In the beginning, writers wrote for readers. When readers hoped they might get something from reading, they were likely to sacrifice more than time and pay upfront for the promise of reading something of value. A monetary exchange based on a hope and a promise created livelihoods for writers and their publishers.
Later, after this supply-driven market took hold and flourished, a second pillar of income for writers and publishers emerged. Those with a commercial interest in reaching readers purchased ads that appeared alongside the good (reading) stuff. If played right, this revenue stream was not cannibalising but incremental to the pockets of publishers and writers. Ad revenue made it possible to reduce the sacrifice (price) a single reader had to pay. This broader reach (lower rates --> affordable by more price-sensitive segments --> bigger circulation) drove additional revenue for writers, publishers and advertisers. Various points of equilibriums of the above developed and evolved, delineated by topics and geography.
Ads improved the possible livelihoods writers and publishers could attain. Not just of news, magazines, catalogues, directories, pamphlets, etc; books too. Back in the day, before putting down an affordable, mass-market Penguin novel, I always had a look at the list of other novels the publishers thought I might be interested in. There, the myth of Amazon inventing, “you bought this, you might like this book too…” is busted.
In the mid-twenties of the 21st century, writers and publishers can no longer afford to write just for human eyes. Increasingly, writing is done for satisfying machines. In a recent report, Jerome Segura from DataDome writes about the online demand for content. To be clear, it is not just for text, it is also for audio and video offered alongside. He writes that demand by machines is already vast and its growth is accelerating. By some measures, it has already overtaken demand by humans.
The beginnings of a new, additional demand signal is detectable: AI agents on behalf of humans, sometimes one or even many steps removed. The AI agents are looking for the best way to quench thirst for information. Recall, the “hoped-for” promise of the writer-reader market. The demand discerned originates from Model Context Protocol (MCP) servers. This is how things will play out in the late 20s: as AI agents will become an important (maybe even primary) intermediary of published content, writers and publishers will begin to realize a third revenue source: getting paid (+ attribution) for content offered in sample, summary or in full to AI agents.
In recent years, LLM labs have created valuable capabilities by training their models on authors’ and publishers’ content either at zero or near-zero cost. If deals had a monetary component to them, the terms agreed were more about avoiding litigation. The exchange had very little to do with the value of the data, the copy. The livelihoods damaged by this Wild West (and East) episode are real, the economic benefit to some untold.
Marketplace for content
It’s time and it is possible to turn the tables. We are in the middle, not even the end, of the first scene of an Odyssey in many acts. Expect some sequels too. What happened at the beginning of Scene 1 – free for all – doesn’t mean content served to AI agents in the future must remain uncompensated. To keep writing flourishing, AI agents coming in good faith, must and can be made to pay. AI agents coming in bad faith must be blocked. All of them.
As restitution claims for the past are worked through, two clear openings are emerging. First, there is real potential to monetise future ongoing training, calibration, and fine-tuning of LLMs, and maybe even more interestingly, SLMs. A marketplace for content, archives, data sets at varying levels of raw or processed copy and quality is in its early days. Think of the Toyosu fish market in Tokyo or the Aalsmeer Flower Auction, but for content. Fresh, new catch will be made available by those who have rights to sell, bulk distribution-type deals will be made. They will also encompass the value of attribution which allow for machine and human discovery of the source material.
The second opening is Retrieval Augmented Generation oriented. In this new area of machine-related revenue for writers and publishers, each time information is needed and served will be a discrete transaction, with a known AI Agent. Metering of (or micropayments by) the querying AI agent, and by extension, the tech platform behind it, as it requests, pays and then gains access to the specific content it wants to pull, at the moment it pulls it. Sometimes it will be an AI Agent with a wallet + some limits given to it by an individual human buyer. Other times, the access request has been paid for by another mechanism (bulk, subscription, etc.). Regardless, consumption will have to be metered accurately for later reconciling and settlement by the publisher’s accounting department. To make this mass market imaginable, take the example of mobile phone contract or a residential data / tv access bundles on offer today; typically, a flat monthly fee comes with an amount of movies, call time, data volumes, roaming, etc. In the future, there will likely be an additional component there like a million LLM tokens.
In recent drop-ins with publishers, I made a point to ask about the challenges in preparing (technically and commercially) for the new economics of this third revenue pillar. Sometimes, the conversation starter was a quote from a speech made in Davos earlier this year: “We are in the midst of a rupture, not a transition.”
Publishers’ main concerns
As I sat down to write this, I saw notes of calls with 20+ publishers across Europe (Europe is defined generously, UK included) and the Middle East. One thing stood out as common thread: While each publisher has a unique set of assets on which it builds future cash flows, all are asking similar questions about the risks of making AI agents pay. It ranges from, “I’m blocking all non-human traffic already. How do I safely open the tap to those willing to pay?” All the way to, “We’ve only ever managed to generate revenue from print and online ads. We’ve never gotten humans to subscribe or micropay at scale, so why would AI agents pay for something we never get humans to pay for?”
Beyond the infosec tech (of detect, block, let in), a host of commercial questions surfaced: “How much will it cost me to enable machine payments?” The short answer is well below the revenue potential when priced properly. Cost+ is key here. “Who sets the prices and how do I build pricing models for machine discovery and yield optimisation? What parameters to go for, per article, per word count, per query?” The answer to the first question is short and simple. The publisher sets prices. A few are already using existing configurations to shape their own take on the pricing structures that can be deployed with today’s tech. Failing small and fast is the way to go. Begin testing and iterating to see what drives subsequent behaviour by which type of AI Agent and perceived AI Agent intent. And of course, this will drive editorial decisions down the road.
“How do I govern for rights, attribution and royalties’ distribution for content I’ve transacted with AI agents for a profit?” Publishers already a bit further down the road are defining new meta data fields for the content management systems in use, extending provenance capabilities designed for the purpose of agentic payments and attribution.
The most common question: “What are other publishers doing, who’s already making revenue and how will the search giant (and traffic director) of the web react when I begin to change how I grant access to my content?” From the line of visibility I have, more publishers are realizing this is not a topic to be promiscuous about. Hard thinking, decisions, and strict adherence is what will create desirable behaviour by readers and by the tech providers leading to monetisable value for the publisher (and for the industry as a whole) in the long run. We’re back to the first paragraph of this note: in this market, supply creates the norms under which it operates. Demand reacts.
Something is surely on the right side of history when a bipartisan bill for its advancement gets introduced in the 119th U.S. Congress. That happened on July 23rd. It’s encouraging to see publishers building the capabilities to make demand for machine-consumed content pay its way. A substantial share of the economic rent generated by the exchange of information needs to flow back to where costs are incurred. Plural societies are better served when their publishing markets sustain the livelihoods of thinkers, writers and creators.
About us
By lowering the barriers to entry for cloud computing and GenAI for organisations of every size, AWS has built one of the fastest-growing enterprise technology businesses in history. Customers across every industry use our infrastructure to accelerate innovation, transform their businesses, and shape their future. Publishers around the world trust AWS to deliver content to millions of eyes and ears, every second. We are guided by customer obsession, pace of innovation, commitment to operational excellence, and long-term thinking. Learn more at aws.amazon.com/media and follow our Media & Entertainment page on LinkedIn.
