Taboola Presents DeeperDive Generative Chatbot Findings

Taboola product manager Kurt Gessler delivered a session titled "A deep dive Into DeeperDive: What we've learned experimenting with agentic experiences on traditional publisher websites" at Mumbrella Publish on 28 October 2025, per Mumbrella. The article states Gessler was roughly four months into building and launching DeeperDive, described as a generative-AI-powered chatbot for publishers that allows readers to ask open-ended questions and receive answers gleaned from the publication's own journalism. Mumbrella records a direct quote from Gessler: "We wanted to know, would this work on a traditional publisher site? We got in very early, got live, and we have a lot of things to share."

Companies deploying generative chat on publisher websites typically combine an index of the publisher's corpus with a retrieval layer plus a generative model to produce contextual answers. Industry-pattern observations suggest common engineering focal points include content ingestion and canonicalization, retrieval quality and prompt design, and provenance or attribution mechanisms to surface source articles alongside generated answers. Handling hallucination risk and ensuring up-to-date content are recurring technical challenges for publisher-facing conversational features.

Industry reporting frames experiments like DeeperDive as part of a wider wave of publisher-focused applications that aim to increase engagement by letting readers query a site's journalism directly. Observers note that early deployments often trade off answer depth for editorial control and require publisher-specific tuning to match tone, paywall rules, and licence constraints.

Indicators useful to practitioners and observers include metrics Gessler or Taboola publish for engagement and accuracy, whether the product surfaces inline citations or links to source articles, how the system respects paywalled content, and any post-launch findings Taboola publishes about moderation, hallucination rates, or retraining cadence. If Taboola or partners publish technical notes or case studies, those will clarify architecture and data-handling choices.