The Game Theory of News Publishers’ AI Dilemma
This week’s newsletter isn’t about tech comms per se but about what I think is the most interesting issue right now at the intersection of tech and media: the combination of dealmaking and conflict between news publishers and AI companies.
If you haven’t been following, Axel Springer, News Corp, and Vox Media have signed muti-year agreements for undisclosed sums with OpenAI to integrate their news coverage into the latter’s models and products. (It’s rumored Apple and Google are also negotiating similar agreements, though none are announced as of yet.) Meanwhile, other publishers, most prominently the New York Times, have sued OpenAI on copyright infringement grounds, hoping to foreclose any use of their work for AI model training without their approval.
Adding fuel to the fire was news this week that Perplexity is allegedly bypassing publishers’ Robots Exclusion Protocols to crawl their content, even when those publishers’ robots.txt files explicitly prohibit Perplexity’s crawlers. (In the interests of fairness, Perplexity’s CEO responded in an interview with Fast Company.)1
There has been much rending of garments about the deals among Media Twitter—about whether these publishers are signing their own slow-death warrants, about journalistic ethics and norms regarding proper attribution of information, and not-so-thinly-veiled loathing by journalists of the suits in their companies making the call to take the cash.
But I haven’t seen much discussion of the actual decision-making factors facing publishers. It’s a fascinating example of game theory playing out in real life. And in playing out the game, it seems pretty clear: the most rational move any individual publisher can make right now is to take the money and free-ride on others’ legal challenges against AI companies.
Let’s look at the three basic choices sitting before publishing executives:
Take the money: You get a nice chunk of change upfront with some ongoing royalties, perhaps even a seat at the table in discussions about how your content is used. It immediately removes any ambiguity and comes with the promise of future collaboration. The downside is that your deal could look worse over time due to the immaturity of the market and the relevant nascency of these AI products in terms of market penetration. Your board is probably happy, but your journalist employees are pissed.
Sue the bastards: This offers the potential of large settlements (larger than the deal amounts on the table pre-litigation, at least) or court-ordered compensation. The tradeoff? High legal costs, years of uncertainty, and the possibility of an adverse court ruling that leaves you with nothing. This choice will likely fire up your journalist employees but slightly terrify your board, given the risks and costs.
Do nothing: Sit back and hope the market and legal landscape evolve in your favor without having to take any action. It's low risk but low reward, too, since you’re forgoing both certain revenue and potential settlement money. Probably untenable as a strategy with both your employees and your board on a medium- to long-term basis, but acceptable in the short-term in that you keep all your options open.
There’s no world where publishers all make the same choice monolithically (and there are legal bars to that level of coordination anyway), so we’re dealing with some mix of choices among these options.
Legal challenges on copyright grounds like the one the Times is mounting against OpenAI will be lengthy, costly, and ultimately destined for the U.S. Supreme Court to adjudicate. Enormous corporate cases like these can take seven to ten years (or more) to fully resolve absent settlements. And given the players, the stakes, and the complexity of these cases and inevitable appeals, each side’s legal bills will likely go well into nine digits when all is said and done.
But let’s say the publishers ultimately win in court. Sure, the Times stands to reap larger financial rewards, but so does every other publisher in a world where using copyrighted material for model training breaks the law. It’s a classic economic free-rider problem: everyone can reap the benefits whether they fight or not, so there’s little to no incentive for any individual publisher to join in.
U.S. Congressional action in the next decade to change copyright law and draw a bright line on the issue of model training is possible but not probable. Democrats would seem most likely to take this sort of action if they were able to attain majorities in both chambers. Still, it’s very unclear how high up the list this issue would be among all other legislative priorities (healthcare, abortion, immigration, housing, etc.). Republicans seem unlikely to act—if there’s any sector of the economy Republicans hate more than Big Tech, it’s journalism and the news media. And at least some of the tech folks are good for big campaign checks.
Just as with the courts, though, let’s say Congress does act, and the White House signs a copyright law change into law. If you’ve taken the deals on offer from AI companies, you have the benefits of 1) having money from the deals and 2) increased leverage in your next negotiation due to the legislative change. Not taking the deals now just makes you that much poorer later.
While the growth outlook for digital advertising over the next decade looks pretty robust (let’s say a 10% compound annual growth rate to ballpark it), vanishingly little of it is likely to be captured by news publishers, as is the case today. Their growth, optimistically, will likely be flat to slightly up but still sub-inflation. This is all to say that publishers need diversified revenue streams that have real possibilities for growth, now. These AI deals offer one possibility.
The objection I anticipate to all this from journalist friends who are vehemently against these deals (and to be clear, that’s not a crazy position, it’s just a sub-optimal one strategy-wise in my estimation) is that, in making their organization’s reporting accessible in some form on these AI platforms, they’re helping habituate consumers to seeking out news there instead of on publisher properties.
Maybe? It’s not as if there aren’t already thousands of places (some very popular) to get poorly attributed, aggregated news online that also work around paywalls. Consumer habits will likely come down to how product experiences evolve, an area in which these AI-publisher deals could be determinative. And there’s still a vast and still-widening consumer trust chasm AI platforms have to overcome generally, not just for news.
Ultimately, the best path forward for news publishers, deals or no deals, AI or no AI, is doing everything you can to build a direct connection with a brand-loyal audience. Whether it’s Facebook or OpenAI, distribution constantly changes. The control you as a publisher can exert over the presentation of your content doesn’t.
As a corporate comms aside: it seems really fraught to trust Fast Company or any publication to carry your message on this issue with full fidelity? At least publish something on your blog that lives alongside the interview, contains everything you want to get out publicly, and can be a single source of truth for others. My email’s on my site, Aravind.