In a recent NPR report, Stephan Bisaha asks a simple question: are Americans actually paying for AI? Most use it, he writes, but few pay.
Bisaha opens with Kirby Plessas, a self-described technophile who pays about $40 a month for both ChatGPT and Gemini, and expects to add Claude within a year. She is the exception, not the rule, Bisaha notes.
Bisaha reports that only about 3 percent of U.S. households were paying for AI in February, citing the Bank of America Institute, which tracks consumer transactions. The number is small but reportedly climbing, with roughly 10 percent more households paying than a year earlier. In the story, Sekoul Krastev of the Decision Lab compares it to the slow early days of streaming, before subscriptions took off.
Subscriptions get you more
Most Americans already lean on the free tools. Bisaha cites a March Quinnipiac poll finding that 51 percent use AI to look things up.
The free versions have limits, though. Bisaha notes that OpenAI caps its default model at 10 messages every five hours before dropping users to a weaker one. Pam Dean, who pays for both ChatGPT and Claude, recalls hitting a “your time has run out” notice in the middle of a task.
Paying buys smarter models and more questions. A $20 ChatGPT Plus plan allows up to 1,280 messages a day, Bisaha reports, plus features like custom chatbots. He describes how Jim Arnold built “Francisco,” a ChatGPT Spanish tutor tuned to speak slowly, to prepare for a possible move to Mexico.
Today’s ChatGPT stays free
OpenAI has about 50 million subscribers, Bisaha reports, and Nick Turley, who heads ChatGPT, thinks that could quadruple. Even so, Turley told NPR that paying users will always be a minority, and that keeping a free version is part of the company’s mission.
The first subscriptions, Turley explains, were less about money than about keeping the site usable when computing costs ran high. “We weren’t even trying to make money,” he told NPR, though he admits the revenue became a welcome side effect.
The next version may not be
Turley expects a free ChatGPT to stick around, but he will not promise the same for its successor. Bisaha explains that the successor is likely to be an agentic AI, one that can act on its own, say, planning a trip or doing your shopping.
Turley imagines a personal-assistant version so capable, and so costly to run, that it may stay behind a paywall.
Could AI companies use ads?
Bisaha explains that Ads are another way to cover costs, as they do for sites like Facebook. But Jeff Hancock of Stanford’s Tech and Impact Policy Center warns against it bluntly: “People hate it.” Ads reward endless scrolling, he notes, which has fueled worries about attention and mental health. A good AI assistant works in the background and keeps users off the screen, Hancock argues, a poor fit for an ad business.
Some users are already wary. Bisaha describes Sarah Womer, who rotates several AI subscriptions but turns to a privacy-focused tool, Kagi, for product research, worried that OpenAI’s ad tests could tilt answers toward advertisers. OpenAI has said its ads will be clearly labeled and will not influence ChatGPT’s responses.
Or bundling
With the model still unsettled, Bisaha lays out other ideas. At a March BlackRock summit, OpenAI’s Sam Altman floated charging for AI like electricity, by how much you use.
Greg Portell of the Kearney consulting firm told NPR that bundling instead, folding AI into subscriptions people already hold, like Amazon Prime or internet service, since Americans are tired of signing up for new ones. He even wonders whether employers might one day cover personal AI access as a perk.
Acknowledgement: This summary was generated using AI writing tools as part of my AI Explorations process. For more, see the original article linked below.