Sign or Sue? Tollbit’s Third Option for Content Monization

Hi Toshit, thank you for joining us. To kick things off, tell us about the origins of Tollbit. What drove you to start the company, and what mission were you aiming to accomplish?

We started Tollbit in late 2023. At the time, ChatGPT was blowing up and there was a lot of talk about copyright infringement and IP theft with regards to AI and training.

Yes, this is an IP, copyright, and regulatory issue, but it’s also more than that – it’s a monetization issue. Publishers can litigate or do their one-on-one deals, but that’s not going to run their newsrooms. Those are Band-Aid solutions. The reason we got started was because publishers need a recurring revenue stream, and we believe we have the know-how to make this possible. 

We believe that RAG (retrieval-augmented generation) is going to be a big part of the future. It means that for an AI engine to answer a question, it has to go out and scrape a bunch of sites and read a bunch of documents to give an accurate answer. But there are still issues with this. For one, the models don’t know what happened the day after they were trained and, even if they were trained recently, they can hallucinate. So, the only way to correct that is to go out and fetch that information in real-time. We saw that as an opportunity to monetize by charging them every time AI comes in to scrape content.  

Is it correct to say that your primary customers are publishers? What kind of publishers are you mainly working with?  

When we were first fundraising, the first few venture capitalists didn’t want to invest because they thought we were a media-adjacent company or they had this thesis that media is dying, and AI is going to make it happen even faster. What they didn’t really understand is that this is not just a media problem. This was coming for everyone on the Internet.  

To us, a publisher is anyone who produces content or data on the Internet. Everyone gets scraped. This includes newspapers and digital publishers, but it also applies to the Airbnb and Shopify of this world.  

We have, until the beginning of this year, solely focused on digital news publishers of all sizes. We’re especially excited about local news and niche news sources because some of the most interesting content out there is unique content. Uniqueness commands a premium because it’s not replaceable.  

How are you processing payments between AI companies and publishers? What is Tollbit’s business model? 

The old way of doing partnerships was paperwork and human-heavy. AI companies cannot strike deals fast enough in that old world. Imagine you are an AI company: How would you know who to make deals with? You could not possibly make deals with content providers fast enough to cover the content for all the questions that you, me, or the person down the road is going to ask.  

What was missing was a machine-to-machine exchange of value without a human in the loop. For instance, for the AI company to say, “hey, I need this content from you. Tell me your price and I’ll pay for you to give me the content immediately.”  

This has to be a very fast system, and the payments have to be in very small amounts because the AI company is not going to pay if it’s too big. But if we could get it down to the scope of micropayments, it’s possible. That’s where Tollbit helps determine the right fee to charge per access. 

We have publishers who use this in slightly different ways. But I think the key theme through it all is that Tollbit is like a front door for AI: It is a sanctioned way to access content. Some publishers charge for it, others don’t. We charge a very small transaction fee on top of each of those content accesses. 

Building on the question of payments, how do you determine the value of each transaction?  

We are making the market for this in real-time, so there is a lot still to be determined. A lot of the exercise that we do with both publishers and AI companies is price discovery.  

Both sides seem to agree that a more usage-based pricing makes sense. It’s easy to understand why the alternative wouldn’t make sense: If a publisher goes to an AI company and says I want 500,000 dollars for my content, it’s going to take a long time to get the deal done. You have to prove that content’s worth that much and there has to be willingness to pay.  

From the AI company’s perspective, they’re thinking, “I don’t know how much I’m going to use your content so I’m not going to pay a lump sum ahead of time for that.” In 2023 and early parts of 2024, some publishers were able to strike these kinds of deals. Those have dried up now.  

So how does typical price discovery happen? If you are a publisher, your concern is that if you get scraped by an AI company, your content shows up in Perplexity or You.com or Google’s blue box at the top of search. You’re afraid that you’re not getting any traffic.  

What is your lost cost for not getting that traffic? Every publisher knows their ad CPM rates and their RPMs (how much they’re earning per thousand impressions). This gives us a basis to say, “Okay, so you lost a page view. What is your RPM?”. Maybe that’s the right place to start, to compensate them for not sending them that traffic.  

We then take that back to AI companies, and they will tell us that they can’t conceivably pay every publisher this rate. Why? Well, you have to consider the difference between how AI consumes content different from humans. When we (as humans) Google something, we click the first link, maybe the second, and then move on. AI doesn’t do that. It will read all 10, 20, 100 links if it has to, to get your answer. So, the amount of content access is going to go up. AI company’s pushback is that they can’t pay the level of what one human access would have cost to 10 or 20 publishers, because the 10th publisher on that list may never have even got traffic from humans anyway.  

So now we come back to saying, “Okay, well, let’s take your RPM rates and let’s apply a little bit of a discount on it”. The revenue per AI visit might be lower than what one human visit nets you. This is how we’ve figured out what that pricing might come out to be in the end.  

There is also the question of uniqueness. If I can find that same article in five or ten different places, that article is very replaceable. 

I’m curious to know, can publishers determine with you what kind of content they want to be accessed by AI companies versus not accessed? How does Tollbit manage what gets scraped versus what doesn’t?  

We give publishers the ability to set rates and rules for what can and can’t be accessed, even down to specific pages. We track the visits to the site, then we index them to figure out which of them are scrapers. We show them visits per page for their content and then they can drill down further. They can see for any particular page what the breakdown was between Anthropic or OpenAI or Microsoft.  

We also added referral data so they can now see in the product how much referral traffic they’re getting from any of these AI platforms. One of the things AI companies are claiming with their search products is that it’s going to send referrals back, that it’s going to be like normal search. We have a report coming out in a few weeks on this, but we can say that it’s not like your typical search. 

Was there any surprising discovery you gathered from your analytics?  

We made the analytics tool for the business side of operations to understand who to do deals with and who is scraping you. But now the editorial side uses the analytics tool because they want to see how soon after they publish something does it show up in these AI systems and what articles are being scraped more often. This gives them clues into what people are asking on these platforms. So now, they can craft their content strategy around this, which has been cool to see. 

Thanks again for taking the time to share your insights with us, Toshit. Before I let you go, is there anything you wish we talked about or that you would like to share?  

I think what’s important to understand is that we’re really just at the start. We don’t know how quickly the market’s going to open up. At the beginning of the year, people still didn’t know what Perplexity was. We would go out to publish to every media executive in New York and they would never have heard of it. Now, everyone does. I think it’s a testament to how quickly this world is evolving and what sort of opportunity there is. 

I think there is an arbitrage on content right now. We tell publishers we have to do this now because if you wait two or three years from now, it will be too late. AI companies will have built their businesses too far along, the business models would have been set already. And then, unfortunately, I think they can dictate the terms on which they decide to give you revenue instead. So, now really is the chance to get ahead of the curve.  

This interview was edited for clarity. A special thanks to Toshit for taking the time to chat with us.