A few weeks ago, I reported on the new frameworks we’re starting to see in publisher-AI relationships. Publishers are currently not seeing much revenue from AI answer engines’ use of their content, and industry insiders told me about three main compensation models they’re starting to see develop. The third of which, shared with me by Joe Marchese of Human Ventures, seemed by far the least plausible.
Basically, Marchese told me that publishers could work with AI firms to create branded answer engines—imagine ChatGPT powering a New York Magazine chatbot. Rather than scrape the entirety of the web to populate their answers, these answer engines would restrict their searches only to content from their specific host publisher.
This seemed wild to me because why would anyone willingly restrict the scope of their search to a website, when an internet-wide search would necessarily include that website, plus a trillion others?
FOR Scoops, Sharp Analysis, and Insider Context
By submitting your email, you agree to our Terms of Use and Privacy Policy . You may opt-out anytime by clicking ‘unsubscribe’ from the newsletter or from your account.
Subscribe
Loading…
Since then, I’ve had some pretty compelling conversations. First, I spoke with Adam Singolda, the founder and CEO of Taboola, one of two companies responsible for the “Doctors Hate Him” and “Here’s What The Kids From The Goonies Look Like Now” native ads you see on sites across the web. My British editor called these “chumboxes.”
The Taboola business is predicated on the health of the open web, which is not looking so healthy lately. As necessity is the mother of invention, Taboola recently unveiled a new product, called DeeperDive, which effectively acts as an answer engine that lives on publisher websites, ingests only their data, and produces answers that surface other relevant content from those websites. It is essentially a recirculation widget for the chatbot era. So far, the tool is live on Gannett’s USA Today website, but Taboola has plans to announce expansions soon, Singolda told me.
His argument: All the traffic lost to answer engines is lost to answer engines, and tools like DeeperDive are not remedies for that. Instead, what they seek to do is mimic a new behavior that users are becoming increasingly accustomed to—the chatbot interfaces—and enabling that behavior on publisher websites. This way, whatever percentage of direct traffic a publisher has, the gizmo deepens their engagement, encouraging them to spend far longer on the page and offering up a gold mine of first-party data in the process.
The next day, I had a call with the founder of a stealth startup working on effectively the same product. Their widget is live on two websites, but they have politely asked me not to say more until Oct. 1, so I am reluctantly complying. But they had the same thesis: This does not recover lost traffic, but it does deepen engagement of existing users.
I found this a somewhat more reasonable explanation, so I put it to Gabriel Dorosz, the former head of audience strategy at The New York Times, to hear his thoughts. I had unwittingly stumbled into a trap, however, as Dorosz has recently been doing some consulting work with the International News Media Association (INMA), which, would you believe it, has also recently launched a publisher answer engine of its own. Naturally, Dorosz was bullish.
He encouraged me to think of these publisher answer engines in two buckets. First, there are the free ones, such as DeeperDive and this stealth startup. Second, there are those available on paywalled, subscription-only sites, whose content is theoretically exclusive, un-scrapable, and quite valuable.
He cited INMA as an example of this second tranche, along with The Information, which recently launched its own chatbot called Deep Research. Earlier this year, I wrote about a similar product from Politico, available only to Politico Pro members, that scraped its vast archives of proprietary research to make compiling white papers a breeze. (Max Tani at Semafor has since reported that the Politico Pro product has had some hiccups, generating reports with fake info.)
According to Dorosz, the viability and utility of these products, as is often the case, effectively boils down to quality of the information that they are helping sort. If a publisher deals only in commodity content, which is available in other places across the web, an answer engine that recirculates that material will struggle to offer any benefit that a ChatGPT or other freely available search doesn’t.
If, however, these answer engines help (paying) readers make sense of case studies, white papers, exclusive data, and expert insights, then they add value to subscriptions and utility to users. That will keep people coming back and paying.
There is the natural question of cost and accessibility, however. Singolda, from Taboola, emphasized the point that most publishers cannot afford to build or operate their own answer engines, and they would have little luck charging visitors to use them. So, the best bet for these publishers could very well be partnering with third-party firms like Taboola that shoulder the costs of production themselves, monetize the engagement with ads, and split the revenue with the publisher. It might not be groundbreaking, but it is money.
On some level, this has been the Taboola pitch since the dawn of time. If these tools increase site engagement even 10% or 20%, and they’re free, then it’s hard to make a case against their value.
But that is the kind of logic that animated the old internet, an ecosystem where commodity content produced at scale could feasibly sustain a media business. In this new era, the post-traffic era, you could have an infinite number of recirculation tools and still no one could visit your website in the first place.