The dark side of AI monetization

Ever since OpenAI opened ChatGPT to the public in November 2022, the AI world has served as an information-discovery playground. Dozens of companies followed OpenAI with services based on large language models (LLMs), ranging from chatbots to search engines and everything in between.

Now, these companies’ products — including OpenAI ChatGPT, Anthropic Claude, xAI Grok, Inflection AI Pi, Mistral AI Le Chat, Cohere Coral/Command-R, Adept ACT-1, Glean Enterprise Search Copilot, Character.ai Character, and Perplexity AI Perplexity Search — have attracted hundreds of millions of users.

Behind the scenes, these LLM-based, generative AI tools have sucked up tens of billions of dollars in funding and investment. And the investors want their pound of flesh.

Most users don’t pay anything. A minority pay subscriptions for premium service. But those subscriptions don’t even come close to paying for the service.

So the AI companies have to monetize. But how?

The Truman Show model

In the 1998 movie, The Truman Show — starring Jim Carrey as a man named Truman Burbank who is unwittingly part of a reality show in an artificial world where everyone else is an actor playing the role of a stranger, friend, or his wife — things start to get weird halfway through.

When Truman walks into the kitchen and starts asking his wife, Meryl, uncomfortable questions, she pivots, holds up a box of cocoa, and says with a dazzling TV smile: “Why don’t you let me fix you some of this new Mococoa™ drink? All-natural cocoa beans from the upper slopes of Mount Nicaragua — no artificial sweeteners!”

Two AI leaders are proposing such sponsored conversations with AI chatbots.

On Wednesday in a live broadcast on X, owner Elon Musk told advertisers that brands will soon be able to pay for spots that appear inside answers from Grok, the AI chatbot built by his startup xAI. He said this move will help cover the high GPU costs behind Grok and follows the July exit of chief executive Linda Yaccarino, who had previously managed advertiser ties.

Amazon is thinking along the same lines. In late July, Amazon chief executive Andy Jassy told investors on the company’s second-quarter earnings call that Amazon plans to place paid ads inside answers from Alexa+, the AI version of its voice assistant, so the ads emerge in the middle of user conversations and point listeners straight to products sold on Amazon.

Presumably, these sponsored placements would be unlabeled, and the users would not know their results were bought and paid for.

After all, why would they? Other types of monetization take place on some chatbots without user notification.

The payola model

In the 1950s, payola referred to the illegal and undisclosed practice of record companies paying radio stations, DJs, or their employees money or providing gifts or favors in exchange for playing specific songs on the radio. This ensured a song’s airplay not because of its artistic merit or popularity, but due to financial incentives. Audiences were allowed to believe that the records were chosen for their merit.

Some AI chatbots work this way.

For example, OpenAI and Perplexity AI both run paid prioritization programs for content partners, which means certain publishers get their content prioritized and their brands more visible in AI-generated answers.

Perplexity AI has formed partnerships with publishers like Le Monde, The Independent, The Los Angeles Times, Time, Fortune, and Der Spiegel to supply content directly used for AI answers. These partners are not just scraped for data but receive a share of advertising revenue generated by Perplexity. Their brand names and sometimes logos appear prominently whenever their content is cited in responses, driving user traffic to their sites.

OpenAI has a similar program under the name “Preferred Publishers Program.” Since 2023, OpenAI has secured deals with more than 30 media companies including The Wall Street Journal, The Atlantic, the Associated Press, Vox, Condé Nast, News Corp (owner of The Wall Street Journal and New York Post), and Axel Springer. These agreements provide OpenAI legal access to high-quality, trusted content for training and live AI responses. Participating publishers receive financial compensation, priority content placement, and prominently highlighted links when ChatGPT uses their content to answer prompts.

On the one hand, it’s nice that the AI companies are paying for content used for answers, at least in the minority of instances where they’ve established these partnerships.

On the other hand, it’s also true that these revenue-sharing and content-sharing deals are changing the outcome of results for users, with the partners’ content, links, and branding giving favorable treatment — and this arrangement is not explicitly disclosed to users, who are allowed to believe the chatbot answers are based on merit alone.

The affiliate link marketing model

OpenAI recently announced plans to monetize product purchases made through ChatGPT by integrating an in-chat checkout experience and earning commissions from sales. The new system will allow users to discover, select, and buy products directly within ChatGPT without leaving the chat interface, with merchants paying OpenAI a small percentage as a commission for each transaction. CEO Sam Altman reportedly mentioned a potential 2% affiliate fee on sales generated this way.

While it’s unknown how this policy will evolve over time, it clearly provides a financial incentive for OpenAI to promote products it earns a fee on.

The shrinkflation model

Shrinkflation is a practice where companies reduce the size, quantity, or weight of a product while keeping its retail price the same. This means consumers are effectively paying more per unit of the product without an obvious price increase. It allows manufacturers to manage rising production costs without raising the sticker price, often escaping immediate consumer notice.

There is credible user-based anecdotal evidence pointing to some decline in performance or quality, especially among free or lower-tier access models, as companies may be optimizing costs. This can mean simpler, faster, and less resource-intensive responses at the expense of depth or advanced capabilities for free-tier users.

As the public is generally dazzled by the rapidly growing capabilities of LLM-based genAI chatbots and search, the quality of the free versions appears to be increasingly lagging behind the paid versions.

The charge for everything model

The major AI chatbots monetize via subscriptions, API usage fees, custom AI solutions, affiliate marketing, advertising revenue sharing, digital content licensing, and consulting services.

And it’s still not enough. They’ll have to come up with even more ways to monetize. I worry about what comes next.

The truth is that AI is expensive. It costs a fortune to crunch those numbers and perform those calculations to answer simple chatbot questions.

And there’s no such thing as a free crunch.