Taboola's DeeperDive Reports 15–25% CTRs via Conversational AI

Full analysis

Taboola's Adam Singolda is telling anyone who'll listen that DeeperDive, an ad-supported chatbot bolted onto publisher pages, is pulling 15 to 25% click-through rates and fielding about 333,000 questions a day. He calls LLM monetization "the next CTV." The question for operators: is this a real new revenue layer for the open web, or a founder's conference-stage number that mean-reverts by spring?

What's actually being decided. Not "should you use DeeperDive." It's whether publishers, DSPs, and data buyers should start treating on-site conversational search as a distinct inventory and data category worth building pipes for. Type 2 decision, easy to reverse: nobody's betting the company on a pilot. Forcing function is the 2026-2027 upfront conversations, where "curiosity data" either shows up in a negotiation or it doesn't.

The Market Analyst. Singolda is planting a flag on the on-site answer layer before Google's AI Overviews and Perplexity swallow it at the browser level. In plain terms: he wants Taboola to own the little "ask a follow-up" box on publisher pages before the big platforms make it free everywhere. That's a smart land-grab, because it repositions Taboola from a traffic middleman into an intent-data broker, which is a better business if it holds. But a CEO quoting his own unaudited CTRs at Beet.TV is marketing, not evidence. Watch the next earnings call: if DeeperDive is real, it becomes a named revenue line. If it stays a slide, it's a narrative to prop up the stock.

For the generalist: Taboola is trying to sell what readers are curious about, not just what they clicked.

The Skeptic. 333,000 questions a day sounds big until you set it against Taboola's network reach, where it's a rounding error. The 15 to 25% CTR almost certainly reflects the novelty of early, opted-in, hyper-engaged readers on a handful of friendly publishers. New ad formats always overperform for two quarters, then settle toward the mean. "Curiosity data" is session-depth signal with a fresh coat of paint. And the structural tension nobody's naming: Taboola's core business sends readers away for money, while DeeperDive keeps them home. Those two revenue models fight each other. Singolda's own framing that "AI won't kill agencies but programmatic might" is a man talking his book against the channel that made him.

For the generalist: a chatbot's first users are its fans, so early scores flatter it.

The Operator. Forget the CTR. The Tuesday-morning problem is plumbing. That "first-party data asset" is worthless until a publisher's question logs actually flow into their clean room or DMP in a shape a buyer can target against. Most mid-tier publishers do not have the engineering bandwidth to build that pipe, which means the curiosity data sits in Taboola's environment, not the publisher's. And that's the catch. If Taboola holds the data, Taboola captures the value, and the publisher is renting intent signal about its own readers. The second break point is attribution: when an LLM-assisted click converts, whose win is it, the publisher's or the network's? Nobody's answered that.

For the generalist: the data only pays off if it lands somewhere the publisher controls, and right now it doesn't.

The Customer / End User. Two customers here, and they want different things. The reader wants a fast answer without leaving the page, and DeeperDive genuinely delivers that. Fine. The advertiser is the one being sold to, and no media buyer has asked for "curiosity data" by name. They'll try it because 20% CTRs are irresistible in a deck, then they'll ask the boring questions: is this incremental, or is it stealing credit from clicks that would've happened anyway? Does the conversion survive an audit? Until a buyer runs a holdout test, "the number one conversion rate comes from LLM" is a sentence, not a media plan.

For the generalist: buyers love the number until they check whether it's real new business.

Where the council splits. The Strategist read (from the prior lens) and the Market Analyst see a genuine repositioning: Taboola escaping the arbitrage business and owning reader intent on the open web. The Skeptic sees a retention feature that Taboola is pitching as a data product, while it fights Taboola's own revenue model. The Operator sits in the middle and says it doesn't matter who's right until the data pipe exists, because whoever holds the logs holds the money. That's the real disagreement: is DeeperDive a new asset class, or a Taboola feature that mostly benefits Taboola?

What it hinges on. Three things. One, does the CTR survive contact with scale and an audited holdout, or does it mean-revert. Two, does the curiosity data land in the publisher's environment or stay in Taboola's. Three, does any buyer commit disclosed spend against it. The council leans skeptical on the headline number and interested in the intent-data thesis. The move for operators is cheap: run a pilot, but demand the question logs land in your clean room and insist on an incrementality test before you believe a single conversion claim.

Prediction: Through Taboola's next two quarterly earnings calls, DeeperDive will not be broken out as a separately disclosed revenue line, and the 15 to 25% CTR figure will not be repeated with audited, at-scale numbers.

Confidence: Medium. Early founder-stage metrics rarely survive scale or make it into audited financials.

Why: The 15 to 25% CTR and 333,000 daily questions come from Singolda pitching at a conference, not from a filing or a third-party study. That is the classic shape of a number that shrinks when it scales past opted-in early adopters. New ad formats reliably overperform on novelty for a quarter or two and then settle toward network norms, so a repeated, audited, at-scale figure would be the surprise, not the base case. And companies break out a new revenue line when it's big and durable enough to move the stock; if DeeperDive were already that, Singolda would be citing dollars, not click rates. The opposite outcome, a clean audited number and a named revenue line within two quarters, would require the pilot to hold up under exactly the pressure that usually deflates these figures.

Revisit by 2026-12-31: We're right if Taboola's next two earnings reports mention DeeperDive only qualitatively with no audited CTR at scale and no standalone revenue disclosure. We're wrong if Taboola discloses DeeperDive as its own revenue line or publishes an audited, at-scale CTR in that window.