The incentive structure around grounding AI outputs on real-time related issues looks busted, and PR and other forms of strategic communication may increasingly drown out news content if the incentives for news publishers remain unclear.
As with so many other topics around AI, the underlying issue isn’t new, exactly. PR professionals already outnumber reporters by a ratio of 6.4-to-1 in e.g. the United States.
The AI angle on that ratio is this: Every kind of strategic communication from advertising to marketing to public relations seems to be increasingly investing in “Answer Engine Optimization” (AEO) to increase their chances of featuring in and influencing output (remember, it’s only called “LLM grooming” when others do it). Meanwhile, publishers are often blocking AI crawlers in frustration over what many of them experience as increasingly rapacious extraction of their intellectual property that delivers very little of the referral traffic that underpinned the de facto content-for-reach terms of trade they acquiesced to around search engines and social media for decades, and no other clear incentives either.
Strategic communications is supply-driven, and has incentives to lean in to AI
Strategic communications broadly speaking is supply-driven and underwritten by those who have money to spend on getting their message “through the right channel, to the right person, at the right time” (as the mantra goes). AEO professionals are increasingly helping their clients do that by recommending e.g. making content that is relevant to target audiences’ likely information needs and questions, building topical authority, making the content crawlable and presented as structured data so it is easy for AI crawlers to access and extract, etc.
So strategic communicators want in to AI output, because it helps them achieve their goals. (Elsewhere, I have written about this broad and heterogenous category as “promoters”.)
Publishing is demand-driven, and the incentives are unclear
Publishing, in contrast, is demand-driven and underwritten by paying users and advertisers who want to pay to reach users. Right now, crudely put, publishers invest in content, but often don’t see any real return on that investment from AI, because AI for most of them deliver tiny amounts of referrals (the traffic from search and social that could, in the past, help sell subscriptions and generate advertising revenue), and because the vast majority of them are not paid anything by any AI company.
Referrals first. Late last year, Google delivered about 6 billion referrals to the 2,576 sites in the Chartbeat network, and Facebook about 1 billion. (Numbers from data shared with the Reuters Institute.) Of course publishers would love to be paid as well. But even in the absence of payment, most will want the referrals. By comparison, in the same period in the same population of publishers, ChatGPT delivered 4.6 million referrals, and Perplexity 449,422. That’s more than three orders of magnitude between ChatGPT and Google. Referrals from by far the most popular consumer-facing AI chatbot are only little more than 1/10 of referrals from X, themselves a rounding error of overall traffic and of limited commercial value to most publishers.
Licencing deals next. Depending on what you consider a deal there are, according to Cloudflare, about 50 or so publisher-AI agreements. That’s not nothing, and some of the individual publishers are privately or publicly happy with the deals they have. But there are thousands of major and medium-size publishers globally (WAN-IFRA alone organizes more than 3,000 news publishing companies), and tens of thousands smaller ones. Considering only the major and medium-size publishers, that means less than 2% of them have any kind of deal. Meanwhile, AI companies and the various third parties who crawl content for them are making an aggressive bet premised on the claim that all or most of this is fair use and/or transformative and/or covered by text-and-data-mining exemptions etc. etc. etc., and whether that is so or not will probably take years and millions in legal fees to establish.
Given this lack of clear incentives, it is not surprising that, according to Buzzstream, 79% of top news sites block AI training bots and 71% also block AI retrieval bots.
So many news publishers do not want to go in to AI output, because it is not clear it helps them achieve their goals, and may in fact undermine them.
What does strategic communications leaning in, and many publishers blocking, entail?
I am not suggesting all strategic communications, or other content that those behind it actively invest in and make freely accessible for AI use (as they do for many other uses) is necessarily always bad for grounding, or that publisher content is always intrinsically good for every purpose or all users, or necessarily better than alternatives from other sources. The European Union, the US National Institute of Health, NASA, the US National Oceanic and Atmospheric Administration, and the World Bank are among the top 500 registered domains of the last main/monthly crawl from Common Crawl.
