If Chatbots Replace Search, Who Pays for the News?
AI is the future of information, journalism is its infrastructure
Thomas Germain, on the BBC, recently bragged that he’s hacked ChatGPT and other generative AI models into thinking that he’s an amazing hot dog eater. I can’t say how good or bad he is at eating hot dogs, though he admits to lying about the topic.
And just like a person who eats too many hot dogs, an AI can get “sick” by ingesting too much bad content. The obvious one is that anyone looking to manipulate results can feed the AI junk food and get it to say whatever.
People have studied this and gained a strong understanding of how to influence the bots. So the issue here for the AI companies is much more existential. In fact, it’s such a central concern that tech leaders like Alexis Ohanian and OpenAI CEO Sam Altman talk openly about “dead internet theory.”
The vulnerability that Germain lays out above is one of trust. That is, can they trust that the content they’re getting is truthful and accurate? AI companies need a way to ingest the right content that they know has been produced by humans.
For that, the AI companies need to purchase content, in real time, from trusted news organizations through an open AI Content Marketplace.
Today AI companies are happy to scrape the web and grab what they can, a concept that has its own ethical issues around content ownership, but also creates ethical issues within the models themselves. Generative AI is, in many ways, a reflection of who we are as indicated through our language, and the biases we have built into our own writing come out in the results we see. (more on this another time).
The web scraping process creates additional problems for the BBC and other news publishers. Today, subscription revenue keeps brands like News Corp and the New York Times company alive, while a robust search program works to send new subscribers in through Google links.
Now, with AI, people are getting their answers directly from AI models instead of going through Google, threatening this vital customer acquisition tool, which means Google Zero is on the horizon, a time in which Google sends little to publishers because people get most of their answers right from the chatbots. Referrals started dropping fast just as Google rolled out its AI chatbot. Cloudflare CEO Matthew Prince sounded the alarm at the Council on Foreign Relations.
“AI is going to fundamentally change the business model of the web. The business model of the web for the last 15 years has been search. One way or another, search drives everything that happens online,” he said. “Today, 75 percent of the queries that get put into Google get answered without you leaving Google, get answered on that page.”
The AI Content Marketplace could be a part of that new business model by encouraging the LLMs to pay news organizations for content in real time. This link isn’t just critical for news publishers, but also for the AI companies themselves.
AI experts fear model collapse, that is, when AI is eating its own content and the model itself becomes useless. This technology needs ongoing human-created information, and that’s just what news publishers have.
AI companies need this content not just to train their systems but to ensure answers come from reliable, up-to-date sources. In this context, journalism becomes a trusted Verified Content Layer behind the AI chatbots. Imagine a user wanting to know more about an upcoming local election and using AI to summarize the issues. The chatbot needs truthful, trusted information in it to help keep that voter informed. Otherwise, anyone can claim to be a hot dog eating king and AI will just believe it.
The AI content marketplace is already emerging, but it’s not yet reached critical mass. Reddit made a deal to sell training data to Google for $60 million a year, a deal it says is worth more than $200 million over the next several years. The New York Times completed an AI deal with Amazon while companies like the Washington Post and Conde Nast signed on with OpenAI.
These partnerships work. A study from SearchEngineLand and Fractl found that the content provided directly to AI shapes how the model understands information. “When a publisher network becomes a model partner, it stops being just another website the AI can read – it becomes a trusted source the model actively learns from and reuses.”
There are startups coming onto the scene like TollBit and ProRata.ai, but both are very early-stage. TollBit recently raised a $24 million Series A round, while ProRata closed a $40 million Series B. Microsoft is also getting into the game, rolling out a pilot of what it’s calling the Publishers Content Marketplace (PCM).
The one-off deals or exclusive arrangements happening today come with the danger of inserting bias or limiting what information the AI models have.
The digital revolution, and the social media dominance that quickly followed, left media companies searching for new business models and new ways to distribute information. Gone were the captive audiences who received a printed newspaper on their doorstep, replaced instead with fickle users who clicked on one article at a time and only wanted to pay for what they read. That is, publishers lost control of the syndication model even as they maintained the cost of production.
Now, with an AI Content Marketplace, they can start building it back.