What the Dummy Pill Does

That's the finding, stripped of its hedging. A meta-analysis in The Journal of Clinical Psychiatry pooled nine randomized controlled trials and looked at what happened to the people who got nothing: 495 of them across the placebo arms — 300 with alcohol dependence, 88 with cocaine, 107 with methamphetamine. Four weeks minimum of swallowing a capsule with no active ingredient in it. In the alcohol group, craving scores fell consistently. Not a cure. Not dramatic. But measurable, in people whose treatment consisted entirely of a ritual.

The stimulant groups drifted the same direction and didn't clear significance, which matters and which the authors say plainly. They also say — properly, scientifically — that the drop is a blend of true expectancy effect, the natural course of the condition, and regression toward the mean.

That's the honest objection, and it deserves better than a nod: maybe most of that movement is statistical weather. Fine. Concede all of it. There's a claim underneath that doesn't need the number to stand.

Every one of those nine trials was built around the assumption that ritual and attention do enough clinical work to threaten the result. That is what a placebo arm is for. Nobody spends money controlling for a variable they believe is inert. The apparatus conceded the point in its architecture — before any data came in, before regression to the mean got a vote. Whatever the effect size turns out to be, the design is a standing admission that being enrolled, scheduled, and asked about does something worth subtracting.

So the question isn't how big it is. It's what we do with something we've already admitted is active.

Because everyone in that placebo arm got something. Not the drug. They got a diagnosis somebody wrote down and took seriously. A daily ritual with a beginning and an end. A bottle on the counter that said, every morning, you are a person currently under treatment. And a researcher who would call and ask how the week went, and record the answer, because the answer mattered to someone other than them.

The trials aren't the villain here. Subtraction is the correct way to find out what a molecule does, and no one running a study thinks the remainder is worthless. The problem starts one layer over, where the money is.

There the rule is simple and almost never said out loud: a thing has to be separable before it can be paid for. A molecule can be isolated, named, coded, billed. Being checked on cannot — it arrives bundled with everything else in a human relationship, and the billing layer has no line for a bundle. So it rides along as packaging.

That's changing, barely, and the shape of the change proves the rule. Attention did finally get insurance codes — the collaborative care codes, 99492 through 99494, which reimburse a behavioral health care manager's follow-up time. But look at what had to happen first: it had to be converted into metered minutes. Seventy of them, then sixty, then thirty at a time. Attention became billable by becoming a unit.

And contingency management — giving people small incentives for clean tests, among the best-evidenced treatments we have for stimulant use disorder — spent years effectively capped around $75 per patient per year, because federal anti-kickback rules read giving a patient something for showing up as potential inducement rather than as treatment. SAMHSA lifted the ceiling for its grantees to $750 in January 2025. Medicaid coverage exists in five states, by waiver. The treatment had been sitting there, evidenced, for decades. What was missing was permission from the payment layer.

That's the mechanism, and it's duller and worse than malice. Nobody decided relationship doesn't matter. What can be isolated is what can be funded, and what gets funded is what comes to exist. The market's ontology quietly becomes care's ontology.

Which is why the next part should land harder than it does. The industry has now found a way to manufacture the packaging at zero marginal cost. An app that pings you daily, asks how the week went, remembers the answer, and tells you it's glad you checked in is — structurally, precisely — the placebo arm: the ritual, the schedule, the sense of being on somebody's list. Finally separable. Finally sellable. If that arm was doing real clinical work, this is either the largest expansion of it in the history of the field or the most efficient hollowing-out of it, and nobody honestly knows which yet. What it certainly is: the first version of being checked on that can be unbundled, priced, and scaled — which by the logic above means it is the version that will get built, whether or not it's the version that works.

We'd rather believe a molecule fixed us anyway. A molecule can be purchased, and it doesn't require anyone to show up. Say the improvement came from expectation and ritual and being asked about, and the culture hears fake — as if a change in someone's actual craving becomes less real once you learn what produced it. We reserve the word "real" for the part that works with nobody in the room. Strange thing to want, for a species that gets sick from isolation.

None of this argues against medication. The alcohol pharmacotherapies beat placebo; that is the entire point of running the trial. If you're on something, stay on it.

It argues against the accounting. A system that could actually pay for attention wouldn't look mystical, it would look tedious: the collaborative care codes funded and used instead of stranded on the fee schedule, the contingency management ceiling raised and the waivers extended past five states, peer recovery support billed as treatment rather than tolerated as charity. None of that requires believing anything exotic about the placebo effect. It requires the billing layer to admit what the trial design admitted years ago.

That work is slow, and it isn't yours to finish this week.

What is yours: you already know who's going through something and hasn't heard from you. That isn't the fix, and it shouldn't be dressed up as one — it's what's available while the fix takes a decade. But it was never the hard part of the protocol. It was never patented, it's still unprescribed, and it is apparently not entirely inert.