These are different kinds of public institutions that invest in content and make it available, also for AI. Many top universities also feature. Some are private, some are public, again, they invest in content that is made available, also for AI. Numerous private companies that aren’t publishers also figure on the list, all the way down to Deloitte at number 500. That’s good for AI companies, and arguably good for users.
At the same time, when it comes to news, the overall incentive structure looks kind of busted to me, at least for real-time output. It may only be a small part of overall use, and not the most important. But it is arguably still important, at least in a public sense. PR and other forms of strategic communication are accustomed to spend money, sometimes a lot of money, to get their content, their narrative, their version in front of people. Publishers, most of whom are for-profit, and others of whom are non-profits with bills to pay, try to make money from the content they offer to people. The first group can live with the current terms of trade around AI. The second group? Unclear. Over time, this does not point to a great equilibrium.
What are some possible medium-term scenarios for news?
Leaving strategic communications/PR aside and focusing on news, in terms of supply (publishers) and demand (AI companies willing to strike deals), abstractly, we can map the possibilities along two axis – few deals versus many deals, and limited value deals versus high value deals.
(“Value” here could potentially be referrals, the traditional currency of content-for-reach, but across both AI, social, video, and more platforms we seem to be moving away from referrals so unless other incentives are provided, value here would mean some kind of payment.)
That gives us four scenarios. They are not mutually exclusive, but provide a stylized sense of different landing zones. Given current dynamics, one of them (many deals/high value) seems to me to only be realistic if AI companies are basically forced to pay in some way.
So here we go –
Scenario 1 – Continuation of the status quo (few deals, of relatively limited value for most those who have them)
This is a scenario where the dominant trends are the current ones. So more PR, perhaps a growing market for content farms willing to enter on terms many other publishers find unattractive, few deals and in most cases of relatively limited value to participating publishers. (We fundamentally don’t really know what the current deals are worth, but in the cases where do, whatever other value they may hold they seem to represent incremental revenue.)
In this scenario, AI output at least around real-time issues is increasingly grounded in content that mostly comes from those willing to actively invest in making information accessible to AI companies, new AI-directed content farms driving “market for lemons” dynamics, and only subsidiarily a limited number of AI-darling publishers with deals. This is, by and large, a market that treats news as an information commodity, and, as was pointed out almost thirty years ago, information commodity markets don’t work.
Scenario 2 – “Programmatic-but-for-licencing” (Many deals, relatively limited value for most of those who have them)
Here we would see more licensing, mostly at very low rates. This would be a world where the marketplace begins to scale beyond a few dozens of often bespoke deals to include a much, much larger number of publishers. Cloudflare, one of the companies trying to build the infrastructure for such transactions, argues that a “licensing economy is emerging”, Microsoft is offering its “Publisher Content Marketplace”, Amazon has reportedly been planning one.
All of these (and several start-ups) come out of the technology industry. Meanwhile, the SPUR coalition of publishers – perhaps informed in part by publishers’ mixed experience with programmatic advertising infrastructures developed by tech companies – is, among other things, intending to “evaluate existing industry infrastructure and assess where new technologies or approaches are needed”, suggesting the possibility of a marketplace infrastructure developed from a publisher/supply side perspective rather than from a technology industry/demand side perspective.
In this scenario, licencing could work akin to programmatic advertising, something many publishers have as part of their revenue mix, even though the money involved can be limited.
Scenario 3 – AI companies as anchor customers for e.g. news agencies and a few others (Few deals, relatively high value for those who have them)
We might also see few, but much bigger deals. Here, deals continue to be the exception, but those who do get deals scale up what they offer to AI companies, and how they offer it.
It is noteworthy that news agencies – long accustomed to B2B wholesale deals with the requisite B2B technical infrastructure – feature prominently among those who have already struck deals. The three biggest ones, AFP, Associated Press, and Reuters all have one or more deals. EFE is also part of a deal. AFP and Associated Press both publish more than 300,000 text stories every year, and Reuters over 2 million, and the content is essentially all news reported to very high professional standards. (By comparison, a big general interest news title might publish something like 50,000 units, and much of this will not be news but entertainment, lifestyle, opinion, recipes, travel, etc.)
