TMB – which owns brands like Reader’s Digest, Taste of Home, and user-generated video licensing company Jukin Media – doesn’t want to give LLMs wide access to its content. “We just don’t want to hand over the keys,” said Salamon at Digiday’s Publishing Summit in Miami, Fl., in September.

So far, the only AI company TMB has a deal with is Prorata, which gives half of its total revenue to publisher partners on a recurring basis, depending on how often their content powers AI responses. So far, TMB hasn’t made any revenue from that deal, Salamon said.

But Salamon understands the significance of these deals for the future of TMB’s business. “It’s existential in terms of: which direction do we want to go and where do we not want to go,” he said.

For now, TMB is “monitoring” the AI dealmaking landscape — seeing which tech companies start favoring a pay-per-usage model over lump sum deals, which ones start giving publishers more control over how their content is used, and waiting for the outcomes of a number of legal proceedings against AI companies for scraping publishers’ content without compensation.

Companies from The New York Times to Penske have sued tech companies for copyright infringement, claiming that they are taking publishers’ content and using it to train or supply their LLMs with the information. Those cases are ongoing.

Here is more of what Salamon shared on stage about TMB’s approach to AI licensing at Digiday’s Publishing Summit last month:

What TMB doesn’t want in an AI licensing deal

Numerous companies have signed AI content licensing deals with companies like OpenAI and Amazon. What’s TMB waiting for?

Salamon said he’s closely watching the copyright infringement lawsuits brought by publishers against these companies – as well as waiting for a deal to come his way that seems “fair.”

“AI companies really want everything. Often it’s like, ‘we want it all.’ We want to do whatever we want with this in perpetuity, one time, here’s your check, and you never need to work with us ever again,” he said. “They want to [train their LLMs on our content and] be able to create replicas of our content or build them into generative models, so that our recipes, our DIY steps or instructional guides, our 100 years of stories, or all of our video content can just be generated and new versions can be created.” The worry is that TMB’s content can be replicated – and eventually replaced – by generative AI, he added.

So far, those unfavorable usage terms have been the dealbreakers, Salamon said.

“[Right now,] it’s unlimited usage. We do that occasionally in some licenses, if we’re going to a movie or a TV show or a film – if this piece needs to exist in perpetuity in the world, and that’s a license that you can buy. But again, when it’s any application, any usage… that’s the crux of the problem,” he said.

What TMB wants in a deal

So what would an ideal AI content licensing deal look like, in Salamon’s eyes?

“We’re getting paid for access to our content as it’s getting accessed. We have an ongoing, recurring revenue stream. And if there is training, we are able to create constraints around how that is used and where it will be placed, and how that content will be leveraged,” he said. 

While a lump sum AI content licensing deal from a tech company “would make me look great,” it would need to come with “recurring revenue,” he added.

Salamon said some tech companies (he declined to say which ones) are having more open conversations around pay-per-use or RAG (AI systems retrieving live content from publishers in real time) deal structures.

“It mirrors more what we’re used to…. We get paid, or we get a visit, for every time someone finds our content… It just mirrors what we’re comfortable with. It’s a longer-term solution,” he said. “It gives us this ability to work in a way that’s actually a lot more sustainable long term, so we’re not disintermediating ourselves completely.”

But onstage, Salamon said he was still unsure about these models working the way they’re supposed to, and actually bringing in revenue or traffic to publishers.

Other things on Salamon’s wishlist for an AI content licensing deal? Being able to revoke or withdraw content accessible to an AI model, as well as having a clear understanding of which content is being used for training and how long that content gets stored.

“Given the history we have as a licensing company, we’re not seeing that kind of detail and that kind of surrender on the side of the LLMs that we’ve spoken with,” he said. “Especially for training, the biggest interest [is in our content library,] especially around video… We’d like to see, what is this being used for? Is it being used to create new versions of user-generated clips? Is it being created for filters and effects and something more imagery-based?… Ultimately, [is this going] to make our licensing business irrelevant? That would not be very strategic for us.”

Blocking AI bots may give TMB some leverage

TMB uses Tollbit and Cloudflare’s AI bot-blocking mechanisms to put up some barriers to LLMs scraping content without payment, according to Salamon.

“Tollbit is allowing us to set up a signpost saying, ‘if you want to scrape our content or access our content, here’s the price you’re going to get,’” he said. “Cloudflare has the ability to enforce that.”

Though some AI bots are bypassing these bot-blocking efforts, these “gates” help give TMB a bit more leverage when negotiating AI content licensing deals, Salamon said.

“It asserts us legally… If there is a major settlement or a major [legal] decision, we can say, ‘Look, we’ve established those rates. We were bypassed. And now there’s much more ground for a claim,” he said.