Billing the Bots

As A.I. search reshapes the internet, publishers are facing a new reality: fewer human visitors and more bots consuming their content. A recent study by Tollbit found that A.I. search referrals generate 96% less traffic than traditional Google searches—meaning publishers face a real possibility of being left out of the loop entirely unless they find a way to get paid.

Enter Olivia Joslin, co-founder of Tollbit, a VC-backed startup helping publishers charge A.I. companies for access to their work. As big tech players like Google and OpenAI negotiate licensing deals—and fight legal battles—Tollbit is pushing a different model: charging A.I. bots when they scrape content, much like a digital toll system.

I spoke with Joslin about why publishers need to rethink their business models, what types of content A.I. companies value most, and whether the industry is heading for a Spotify-style solution, or a Napster-like free-for-all.

Our conversation below has been lightly edited for style.

A recent study by Tollbit found that referral traffic from A.I. searches are about 96% lower than traditional Google search. Were you surprised at the results of this study? And what does it mean for publishers going forward?

We were not surprised that A.I. bot click-through rates were tiny in comparison with traditional Google search, ranging from 0.36% to 0.7%. This aligns with most users' anecdotal experiences; A.I. chat interfaces are great at directly answering your questions, saving you the effort of clicking through multiple links.

I don’t think human curiosity goes away—the amount of questions we ask won’t decrease. In fact, A.I. consumes far more efficiently than humans, so we foresee content access exploding.

However, what this means for publishers is that humans may never make it to sites. As evidenced in our report, increasingly site visitors will be A.I. bot “eyeballs” which are accessing content to answer questions. The publishing industry needs a way to charge these A.I. “readers” so they can be properly compensated if humans aren’t the ones visiting sites.

This week, both Google and OpenAI submitted policy proposals to the Trump administration, advocating for more flexible copyright laws to facilitate A.I. training on copyrighted materials. What did you make of that? And if A.I. companies successfully push for weaker copyright protections, does that undermine the business model of licensing content?

The legality of training is a complex issue that will ultimately be decided in courts. We’ve always told publishers to pursue a deal for training if they can get one, but that it won’t be enough to run their business. There’s a bigger, recurring revenue opportunity for publishers that falls beyond the scope of training models.

A.I. bots and agents built on top of models need to continuously access content. No one can store the entirety of the Internet and these bots need to ensure they have the latest accurate information. This is known as RAG, inference, or grounding. These A.I. bots can be charged every time they come to scrape content, which is the model we’ve pioneered at TollBt.

Some media companies have struck licensing deals with OpenAI and Google, while others are suing. What do you make of these two divergent approaches?

This is an interesting question because we have a handful of publishers on TollBit that fall on either side. There are publishers whose 1:1 licensing deals with big tech companies flow through us and publishers in high-profile litigation who are on the platform.

Publishers that take either approach are some of our most powerful thought partners because they see that a better, scalable solution needs to exist. Litigation and 1:1 licensing deals don’t scale. To truly have a thriving content creation & A.I. ecosystem, a solution needs to work for the millions of A.I. agents that will need to access information from tens of thousands of content providers.

How do you respond to critics who say that A.I. companies should be paying more for publisher content—or not using it at all?

Early into TollBit, we investigated how these publisher deals were valued. A.I. companies understandably struggle to predict all the sources of content they need and how frequently they’ll use it. Through these conversations, we realized there wasn’t an unwillingness to pay, there was just significant friction for both sides in formulating large upfront partnerships given this lack of clarity. This is why we created a pay-per-use model.

We’re in the Napster days and scraping/stealing is rampant due to this partnership bottleneck problem. The industry needs a Spotify-like solution for accessing licensed content.

What types of content do A.I. companies find most valuable, and why?

You probably went out and read the news this morning. If you consider a thought experiment where all of our minds were “trained” on every single fact about humanity. None of us know what happened today. We all had to go out and read.

In order for A.I. apps to displace traditional search, they must be able to do the same—otherwise their users will just return to Google. Therein, publishers have significant leverage. One of the most common requests from A.I. companies we hear is for a real-time feed of news. We’ve also seen that content that is truly unique or not substitutable, like local news, commands a premium too.

Can publishers realistically make meaningful revenue from licensing deals, or will this be a small fraction of their business?

As A.I. adoption increases, there are one of two outcomes. In one, publishers continue to focus on eking out an existence by monetizing their declining human visitors. In the other, publishers lean in to set up a new economic model that cashes in on charging the explosion of A.I. apps and agents accessing their content.

Thousands of apps are built on GPT tools that fetch content. Every company is going to deploy A.I. agents for automation. Many of these agents need to get content and data from the web. This is a new class of visitors alongside human visitors.

We work hard to help publishers understand that the right answer to this is not “licensing” in the traditional sense. As an analogy, you wouldn’t call paying the toll on a highway a “licensing deal.”

With human visitors, it was possible to monetize by showing an ad. Our technology makes it possible to scalably levy a “toll” on A.I. apps and agents for accessing content because you can’t show ads to these visitors. Without a standard like TollBit, A.I. companies, indeed every company, will have an arbitrage on content in the future.

• The Trump administration put all full time Voice of America employees on paid administrative leave, a move that effectively shut down the outlet. The dramatic action, relayed to employees by Kari Lake, came after Trump signed an executive order calling for parent organization U.S. Agency for Global Media to shut down activities not required by law. [NPR]

  • "The Voice of America has been silenced," a VOA correspondent told Brian Stelter. [CNN]

• For the first time in 140 years, the Gridiron Club Dinner ditched its toast to the president, opting instead to honor the First Amendment—yet another sign of the news media’s strained relationship with Donald Trump. [POLITICO]

  • Judy Woodruff opened up the dinner cracking jokes about Elon Musk and Pete Hegseth, per Shawn McCreesh. [NYT]

• Protests against Musk continued to grow this weekend outside Tesla showrooms. [CNN]

• As was expected, the "Snow White" premiere was heavily scaled back, with the only press on the red carpet being Disney employees. [Variety]

  • The first reactions for the film, though, are positive! [THR]

  • ICYMI: Jon Passantino wrote in his new Saturday column that the Bob Iger-led Disney buckled “under sustained right-wing pressure,” instead of standing with its creatives. [Status]

• Joe Russo ripped Harvey Weinstein, telling Jonathan Dean that the disgraced filmmaker "vilified mainstream movies" and kept popular films out of the Oscars. [Sunday Times]

• Asked about the future of Ant-Man, Paul Rudd put the ball in the court of the Russo brothers, who are directing the next "Avengers" movies: "They know where I'm at, and that's all I can say." [Collider]

• More synergy at Warner Bros. Discovery: The company announced plans to air "Hacks" on TBS before its season four premiere on Max. The conglomerate also said it will broadcast "True Detective: Night Country" on TNT. [Variety]

• 🚨 We're getting a direct sequel to "Cloverfield," per Babak Anvari, who said we're "going to get something amazing." [Deadline]

• Former Obama White House spokesperson Eric Schultz spoke to Marc Malkin about what it was like working with Robert De Niro on “Zero Day.” [Variety]

• Will Smith announced he will release his first album in 20 years. It's due out March 28. [Billboard]

• “Novocaine” topped 2025’s slowest box office weekend of 2025, pocketing $8.7 million. Meanwhile, “Black Bag” debuted to $7.5 million and “The Day the Earth Blew Up: A Looney Tunes Movie” counted $3.2 million in receipts.

• “Mickey 17” declined 60% in its second weekend, adding $7.5 million to its coffers.

• “Captain America: Brave New World” pulled in $5.5 million.

[Data via Box Office Mojo]