News agencies are also not unaccustomed to having a few key customers/ sources of revenue. The London Stock Exchange Group pays Reuters News $325 million per year, somewhere in the region of half of Reuters’ revenue. The French state provided AFP with €120 million in funding in 2025 as compensation for fulfilling a mission of general interest, about a third of AFP’s revenues.
Compared to AI investments projected by some analysts to exceed $1 trillion in 2026, and packages for some individual AI engineers reportedly running into the hundreds of millions of dollars, if an AI company believed exclusive access to news content would give it a competitive edge vis-à-vis its rivals, surely they would spend as freely on that as they do on talent, data, and energy? (Some primarily B2C news publishers could try to get in on this, as some have in the past operated what was effectively their own news agencies. You could also imagine vertical integration, but I suspect AI companies, conscious of the scrutiny this would draw from politicians and regulators, and the legal liabilities that would surely follow, will hesitate – Google considered this almost twenty years ago but decided against it.)
In this scenario, you might see a lot of content made available by a few respected major publishers.
Scenario 4 – something for everybody? (Many deals, relatively high value for those who have them)
Right now, I just do not see effective demand for this from the AI companies. Big or small, old or new, they do not seem to think they need lots of significant content deals to compete with one another. They do not seem to think they are under any legal or regulatory obligation to strike lots of deals. Unless that changes, that means this scenario seems to presuppose some kind of political intervention.
That’s a matter for citizens and their elected officials. Both the domestic politics (especially if a country has, or wants, AI companies) and the geopolitics (if they want to be on civil terms with the few, big countries that have AI companies) of this are complicated. But many politicians and some governments around the world are clearly signaling that they regard the current trends and status quo as problematic.
If elected officials want to do something, there are any number of possible interventions here, from facilitating collective management organizations to final offer arbitration to cultural levies to hypothecated taxes to statutory licensing regimes. Almost all would require legislation which is slow, uncertain, and relies on elected officials and the government of the day (including their willingness to tangle with, for example, a US government that has, under both for example the Obama administration and more vocally the current Trump administration, regarded much tech regulation in other countries as thinly-disguised protectionism). And as with all public policy, the devil is in the detail, and every intervention will necessarily have winners and losers, in publishing, and in society at large.
Where might each lead?
These four scenarios are not all mutually exclusive, but each foregrounds a different aspect and dynamic.
Under scenario #1, publishing arguably plays a smaller and smaller role in grounding AI output on real-time issues, leaving more and more space for the strategic communicators and PR professionals I started with. That seems to be the current direction. (Surely AI companies are at least aware of the risk this could pose to them over time?)
Under scenario #2 more publishers are paid, and the news they produce serve to ground AI output on real-time related issues, though on terms that may not be what they had hoped for. It could help reduce the relative role of strategic communications and PR, and increase supply of news, but if rates are mostly (very) low, may still be subject to the market for lemons problem.
Under scenario #3, a few publishers really scale up what they offer and how they offer it, paid for – at a completely different scale – by competing AI companies. This could supercharge existing winner-takes-most dynamics in publishing (this report points at related dynamics) and simultaneously provide a generally strong basis for grounding AI outputs on real-time issues in key markets, strengthen a few publishers, while potentially making AI a (even greater) challenge for many other publishers.
Under scenario #4, publishers probably have to rely on politicians to change the terms of trade in ways that are to their benefit – and, if done well, hopefully also serve a wider public. This one is, to an even greater extent than the others, incredibly uncertain, so who knows that that would look like (other than probably necessitating a more direct political intervention in the news media market).
What have I missed? (Other than the data here being limited so a lot of this necessarily speculative, and the majority of these scenarios mostly about the context of high-income democracies I am most familiar with, not representative of the situation and possible paths elsewhere.) Other plausible scenarios for where this goes? It is early days yet, and while there are observable dynamics, the situation remains shifting and contested. Let me know where you think it is heading.
rasmuskleisnielsen.net 