosure goes against the code. If a specific clause matters to a decision, read the code itself or ask the Media Council.

What Kenyan audiences say

Kenyan audiences are already meeting AI, and many cannot spot it. The Media Council's State of the Media 2025 report, based on a survey of 3,774 people across all 47 counties in April 2026 and released on 4 May 2026, found:

• 59% of respondents were aware that AI is used in the media.
• 63% could not identify AI-generated content.
• 27% used AI platforms for news regularly and 34% occasionally.

The global picture fits. The Reuters Institute's Digital News Report 2026 found that concern about fake news is highest in Nigeria and Kenya, and that more than half of Kenyan respondents use TikTok for news. Across all markets, only 20% trusted answers from AI chatbots, against 37% who trusted news overall.

A July 2026 study in Media and Communication, based on 94 interviews in Kenya with audiences, journalists, educators and others, found what its authors call a "double-edged sword". Most of the people interviewed supported disclosing AI use, but many journalists fear disclosure makes their work look less authentic.

What audiences are comfortable with

Audiences accept AI behind the scenes and push back when it replaces people.

The Reuters Institute's Generative AI and News Report 2025 surveyed people in six countries (Argentina, Denmark, France, Japan, the UK and the US) in mid-2025. Kenya was not included, but the pattern is clear:

By task, comfort was highest for spelling and grammar editing (55%) and translation (53%). It fell to 30% for rewriting content for different audiences, 26% for creating a realistic image where no photo exists, and 19% for artificial presenters or authors. Only 33% believed journalists "always" or "often" check AI outputs before publishing.

The lesson: start where audiences are comfortable, and earn trust before you go further.

What can go wrong: five real cases

Every risk below has already hit a real newsroom.

1. Invented facts in print

In May 2025, the Chicago Sun-Times printed a syndicated summer supplement whose reading list recommended books that do not exist. The Philadelphia Inquirer used the same list. The distributor, King Features, said a freelancer had used AI without disclosing it. The Sun-Times said the section ran without review by its editorial team, and removed it from its website.

The lesson: syndicated and supplied content needs the same checks as your own copy. AI makes up titles, quotes and references that look completely real.

2. AI summaries that change the story

In December 2024, Apple Intelligence sent a notification under the BBC's name falsely saying that Luigi Mangione, a US murder suspect, had shot himself. Other false summaries hit the BBC and the New York Times. In January 2025, Apple said it would pause AI summaries for news and entertainment apps.

The lesson: a summary is a new piece of content. If AI summarises your story, someone must check that the summary says what the story says.

3. AI assistants getting the news wrong

In October 2025, a study coordinated by the European Broadcasting Union and led by the BBC had journalists from 22 public service media organisations in 18 countries check over 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity. 45% had at least one significant issue, 31% had serious sourcing problems and 20% had major accuracy problems, such as invented details.

The lesson: never use a chatbot answer as a source. Go to the original document, person or record.

4. Deepfakes in Kenya

AI-made videos and images can look like real news, and they are hard to spot: in the Media Council's State of the Media 2025 survey, 63% of people said they could not identify AI-generated content.

In February 2026, the head of the Media Council of Kenya said technology-facilitated disinformation is the biggest challenge the Council sees.

The lesson: a fake can now carry a real logo and a familiar face. Newsrooms need a verification habit, not just a fact-checking desk.

5. The prompt that went to print

In the same INMA piece, TUKO's Jacob Otieno Onyango wrote that in November 2025 a mainstream Kenyan newspaper printed an article that still contained an AI prompt. He called it "a glaring lapse in editorial oversight".

The lesson: AI does not remove the sub-editor. It makes the sub-editor more important.

What Kenyan editors can copy, starting this month

You need clear rules and good habits more than a big budget.

