Deal or No Deal - Big AI Meets Media

Deal or No Deal.

As a long time biz dev guy, my first inclination is always to do the deal. A smart, winning deal, but a deal nonetheless. Sharks die if they stop swimming or so the saying goes; though I’ve never actually bothered to check if this is actually true.

As a deal and media junky, I’ve tried to parse reporting on the discussions between media companies and the major LLM developers, or Big AI as I'll call them here, to get a sense of the negotiations underway and what deals will continue to emerge. And I’ve been thinking about what I would recommend to my CEO boss or clients weighing their options, given all the second guessing that accompanies any new partnership announcement.

First my observations:

• Replacing Meta Facebook News Dollars - OpenAI is offering publishers between $1MM and $5MM annually to license content for its models, The Information reported. To put this in context, OpenAI’s licensing fees could replace or (for some) even meaningfully exceed the dollars lost when Facebook shut down its news service. That is high-margin revenue to be sure and if publishers can stack deals from multiple Big AI players, perhaps it's the difference between breaking even and losing money. But just as with the Meta deals, it won't solve underlying business model challenges. At the current moment, AI feels like a more existential threat than Meta or Alphabet Inc. ever has. And while we could spend a lot of time debating that point, the behavior of many media execs appears to make that very case.
• Matthias D?pfner is No Dummy - Existential threat notwithstanding, Axel Springer appears to be reaping the rewards of signing a second (if you count the The Associated Press ) to market deal announced in December 2023 and, according to the FT, “worth tens of millions of euros a year.” That seems like a lot of money compared to what the general market is seeing, but D?pfner appears to have won a premium as one of the few executives who can deliver brands in followed by regulators and lawmakers in both the EU and U.S. The deal reportedly includes a flat payment for historical content already consumed by OpenAI and included in its models. This sets a precedent for retroactive payments for use of old content outside of an agreement and appears to undermine OpenAI’s fair use arguments. Perhaps, OpenAI was willing to hand waive on this point to demonstrate on two continents that they are actively supporting the news business and, thus, legislative intervention is unnecessary. “The market is figuring it out” reads well in German and English. The FT also reported that “there are … ‘kickers’ — in effect extra payments — for popular content, meaning the media group will be paid more each time its articles are used by AI.” I’m curious how this will play out and if it involves some venn diagram of Axel topical content and user queries or if it is based on how often a user specifically references an Axel brand. Either way, it could help both sides, by cushioning OpenAI from overpaying for content that isn’t used, while providing value to Axel to offset from the reality that the interactive nature of chat-based AI will prevent users from interacting with or visiting the underlying sources. Finally, OpenAI promised “attribution and links to the full articles for transparency and further information.” This sounds similar to what 谷歌 provides in its Search Generative Experience, but it's not clear how effective or meaningful this attribution is if users get all their needs solved by the LLM answers. Notably, Axel’s deal is non-exclusive, because what media executive would agree to that? My guess is exclusive access to vertical-specific data and content will be the next battleground for Big AI, but the foundational deals will come first to cover the largest players whose content has already been ingested. Axel appears to have covered all its bases in this deal, though some argue that any agreement with Big AI is a bad deal. The loudest critics seem to be labor leaders who believe individual writers should get paid when their content is relicensed to Big AI, though that is not the trade off salaried staff writers sign up for when they take full-time writing and editing jobs. Would the unions prefer a world entirely of freelancers who are paid less for a one-off content license? Sometimes the rhetoric makes it seem like it.
• The Times Wants Its Premium - Given the New York Times’s singular importance in the English-language news landscape, I would be shocked if it had not received an offer on par with the deal won by Axel, despite only being able to provide content in a single language. The Times is the unicorn of the news business and it is used to securing many multiples of what most other organizations are able to extract from platform publisher programs (the exception has been News Corp. where Rupert Murdoch has used regulatory leverage in Australia to secure outsized deals, though the Wall Street Journal only received $10MM as part of the Facebook News program vs $20MM for the Times). The Suzlbergers, however, chose to sue 微软 and OpenAI, which signals to me that they did not feel like they were receiving their above market offer and needed a new source of leverage. Perhaps the Times wants additional concessions that Axel or the AP weren’t able to secure, but as documented above D?pfner’s deal feels pretty comprehensive. I suppose it is possible that the Times is trying to make a larger legal and moral point about the use of copyrighted content in LLMs, but I think they’d take a deal if the dollars were right. For evidence of that, see its $100MM, three-year deal with Google announced in May 2023. That arrangement covers a wide-range of Google products and services. It also probably prevents the Times from suing Google, if it contains the search company’s standard provisions in such deals. So we know the Times was willing to sit on its hands for the right size check. That said, removing content from Google’s search engine (the only way to opt out of Google’s generative AI experience) is not a business risk that even the Times is willing to take. There’s little such downside though in disappearing from OpenAI’s index.?
• Even the Most Cautious Are Playing Offense - I was a little surprised to see my old boss, Condé Nast CEO Roger Lynch , testify before a Senate subcommittee hearing on the “Oversight of AI: The Future of Journalism.” In my experience, Conde prefers to operate in the background when it comes to regulatory issues. It has been loath to take its criticisms public when it might alienate a major industry player, most of whom are also advertisers and/or strategic partners. But as Lynch testified, AI “threatens the viability of the media ecosystem.” The threat is greatest for a general interest, web-focused publisher like Conde. I read Lynch’s testimony as part negotiating tactic and part Hail Mary to push Congress to act. But my guess is that he’s not holding his breath on lawmakers rallying to Conde’s aid and instead wanted to signal that Conde’s influence is broader than Big AI thinks and therefore the company should be treated more like Axel or the Times. Lynch even namechecked Google in his testimony, which shows the level of concern, but again also the bind media companies face. “Most publishers generate a substantial amount of traffic from search,” Lynch testified. “In order to opt out of the new search engines powered by AI, such as Google’s ‘Search Generative Experience’, we would have to opt out of search, which would materially damage our businesses.” It is one thing to call out OpenAI, but when a company like Google functions as your main traffic source, dominant video distribution platform, ad server analytics provider, and enterprise business productivity suite, things are more complicated and taking them on in public raises the degree of difficulty. Even if legislation materializes to support publishers, I don’t have much faith, given the massive volume of content ingested by Big AI, that any mathematical formula developed by bureaucrats to govern payments for content previously ingested or ingested in the future, will appropriately compensate publishers, especially premium ones. Staying silent, though, doesn’t help Conde either.
• Money Can Buy Friends - Microsoft and OpenAI’s marketing push continued with the announcement that it was paying SEMAFOR to launch, help power and brand Signals, a new breaking news feed. Semafor co-founder Ben Smith emphasized that Open AI’s and Microsoft's tools will only be used for research, not to write the content. Semafor was an interesting choice of partner here since they don’t produce enough content to train an LLM, but they do focus on a thought leader and policy maker audience that Microsoft and OpenAI are looking to influence. And if you’re a start-up brand like Semafor there is no way you can turn down the sponsorship money. Personally, I think AI research tools that serve to amplify the powers and reporting of journalists is the most (only?) promising part of AI as it relates to the future health of the news business. And it hasn’t been clear who would fund development of those tools, since enterprise media software isn’t exactly a focus of VC firms. Maybe marketing dollars can help round out the funding coming from non-profits like the Knight Foundation. Indeed, in addition to the Semafor deal, Microsoft also touted partnerships with the Craig Newmark School of Journalism, the Online News Association, and the GroundTruth Project.?

