Standard demand theory treats price as a continuous variable. Lower the price and quantity demanded rises along a smooth curve, with zero sitting at the end as just another low number. It doesn't behave that way. Zero is a discontinuity. In the experiment most often cited on this, Shampanier, Mazar, and Ariely ran a candy stand at MIT and priced a chocolate at one cent on some days and at nothing on others. At one cent, 58 people stopped. At zero, 207 did. The jump is far larger than a one-cent price change should produce, which is the first sign that a free price gets processed differently from a cheap one. The mechanism is about evaluation, not money. A price, even a trivial one, forces a cost-benefit calculation and introduces the possibility of a bad trade. Zero removes the downside, so the option gets assessed on its benefit alone. This is why people will take a free lower-value good over a discounted higher-value one. In the same research, subjects offered a free Hershey's against a Lindt truffle priced at a few cents flipped toward the Hershey's once its price hit zero, even though the truffle was the better product at a still-negligible cost. The free option carries no perceived risk, and the absence of risk is doing the work. That property is exploitable, and the architecture for exploiting it is well understood: -- Marginal cost near zero. A digital product can serve one more user at almost no incremental cost, which makes a zero price sustainable in a way physical goods never permit. -- The payment relocates rather than disappears. Free tiers recover revenue through advertising, data collection, or the conversion of a small fraction of users to paid plans. Facebook's base product is free because the captive audience is the thing being sold. -- Zero suppresses scrutiny. A price prompts the user to ask whether something is worth it. Removing the price removes the question, so the quality and motive checks a paid product would face often go unmade. -- Free assembles the base that gets monetized later. Usage caps and feature gates convert a minority to paid while the majority subsidize reach through network effects and the data they generate. The current instance of this is the free AI tier. OpenAI began testing advertising inside ChatGPT's free and lower-cost tiers in early 2026, while Perplexity removed advertising on the premise that a paid-for answer erodes the trust the product depends on. The two companies are running opposite bets on the same question: whether an answer engine can carry ads without degrading the thing that makes it valuable. Look at how the ad-supported version is built and the relocated payment becomes visible. OpenAI's stated design keeps advertisements in labeled boxes separated from the answer, with the claim that they do not influence the response. But the targeting draws on conversation topics, chat history, and prior ad interactions, which means the payment is being collected in data even where it is not collected in money. The user funds the free tier twice: once in attention to the ad, and once in the behavioral record that makes the ad sellable. Here the suppression effect matters more than it does with chocolate. An answer engine's value rests on perceived neutrality, and the moment a revenue incentive attaches to the output, there is pressure on what gets surfaced, recommended, or framed as the answer. The zero price has already closed the inquiry into who is paying and why, so the scrutiny a user would apply to a sponsored search result or a paid placement never engages. The problem compounds as people offload more judgment onto these systems. You are not choosing a worse free candy. You are outsourcing a piece of cognition to a system whose incentives you have been conditioned not to examine. The ethical issue is not that free AI is bad. It is that "free" disables exactly the scrutiny a tool of this consequence should invite. So then, it is crucial to approach AI through that lens. If something is free, I am the product, and must acknowledge that fact when interacting with it.