1. Start behind the scenes. Transcription, first-draft translation, tagging, headline options, summarising public documents and style checks. Audiences are most comfortable here, and errors are easiest to catch.
2. Use AI as a checker before you use it as a writer. Follow TUKO's model. Ask AI to flag problems in copy a human wrote, not to write the copy.
3. Treat every AI output as an unverified tip. That is AP's rule, and it works. Check every name, number, quote and date against a primary source.
4. Keep sources and unpublished work out of public tools. Never paste confidential sources, leaked documents, unpublished investigations or personal data into a public chatbot. Our guide to the Data Protection Act and AI explains the legal side. For sensitive documents, some teams run an AI model on their own computer instead; see how to run open-source AI models locally.
5. Label what audiences would want to know. Disclose AI-made or AI-edited images and video, synthetic voices, and stories drafted mostly by AI. When in doubt, disclose.
6. Train everyone. Most journalists are self-taught, so habits vary by desk. Shared training gives everyone the same rules.

Five prompts a newsroom can copy

At FuKazee we teach the four Cs: Clarity, Context, Constraints and Customisation. Use these with public, non-confidential material only, and check every output.

1. Clean up an interview transcript

I am a reporter. Below is an automatic transcript of an on-record interview. Clean it up: fix obvious transcription errors, add speaker labels (Reporter, Guest), and remove filler words. Do not change the meaning of any sentence, and do not add anything that is not in the transcript. Mark any word you are unsure of with [CHECK]. Transcript: [paste]

2. Check copy before it goes to the sub-desk

Below is a news story I wrote. Do not rewrite it. List every factual claim (names, titles, numbers, dates, quotes). For each, say whether the story gives a source. Flag any claim with no attribution, any quote without a named speaker, and any sentence that states opinion as fact. Story: [paste]

3. Headline and social options

Here is a published story. Suggest five headlines under 70 characters and three social media posts under 200 characters. Rules: no claim that is not in the story, no clickbait, no words like "shocking", and keep names and numbers exactly as written. Audience: Kenyan readers on mobile. Story: [paste]

4. First draft of a Kiswahili version

Translate this short public notice into natural Kiswahili for a Kenyan radio audience. Keep names, places and numbers exactly as written. Where an English term has no common Kiswahili equivalent, keep the English term in brackets. Notice: [paste]

A fluent Kiswahili editor must read the result before it airs.

5. Plan the verification of a viral clip

A video is circulating that appears to show [public figure] saying [claim]. I have not verified it. Give me a step-by-step verification plan: what to check in the video itself, which official accounts and archives to search, who to contact for comment, and what would count as confirmation. Do not tell me whether the video is real.

That last line matters: AI cannot tell you whether a clip is real. It can only help you plan the checks.

A checklist for your newsroom AI policy

A newsroom AI policy can fit on two pages. For a general template, see our AI acceptable use policy guide. These are the questions specific to journalism.

• Scope: Does it cover all staff, freelancers and syndicated content?
• Approved tools: Which AI tools may staff use, and on which accounts?
• Approved tasks: Which tasks are allowed (for example transcription, translation drafts, headline options, copy checks)?
• Banned uses: Which are never allowed (for example inventing quotes, generating realistic images of real events, cloning a voice without consent)?
• Never paste: What must never go into a public tool (sources' identities, leaked documents, unpublished investigations, personal data)?
• Verification: Who checks AI output, and against what?
• Disclosure: When will you tell the audience, and how? Does this match the Code of Conduct for Media Practice 2025?
• Sign-off: Which editor approves AI-assisted content before it is published?
• Images, video and audio: What are the rules for AI-made or AI-edited visuals and voices?
• Corrections: How will you correct an AI-related error, and will you say AI was involved?
• Training and review: Who is trained, how often, and when is the policy reviewed?

Questions people ask

Are Kenyan newsrooms allowed to use AI?

Yes. No Kenyan law bans it. Section 27 of the Code of Conduct for Media Practice 2025 sets the rules: disclose AI use in editorial material, have a human editor review and approve AI content before publication, and get consent before cloning anyone's voice or likeness (unless it is justified in the public interest). The Data Protection Act applies whenever personal data is involved.

Do journalists have to tell readers when they used AI?

