Publishers are losing the battle for reader attention to AI search engines, and they're fighting back by building their own AI-powered answer systems directly on their websites. HuffPost UK this week announced adoption of DeeperDive, an AI answer engine developed by Taboola that lives exclusively within publisher websites and draws only from each publication's own editorial archive. The move reflects a broader industry shift as news organizations attempt to reclaim the audience relationships that platforms like Perplexity, ChatGPT, and Google's AI Overviews have progressively weakened over the past two years .

Why Are Publishers Losing Traffic to AI Search Engines?

The numbers tell a stark story. Google Web Search traffic to news publishers collapsed from 51% in 2023 to just 27% in the fourth quarter of 2025, according to NewzDash data analyzed across more than 400 publishers globally . Meanwhile, the scrape-to-referral ratio reveals the core problem: while traditional search engines like Google operate at roughly a 10-to-1 ratio, crawling about 10 pages for every human visitor they send back to a publisher, AI companies operate at vastly different scales. OpenAI's scrape-to-referral ratio sits at 179-to-1, Perplexity's is 369-to-1, and Anthropic's is a staggering 8,692-to-1 . That means for every 8,692 times Anthropic's crawlers scrape a publisher's content, they send back exactly one human visitor.

The impact on click-through rates has been severe. Research from Ahrefs published in February 2026 found that AI Overviews now correlate with a 58% reduction in click-through rates for top-ranking pages, nearly double the 34.5% figure recorded in April 2025 . When Google's AI Overviews appear in search results, only 8% of users click any link compared to 15% without AI summaries, representing a 46.7% drop in engagement . Zero-click searches, where users get their answer without visiting any website, surged from 56% to 69% between 2024 and 2025 .

How Are Publishers Using On-Site AI to Reclaim Readers?

DeeperDive represents a fundamentally different approach to AI-powered content discovery. Unlike conventional AI search engines that aggregate information from thousands of sources and synthesize a single response, DeeperDive confines every interaction exclusively to the host publisher's archive. Readers can type questions into a conversational interface, and the system returns answers sourced only from that publication's own content. The interface also surfaces suggested follow-up questions and links to related articles from the same site, designed to pull readers deeper into the publication rather than toward an external platform .

The technology taps into behavioral data from Taboola's broader network, which reaches approximately 600 million daily active users across 9,000 publisher partners globally. This data informs which questions to prompt, which topics are trending at any given moment, and which answers are likely to satisfy a reader who has already shown interest in a particular subject. Taboola describes this as a "pulse of the internet" capability, a real-time signal that helps the system surface contextual, timely responses rather than relying on static training data .

Taboola

The user interface integrates directly into a publication's existing design. HuffPost UK's implementation shows the DeeperDive bar at the top of the homepage, with sample prompts like "What is the total UK defence spending increase since the Cold War?" alongside suggested follow-up questions about related topics .

Steps Publishers Are Taking to Deploy AI Answer Engines

• Embedding on-site AI interfaces: Publishers are integrating conversational AI directly into their websites rather than relying on external platforms, keeping readers on their own properties where they can control the user experience and monetization.
• Leveraging proprietary content archives: These systems draw exclusively from each publication's own editorial content, creating a competitive advantage by offering readers curated, authoritative answers from trusted sources rather than aggregated information from across the web.
• Capturing search-like advertising revenue: Publishers can insert contextually relevant, high-intent advertisements directly into AI-powered results pages, turning reader inquiries into commercial opportunities while keeping users on the site.

Cate Sevilla, Editor-in-Chief at HuffPost UK, explained the editorial rationale: "The tool allows the publication to use AI in a way that is helpful to our audience, without compromising on quality, so that we can continue to provide UK audiences with the accurate and compelling journalism for which HuffPost is known" .

Cate Sevilla, Editor-in-Chief at HuffPost UK

The monetization angle is equally important. DeeperDive creates a new revenue channel by inserting contextually relevant advertisements directly into the AI-powered results page. Publishers effectively capture search-like advertising revenue within their own digital environments, turning reader inquiries into commercial opportunities while keeping users on the site. Historically, that commercial value has accrued to search engines. Under the DeeperDive model, it accrues to the publisher instead .

The system also gives editorial teams visibility into the specific questions their readers are asking, a signal that has historically been accessible only through external SEO keyword tools. According to Taboola, this data can guide editorial decisions and homepage choices toward topics with demonstrable audience interest .

How Quickly Is This Technology Spreading?

DeeperDive adoption is accelerating rapidly. Taboola first announced the technology on June 11, 2025, with Gannett USA TODAY Network and The Independent as launch partners. Gannett operates approximately 220 local markets across 43 US states and reaches roughly 140 million monthly unique visitors digitally . The Bangkok Post adopted the system on November 18, 2025, becoming the first media organization in Southeast Asia to deploy DeeperDive. By April 2026, the product had reached nearly 7 million monthly active users .

In the UK specifically, Reach, the country's largest commercial news publisher, deployed DeeperDive across Express and Daily Star in early 2026, placing the company among the first major UK groups to deploy on-site AI answer technology. National World, another large UK regional publisher, signed a multi-year deal with Taboola in October 2024 covering AI recommendations across more than 60 news titles reaching 15 million monthly readers .

While DeeperDive's 7 million monthly active users remain modest compared to dominant generative AI platforms, the growth trajectory and publisher adoption suggest this model is gaining traction. For context, ChatGPT leads with an estimated 810 million monthly active users, followed by Gemini at 750 million, Grok at 60 million, Perplexity at 30 million, and Claude at 20 million . However, DeeperDive's embedded engagement model is structurally different from those standalone platforms, operating within publisher properties rather than as a destination service.

The broader context reveals why publishers are desperate for solutions. Organic traffic to US websites declined from 2.3 billion visits to under 1.7 billion between 2024 and 2025 . Traffic to the world's 500 most-visited publishers dropped 27% year-over-year, an average of 64 million fewer visits every month . For ad-supported outlets, this translates almost directly to advertising revenue decline. For subscription-driven publishers, fewer site visits mean fewer opportunities to convert casual readers into paying customers.

The damage extends beyond immediate revenue. Publishers have invested years building first-party data assets including audience login strategies, loyalty programs, newsletter portfolios, identity frameworks, and e-commerce data capture. These assets depend on human visitors actually arriving at the site. When AI intermediates the content consumption experience entirely, publishers lose not just the immediate ad impression but the ability to build the audience relationships that drive long-term monetization .

As one senior publishing executive at The Economist noted, organic traffic has declined sharply as generative AI tools increasingly intercept the path to publisher content. While consumer adoption of AI search has risen, there has been zero value exchange for content creators. The promised upside, traffic from large language models and AI-powered search, has not materialized .

HuffPost UK's adoption of DeeperDive signals that publishers are no longer waiting for AI companies to send traffic their way. Instead, they're building their own AI-powered discovery systems, reclaiming the direct relationship with readers that search engines once provided, and creating new monetization opportunities in the process.