Frequently Asked
Questions

Yes, from the very first click. Froomle's session-based models capture real-time intent signals immediately, so even brand-new anonymous visitors get personalized recommendations the moment they start engaging. There's no waiting period for a model to "warm up."

Froomle uses a proprietary ensemble of multiple AI techniques specifically developed for news: collaborative filtering (user-based and item-based), content-based models powered by LLMs (Google Gemini, OpenAI), session-based deep recurrent neural networks, reinforcement learning (contextual bandits), popularity and trend detection, and diversification models. Rather than a single model, we combine what works best for your specific audience and KPIs.

Yes. We believe the best results come from combining AI with human expertise. You can apply business rules, content filters, boosts, and blacklists — for example limiting recommendations to articles under 24 hours old, prioritizing certain categories, or excluding specific content types. Changes can be tested and previewed in the Froomle Recommendation Lab, a live interactive sandbox on your own data, before deploying to production.

This is something we take seriously and have published academic research on. Froomle actively monitors recommendation diversity and incorporates serendipity and content discovery as explicit optimization objectives — not just short-term clicks. Our models are designed to expose readers to a healthy mix of content, support editorial goals, and avoid the feedback loops that degrade long-term engagement.