The Kenyan code says to tell the audience whenever AI has been used to create, modify or edit images, videos or other editorial material. Some editors, including TUKO's, argue that labelling minor help, such as polishing sentences, adds little for the audience, much like labelling a spell checker. The code's wording is broad, so if you are unsure, disclose.

Will AI replace journalists?

So far AI is changing tasks more than replacing reporters. Most news leaders surveyed by the Reuters Institute called their AI work "promising" or "limited", not transformational. Finding sources, being there and judging what is true are the parts AI does worst.

Can AI help with fact-checking?

It can list the claims in a story and help plan checks. It cannot confirm facts on its own: the 2025 EBU and BBC study found significant issues in 45% of AI assistant answers about the news.

How do you spot a deepfake?

Look for lips that do not match the audio, an odd voice, wrong on-screen graphics, and a clip missing from the outlet's official channels. Contact the person and the outlet directly. Africa Check regularly checks viral Kenyan clips, so search its site too.

What should a newsroom do first?

Write a short AI policy, choose two or three low-risk tasks such as transcription and headline options, and train every desk on the same rules. Review after two or three months.

Sources

All sources last checked 11 November 2026.

• Reuters Institute, Journalism, Media and Technology Trends and Predictions 2026 (January 2026)
• Reuters Institute, Digital News Report 2026, executive summary (June 2026)
• Reuters Institute, Generative AI and News Report 2025 (October 2025)
• IJNet, "How AI is changing journalism in the Global South" (Thomson Reuters Foundation survey, March 2025)
• Nieman Lab, "Not a replacement of journalists in any way: AP clarifies standards around generative AI" (16 August 2023)
• IBC, "BBC updates generative AI guidance" (21 January 2025)
• Press Gazette, "The Guardian's approach to AI" (6 July 2023)
• Nation, "NMG rolls out policy to guide AI use in journalism" (6 March 2026)
• GXpress, "Mishaps that prompted dialogue on AI use", on Jacob Otieno Onyango's INMA Content Strategies blog post (19 August 2026)
• Media Council of Kenya, The Code of Conduct for Media Practice, 2025, full text (section 27)
• The Star, "Media Code ushers in new era for journalism: MCK" (29 July 2025)
• Eastleigh Voice, "Kenya adopts new media code, ushering in stricter AI and child protection rules" (29 July 2025)
• Pulse Kenya, "Kenya media code 2025: key changes" (30 July 2025)
• The Star, "Media Council of Kenya trains journalists on ethical use of AI" (13 February 2026)
• Radio Generation, "MCK poll: Kenya media trust rises as misinformation and AI identification gap grow" (4 May 2026)
• Citizen Digital, "Social media emerges top source of news for Kenyans as radio drops: MCK" (4 May 2026)
• Radio Generation, "Media Council warns technology-driven disinformation threatens public trust" (13 February 2026)
• Zhang, Mudavadi and Zhou, "Transparency is a Double-Edged Sword: Stakeholders' Perception of GenAI Usage Disclosure in Kenyan Newsrooms", Media and Communication 14 (30 July 2026)
• NBC News, "Chicago Sun-Times admits summer book guide included fake AI-generated titles" (21 May 2025)
• Silicon UK, "Apple suspends AI-generated news notifications after errors" (January 2025)
• Infodocket, "BBC: largest study of its kind shows AI assistants misrepresent news content 45% of the time" (22 October 2025)

Where FuKazee fits in

FuKazee trains teams in practical AI for the work in front of them. Since 2022 we have trained more than 3,000 people. Our founder, Rodney Maina Wanjohi, holds a First Class Honours degree in Journalism and Mass Communication from the University of Nairobi.

For newsrooms, PR agencies and communications teams, we build sessions around real desk work: transcripts, headlines, translation drafts, copy checks and spotting fakes. We assume nobody knows AI yet, avoid jargon, show the free version of every tool, and never push one tool. Verification sits inside every exercise. For campaigns and press releases, see how marketing teams in Kenya use AI.

See how newsroom training is set up on our page for AI training for media and PR teams in Kenya.

To talk about AI training for your newsroom, email lead@fukazee.com