12 lessons from news outlets on the cutting edge of AI

Today's (11 November 2025) JournalismAI Festival provided a plethora of new and interesting ways news organisations all over the globe are grappling with artificial intelligence.

We summed up the key ideas, breakthroughs and lessons from Day 1.

Financial Times (UK)
Strategy: Embedded AI into editorial workflows for summaries, translations, and reader engagement, while carefully managing build vs. buy decisions and audience trust. Used AI to generate discussion prompts and experimented with disclaimers for transparency.
Outcome: AI-generated prompts increased commenting by 3.5 per cent - a vital improvement as 60 readers who comment are six times more engaged, but only 60 per cent of readers know they have an option to comment. However, some subscribers feel AI disclaimers “cheapen” journalism, leading to cancellations. FT now uses AI only when necessary and is transparent about its use.
Try this: Use AI to boost reader engagement, but be transparent and sensitive to audience perceptions—monitor for any negative impact on trust.

Read more:

How FT uses polls to engage readers and grow subscriptions

The newspaper has a loyal, well-informed readership keen to share their opinions. Polls in its newsletters record engagement over 75 per cent

Journalism UKGordon Smith

Guardian (UK)
Strategy: Rolled out AI tools - notably newsroom-wide adoption of Google Gemini - focusing on practical integrations and supporting cultural change with training and explicit permission to experiment.
Outcome: High rates of AI adoption for research and workflow; internal survey showed 50 per cent of staff felt the pace of change was right, with the rest split between too fast and too slow.
Try this: Make AI tools easy to access and encourage experimentation, but check in regularly with staff to address concerns and adjust the pace of change.

BBC (UK)
Strategy: Set up a dedicated AI team to pilot translation, content repurposing, and workflow automation projects, always keeping editors involved in reviewing outputs. Rolled out pilots slowly and deliberately, focusing on productivity and audience experience.
Outcome: Successful pilots, like a LLM-powered CMS for BBC Local Democracy Reporters improved efficiency and writing style. Slow, deliberate scaling reassured staff and maintained editorial standards.
Try this: Start with small, well-defined AI pilots, keep humans in the loop, and scale up only when your team is comfortable and quality is assured.

Scroll Media (India)
Strategy: Used AI to personalise news delivery formats (calculators, mind maps, decision trees) for India’s 700 million smartphone users. Ran large-scale headline optimisation (500,000 headlines) and developed a text-to-video tool for multi-language audiences.
Outcome: Headline automation revealed useful patterns but was rejected by editors for being too rigid. AI-generated video avatars were unpopular with Instagram audiences, who preferred authentic human presenters.
Watch out: Over-optimising with AI can stifle creativity and alienate both staff and audiences. Test new formats, but keep editorial flexibility and user feedback central.

Bayerischer Rundfunk (Germany)
Strategy: Used vectorised data from nine broadcasters to create a shared, trusted content pool for semantic search and chatbots. Collaborated with private and public media to build a chatbot for Oktoberfest, combining news, cultural, and service information.
Outcome: The chatbot answered a wider range of questions than any single newsroom could, but still required human oversight to prevent errors and hallucinations.
Try this: Pool content and data with partner organisations to build richer AI products, but always have humans review outputs for accuracy.

Daily Maverick (South Africa)
Strategy: Used AI-driven analytics to study user behaviour on their membership landing page, discovering that 75 per cent of visitors didn’t scroll below the fold. They redesigned the page for more targeted, dynamic messaging and built a community platform and an impact tracker to boost engagement and retention.
Outcome: Landing page conversion rate doubled. The new community forum attracted over 1,000 members, and the impact tracker helped demonstrate the newsroom’s influence on public debate and policy.
Try this: Use heatmaps or AI analytics to identify where users drop off on key pages, then test targeted changes to increase conversions and engagement.

European Correspondent (Switzerland)
Strategy: Developed an AI “companion editor” that integrates with their CMS to provide real-time feedback on story structure, tone, and style, prioritising major issues. The tool also checks for adherence to internal style and audience needs.
Outcome: 96 per cent of participating newsrooms built a working prototype; editing became faster and more consistent, and journalists received more actionable feedback, improving quality and workflow.
Try this: Pilot an AI editing assistant that flags structural or style issues first, helping writers focus on the most important improvements.

Ringier Media (Switzerland/International)
Strategy: Focused on cleaning and standardising data infrastructure to enable AI-driven, audience-centric content strategies across multiple markets. Emphasised the need for flexible teams and a culture of continuous learning.
Outcome: Found that sustainable change came from investing in team flexibility and audience focus, not just new technology. AI was most effective when paired with clear editorial purpose and clean data.
Watch out: Don’t just invest in new AI tools—prioritise data quality and team adaptability to get real value from AI.

Newsquest (UK)
Strategy: Built in-house AI tools to automate the drafting of routine stories (e.g., local events, press releases), freeing up journalists for original reporting. Created “AI-assisted reporter” roles and encouraged staff to develop their own AI tools for tasks like FOI requests.
Outcome: AI-assisted reporters now produce 600–800 stories per month, allowing other journalists to focus on fieldwork and investigations. Subscription numbers and page views increased, and the newsroom shifted onboarding to prioritise original reporting.
Try this: Use AI to handle repetitive content, and support staff in building their own workflow tools—track the impact on both output and staff roles.

Read more:

How Reuters, Newsquest and BBC experiment with generative AI

From letting Chat GPT craft FOI requests to translating articles for international readers, the tech has a lot of potential for newsrooms - but also many limitations

Journalism UKJacob Granger

DPA (Germany)
Strategy: Combined in-house AI development (e.g., headline and teaser generation) with external partnerships (e.g., you.com for archive search) to rapidly launch new services. Built a private, AI-powered archive search tool for clients, updated every three minutes.
Outcome: The new archive tool reduced research time from hours to minutes and was well received by both internal staff and clients. The partnership model allowed for faster, more flexible product development.
Try this: Mix internal and external AI solutions to speed up innovation. Focus on tools that save time for both staff and clients.

Gubbi Labs (India)
Strategy: Used AI to automate the conversion of research papers into news stories and developed a “newsworthiness index” to filter thousands of scientific publications each month.
Outcome: Reduced story production time from 6–7 days to 3–4 hours. Editors with science backgrounds still reviewed all outputs for accuracy and clarity.
Try this: Use AI to triage and summarise complex source material, but always have subject-matter experts review the final content.

El Surti (Paraguay)
Strategy: Partnered with local communities to record and validate Guarani-language data for AI training, using a simple web interface for voice submissions and peer validation.
Outcome: Validation rates for Guarani-language data increased from 11% to 16%. The project built stronger community ties and set the stage for future monetisation through print subscriptions and events.
Try this: Involve your audience in data collection and validation for underrepresented languages or topics—community engagement can improve both data quality and loyalty.

All talks were attended by a human. This article was drafted by an AI assistant before it was edited by a human.