I’ve seen some suggest that media companies are dooming themselves by signing deals with Big AI and that they are repeating the same mistake made in taking money from Big Platform. This may be the case, but you can’t pretend that the environment where you operate doesn’t exist. Yes, AI is the looming existential threat. It promises to be even more destructive than the walled garden and performance marketing engines built by the major platforms that ate the open web’s audience and ad market. The AI threat, however, is still largely theoretical as chat and generative AI have yet to meaningfully cannibalize media audiences (at least I haven’t seen any evidence of this cannibalization; if you have, send it my way!). Taking money from Big AI doesn’t create a new playing field, it only helps cushion media for the one that is to come. Moreover, removing content on a go-forward basis won’t save most media businesses because the LLM models aren’t going backwards and there will be providers in nearly every vertical willing to make the content for free or supply it to Big AI for some cash and the hope of traffic. Unilateral disarmament isn’t a strategy for winning the war.?

So, I think it would be a mistake for most publishers to turn down money from Microsoft / OpenAI, Google, Anthropic or whoever will write the checks. Again, I would make sure to do smart, non-exclusive deals that maximize profitability and whatever protections are possible. To be clear, I think this is the path for most general entertainment, information, and news providers. This is probably also true for most affiliate content brands and recipe providers. Big AI’s content quality is already (or will soon be sufficient) to satisfy users across evergreen-focused categories and, hope for a legislative-driven rescue notwithstanding, I don’t see the genie going back in the bottle.?

The story is different for folks that truly have a differentiated content vertical and/or audience. These publishers largely sit in the B2B space and may be able to build their own product on top of open source or other AI tools. Aviation Week or Oil and Gas Journal should be able to figure out the data product that best serves their audiences. Almost a year ago, Bloomberg announced the development of BloombergGPT and it makes sense for a company that has built a terminal business that is fundamentally an information and data management tool to go their own way. Perhaps, Vogue Runway can leverage its large, historic database of runway images to create a proprietary, AI-driven tool for retail and fashion executives, but I’d take Big AI’s money for Vogue itself.

I’m generally skeptical that media companies can innovate in a meaningful way (the answer is not to refashion all text content into a chatbot) and publishers’ track record of developing innovative products or experiences is bad. There’s a reason Calm wasn’t built by Self or Noom by The Food Network. But maybe unleashing AI on a proprietary data set is enough to unlock additional value for niche business users or enthusiasts. And the last thing I’d want if I controlled a unique data set was for it to become a tool of folks with asynchronous compute, unless I was content with folding shop and collecting a licensing check. It promises to be a fascinating few years, if nothing else.

If you’re interested in talking about media and AI negotiations or you need help develop an AI partner and product strategy, please reach out to me at Mount Prospect Partners.