Inside FAZ’s AI Playbook

Germany’s Frankfurter Allgemeine Zeitung (FAZ) is sharpening its digital ambitions. Having already surpassed its 2025 target of 300,000 digital subscribers, the publisher is now aiming for 400,000 by the end of the decade.

To reach that goal, FAZ is turning to artificial intelligence. At the 2025 Digital Growth Summit, FAZ’s Deputy Chief Product Officer Marina Sorg detailed how they are experimenting with AI to accelerate their subscription strategy and what those lessons might mean for the broader news industry.

Proven Successes With AI Features

Central to FAZ’s AI strategy is the goal of enhancing its core value proposition: the quality of the reading experience. By integrating AI features directly within articles, the publisher seeks to boost reader retention, curb churn, and reinforce its subscription business.

This section dissects two of FAZ’s most prominent AI implementations (article summaries and text-to-speech) and evaluates the tangible results they have delivered in subscriber engagement and conversion.

AI Summaries

For the past two years, FAZ has offered an AI Summary feature in the Rhein-Main section of its edition. Its primary success lies in its impact on existing subscribers. According to Sorg, the data shows that subscribers who use the AI Summary function are “almost twice as engaged as others.” This heightened engagement is a critical metric for FAZ, as it directly correlates with lower churn. The feature has also proven its power as a conversion driver. A single promotional campaign for the summaries generated “200 subscribers in one day” from users subscribing specifically to access the feature, admittedly with the added benefit of a free trial (see image).

However, the feature faces challenges that limit its broader impact. Its availability on only 30% of articles hinders user habituation, as readers don’t encounter it consistently enough to make it a regular part of their routine. Additionally, direct taps on the AI Summary icon by non-subscribers, which then triggers a “feature paywall” screen, show “low performance for gaining new subscriptions.”

Furthermore, FAZ acknowledges a classic “hen/egg problem”: It remains unclear whether the feature actively creates engagement or if the platform’s most engaged users are simply more likely to seek it out. This ambiguity stands in contrast to the text-to-speech feature, whose universal availability provides much stronger evidence that the feature itself can create and sustain user habits, rather than merely attracting already-engaged users.

Text-to-Speech (TTS)

In contrast to the limited rollout of summaries, FAZ’s text-to-speech (TTS) feature has become a standout success, largely due to its universal availability. First introduced in 2018, the subscriber-only feature is now powered by modern generative AI, offering a far superior user experience that, while it “still sounds a bit computer voice-like,” is “getting better and better.” This investment has paid significant dividends, establishing audio as a core component of the FAZ subscription.

The usage data paints a compelling picture of its value:

• Broad adoption: Over a six-month period, a remarkable “34% of our subscribers use text-to-speech.”
• Deep engagement: Subscribers who use TTS are highly active, listening to an average of 14.6 articles a month.
• Consistent growth: User adoption has been “steadily increasing” since the beginning of 2023, with more unique users adopting the feature and existing users listening more frequently.

The key lesson from the success of TTS is the power of consistency. By making the feature available on every article, FAZ has allowed subscribers to form a durable habit, integrating audio into their daily news consumption in a way that remains aspirational for the more sporadically available summaries.

The Experimental AI Lab

Not every innovation initiative results in clean user data and immediate wins. FAZ’s experience with its Article Chatbot serves as a case study in how experiments that “fail” on paper can provide invaluable internal learnings.

For context, the Article Chatbot was launched this past sprint and is populated with articles from the Rhein-Main section. More specifically, it is available exclusively on paywalled articles within this smaller regional section. When asked what user data the experiment had yielded, Sorg’s candid assessment was that they had found “nothing so far.”

The reasons were straightforward: the test was confined to a small user base, limited to a subset of articles, and required proactive user interaction. More importantly, the type of content proved to be a poor fit. Sorg explained that a typical local news article, such as one detailing a delay in metro construction, is too specific and self-contained to invite further questions. In contrast, a broader article on a topic like ETFs would naturally generate more user interaction. The combination of limited scale and unsuitable content resulted in insufficient data to draw meaningful conclusions about user behavior.

However, the project was far from a failure, as it taught the product team crucial internal lessons:

1. The criticality of editorial collaboration: The chatbot had to be manually inserted into articles by the editorial team. This dependency highlighted the necessity of bringing editorial stakeholders into the process at the very beginning to ensure buy-in and a smooth workflow.
2. The importance of scale and context: The team concluded that the Rhein-Main section was ultimately “too small to test this kind of feature” and that the content type was not conducive to chatbot interaction. This provided a vital guideline for scoping future pilot projects.

The New Frontier

Building on its learnings from article enhancements, FAZ is now taking its next major step in AI: moving from augmenting existing content to fully automating the creation of new formats.

The new podcast, titled the “Rhein-Main Feierabendbriefing,” will be “totally generated by AI,” with a human editor in the loop for quality assurance. The end-to-end workflow is designed for maximum efficiency:

1. At a set time each day, a system crawls the first four articles from the Rhein-Main section.
2. This content is automatically transformed into a podcast transcript.
3. An editor from the editorial team reviews the transcript for accuracy and quality.
4. The final script is fed into the ElevenLabs platform, which generates the audio using a cloned voice of the host of FAZ’s daily morning podcast.

The final product will be a daily briefing, a maximum of five minutes long, published every weekday. This initiative marks a significant strategic shift, leveraging AI not just to improve existing products but to create entirely new ones with minimal human effort.

Lessons from FAZ’s Pragmatic Innovation Approach

Three lessons stand out for media professionals:

• Tie AI to core metrics. Features such as text-to-speech and article summaries prove their worth because they demonstrably improve subscriber engagement — a direct lever for reducing churn.
• Embrace internal learnings. The Article Chatbot project shows that not every experiment needs to yield user data to be valuable; insights about internal workflows and limitations can be just as instructive.
• Start small with willing partners. The Rhein-Main model highlights the benefit of finding internal allies who are open to testing, refining, and validating new ideas before wider adoption.