Publishers must build AI features that solve user needs, not just deploy technology.
Nadine Hettegger, Team Lead, AI Products Accelerator, BurdaForward at WAN-IFRA Paris AI Forum, 15 September 2026.
At WAN-IFRA’s Paris AI Forum (Sept 15-16), Nadine Hettegger, Team Lead, AI Products Accelerator, BurdaForward, discussed where publishers should compete in the AI-driven market. BurdaForward is part of Burda Media, a German publishing house with multiple niche and general interest brands. The AI Products Accelerator team focuses on Focus and Bunte, which together reach 550 million monthly page impressions.
Users increasingly gravitate toward short-form content (TikTok, Instagram) and have less time for in-depth articles. They are frequently interrupted (e.g., phone calls, family). When they return to articles after interruptions, they have lost the thread of what they were reading. The loss of context makes sustained reading difficult.
LLMs already address these problems (summaries, Q&A), but publishers must internalise these capabilities or lose ground, argues Nadine: Addressing reduced attention, context switching, and lost context with AI tools keeps users on publisher platforms rather than driving them to ChatGPT.
Nadine made the case that the real competitive contest is "one layer down": The winning strategy is to compete on user engagement by deploying AI features that address changing behaviours and leverage LLM capabilities with proprietary content.
Four AI Feature Demos, All Live in Production
All current features developed by BurdaForward for Bunte and Focus are not just prototypes; they are already tested across the 550M monthly user base.
They fall into four buckets:
Keep Me Up to Date and Quick: AI-powered article summaries surfaced at the article level and category/topic level (e.g., on a celebrity's star profile page). KPIs measured include click-through rates into articles and session length; results show it is a new, effective way for users to discover what interests them.
Give Me More Depth: Star Connections Feature where users on Bunte can click on two celebrities and see AI-generated connections between them, drawn from a vectorised database of all Bunte articles. 80% of users who clicked a first celebrity also clicked a second; 40% repeated the interaction, showing strong engagement.
Listen to Me - Conversational Features: sports journalists watch live games in real time; AI generates live Q&A for users who cannot watch, surfacing the most important moments as they happen - Health chatbot: a chatbot built exclusively on proprietary content, with a prompt injection system that blocks off-topic or inappropriate queries, designed to build user trust.
Show Me the Personal Effect: a Pension Reform Calculator was built in ~3 days (front-end + design) and tested by journalists on day 4 before release. It drove newsletter sign-ups and strong user trust and is seen as a key engagement and potential subscription conversion tool.
Team Structure and Editorial Oversight
The AI product team is only 5 people, including ML engineers, product managers, and a full-stack developer. Small size is intentional: they pilot features on one brand (e.g., Focus) before rolling out to others.
Editorial Oversight: Editorial checks are built in from the start. Journalists review and annotate AI outputs (e.g., summaries) to encode their editorial judgment into a reusable "skill". Summaries are pushed to the CMS via an API but require journalist approval before going live. New features are initially released to a small percentage of users to validate before wider rollout.
What Worked, What Didn’t Work
Worked: Capturing journalists’ judgment by having them document how they evaluate draft summaries, and turning those criteria into prompts, significantly improved output quality.
Worked: Starting with one brand (Focus) as a pilot before scaling to other properties keeps the team small and agile.
Open question: Editorial quality control requires journalist checkpoints throughout, but the team ships fast by limiting initial rollout to small user percentages.
The Three Key Takeaways
Changing user behaviour drives strategy: Three critical user behaviour shifts shape product decisions — reduced attention spans from social media, constant context switching in daily life, and frequent loss of reading context.
Compete on user needs, not technology: Publishers should not compete with LLM providers on models; instead, they should combine LLM capabilities with proprietary content to meet user standards.
Test (learn) more, and ship fast: rapid experimentation with production tests determines which AI features create engagement. Publishers win by treating engagement as the foundation for all revenue models and deploying AI features that solve real user problems faster than ChatGPT can.
