For the past 5 years, Taboola has hosted an annual Publisher Product Council: a forum for our most valued publisher partners from around the globe to learn about our latest product offerings, share their insights and ideas, and explore pressing topics in our industry together.

The latest installment of the council explored the question of how the open web can leverage AI without compromising publisher control, trust, or monetization. The discussion focused on DeeperDive, Taboola’s AI-powered answer engine built for the open web. Leading the session was Kurt Gessler, a Staff Product Manager on the DeeperDive team, with over 2 decades of experience leading digital product and strategy at top publishers.

Over the hour-long session, Kurt explored what early learnings from DeeperDive reveal about behaviour, engagement, and revenue when generative AI is embedded directly into publisher environments.

Rising to the AI Challenge

The rise of generative AI is simultaneously seen as a potentially devastating threat and a rare opportunity for open web publishers. With AI investment expected to exceed $1.5 trillion in 2025, search traffic declining and new user behaviors emerging, the question is no longer whether AI will change how people discover and consume information, but who will be positioned to benefit from that shift.

Taboola is working to ensure that publishers play a central role in the AI-driven future of search and discovery, leveraging their trusted journalism and brand authority to meet users in the early stages of their decision-making journeys.

Deeper Dive: AI Answers, Powered by Trusted Journalism

DeeperDive is Taboola’s response to the growing user demand for instant, conversational answers to any question. Embedded directly on publisher sites, the experience allows users to ask questions (or choose from suggested ones) and receive AI-generated responses sourced exclusively from that publisher’s own content.

Rather than pulling users away to external platforms, DeeperDive is designed to keep the conversation (and the value) within the publisher ecosystem, surfacing relevant articles and encouraging deeper on-site journeys.

Deeper Dive has already been rolled out across multiple publisher sites, including six on which it appears on 100% of traffic, such as USA Today, Daily Star, and India Today.

Built for Publishers: Speed, Safety, and IP Control

Throughout the council, Kurt addressed some of the challenges the team encountered while designing and iterating on DeeperDive. The goal wasn’t just to ship an AI experience, but to build one that could operate at publisher scale, without compromising performance, brand safety, or content ownership.

Key learnings included:

• The importance of having a model-agnostic architecture, allowing DeeperDive to adapt quickly as LLM technology evolves.
• Significant UX and performance optimization, reducing answer load times from double digits to near-instant for common queries, and dramatically lowering bounce rates.
• Building in robust guardrails and human-in-the-loop QA, ensuring brand safety, accuracy, and relevance.
• Clear assurances around content ownership: publisher articles are not saved, reused, or used to train models, meaning that publishers retain full IP control at all times. 

Higher Engagement, Stronger Monetization

Six months after the launch of DeeperDive, results from live deployments on publisher sites point to meaningful impact on engagement and revenue:

• Users who engage with DeeperDive generate more page views per session, return more frequently, and show higher loyalty signals.
• Revenue per thousand (RPM) and average revenue per user (ARPU) are substantially higher, with DeeperDive users delivering nearly 3x higher revenue per user.
• Suggested questions have proven critical, driving click-through rates up to 6 times higher than open text fields alone.

Importantly, cannibalization of existing Taboola placements or other DSP positions has been minimal, reinforcing DeeperDive’s potential as an incremental revenue stream for digital publishers.

Turning Reader Questions Into Editorial Insight

One of the most compelling benefits of DeeperDive is the reader intent data that can be gathered from the tool. Across participating sites, tens of thousands of questions are asked daily, creating a real-time window into what audiences are curious about, confused by, or seeking clarity on.

The data so far shows that readers gravitate toward asking highly practical, time-sensitive questions. Common themes include:

• Weather and local conditions
• Consumer and personal finance topics
• Health and wellness
• Entertainment and celebrity news
• Sports and major live events

In a small but telling reminder of how new these AI interactions still are, the single most common free-text query so far has been a simple “hello”. This underscores the fact that readers are still learning how to engage with AI, making thoughtful prompts and suggested questions critical for guiding meaningful interactions.

Publishers are already using the intent data provided by user queries to:

• Identify content gaps beyond traditional page-view metrics
• Inform story development and updates
• Spot emerging topics earlier

In one example shared during the council, insights from clustered reader questions directly led to the creation of a new article that became a top-performing story for several days, demonstrating how reader questions can translate directly into editorial impact.

What’s Next For DeeperDive?

Looking ahead, Kurt touched on a number of areas of exploration for how DeeperDive could evolve over time. Taboola is having close conversations with leading publishers and product strategists on where AI-powered experiences on the open web may go next.

Among the possibilities shared were:

• Tools to give newsrooms greater visibility into reader questions and intent
• Extending the experience to audio and video content, including transcripts and voice-based interactions
• Integrating structured data sources such as weather, traffic, or market data to answer questions that require hard data answers.

As with DeeperDiver’s initial rollout, any future direction will be shaped by publisher feedback, real-world performance, and a continued focus on delivering trusted, high-quality experiences rooted in publisher journalism.

A Shared Vision For the Open Web

This Publisher Product Council reinforced the central idea that AI doesn’t have to pull value away from publishers. When built collaboratively and deployed responsibly, it can do the opposite. By embedding AI experiences directly into publisher environments, DeeperDive offers a glimpse into a future where journalists and technology work together to create better experiences for readers and sustainable growth for publishers.

This is still the early days, but the opportunity for publishers to shape the next era of the open web has never been greater.