Was it the pill?

Placebo, nocebo and how to tell what a new medicine is really doing to you. A guide for anyone about to start one, for the clinicians who prescribe them, and for understanding why trials are built the way they are.

By Reid Robison, MD. About 30 minutes, with seven things to try, including a worksheet to use the next time you start a medicine.

Here is a conversation I have most weeks. Someone started a new medicine ten days ago. They've had a headache, felt tired, slept badly and been a little queasy, and they've stopped taking it. They've concluded the medicine "didn't agree with them," and they're wary of the next one. Sometimes the medicine really was the cause. Often it wasn't, and the thing that stopped a treatment that might have helped was the expectation of harm, plus a bad week, plus a list of side effects read the night before.

This isn't about blaming anyone's imagination. The placebo and nocebo effects are real, measurable and partly biological; they happen to doctors too. The point of this post is to give you a way of seeing through them: what a new pill is doing, what your expectations are doing, and what the ordinary noise of being a person is doing, so you can keep the medicines that help and drop the ones that don't for the right reasons. The second half explains why clinical trials have placebo arms at all, and what happens to that logic when the drug is a psychedelic and everyone in the room knows who got it.

If you want the research-side companion, with the anatomy of a placebo arm, the rising placebo response and the SAMSON statin experiment, read Sugar & Shadow. This post is the version for the person holding the pill.

Part 1What a placebo effect is, and isn't

Start with two words that get used interchangeably and shouldn't be. The placebo response is everything that happens to people in the placebo arm of a trial: they got better by this much. The placebo effect is the part of that caused by the treatment context itself: expecting to improve, the ritual of taking something, the relationship with the person who gave it to you.¹¹ The rest of the placebo response is made of things that would have happened with no pill at all.

Natural course

Depression, anxiety, pain and insomnia wax and wane. People tend to seek help, and join trials, when things are bad, which means the next few weeks are more likely to be better than worse whatever happens.

Regression to the mean

A statistical version of the same thing. Anyone selected for having an extreme score will, on average, score closer to normal the next time they're measured, purely because the first measurement caught some bad luck.

Being measured

Filling in a depression scale every week with someone who listens is itself a small treatment. In a trial of irritable bowel syndrome, a warm, attentive clinician added more improvement than the sham treatment alone.¹²

Expectation

The true placebo effect. In pain, it releases the brain's own opioids, and the opioid-blocker naloxone can switch it off.^13,14 In Parkinson's disease it releases dopamine. In depression and anxiety it's harder to pin to one chemical, but imaging shows the same frontal expectation circuits at work.¹⁵

Conditioning

If a capsule has relieved your pain ten times, the eleventh one relieves it before the drug is absorbed. Your body learned the association. Conditioning can even be measured in immune responses.¹⁶

How big is the true effect? Smaller than the legend. When the Cochrane group compared placebo arms against no-treatment arms across all conditions, placebo had no clear effect on most objective outcomes and modest effects on subjective ones, chiefly pain and nausea.¹⁷ Two things follow. The placebo response in a trial is mostly not the sugar pill, and the part that is the sugar pill is concentrated exactly where psychiatry lives: in how people feel and what they report.

One more twist. Placebo effects don't even require deception. People with irritable bowel syndrome who were handed pills clearly labelled "placebo" and told, honestly, that placebos often help, improved more than people given nothing,²⁰ and a meta-analysis of such open-label placebo trials found a moderate effect across conditions.²¹ Knowing the mechanism doesn't switch it off.

That gap between open and hidden is the placebo effect, measured without any placebo. It's also the best argument that expectation is not a nuisance to be subtracted but part of how medicines work. The practical consequence for a patient: the same pill does more when you know what it's for and believe it can help. And, as the next section shows, it does more harm when you expect harm.

Part 2Nocebo: the one that stops treatment

Nocebo is placebo's shadow: symptoms caused by expecting them. It's the reason this post exists, because placebo rarely does damage, while nocebo ends treatments every day.

Look at the placebo arm of any drug trial and you'll find side effects. In pooled placebo-controlled trials of escitalopram, 7% of people on placebo reported nausea, 4% insomnia, 5% dry mouth, and 2% stopped the placebo because of side effects.²² Across antidepressant trials, people on placebo report the side effects characteristic of the drug class they think they might be on: placebo arms of tricyclic trials report more dry mouth and drowsiness than placebo arms of SSRI trials.³ Nobody is making anything up. Bodies produce headaches, fatigue, queasiness and poor sleep all the time; a new pill gives them a name and a cause.²³

The cleanest demonstration is the SAMSON trial. Sixty people who had quit statins because of side effects took, in random order, a statin for four months, an identical placebo for four, and nothing for four, scoring their symptoms daily. Symptom intensity was 8 out of 100 with no tablet, 15.4 on placebo and 16.3 on the statin. Ninety percent of the extra symptom burden from the statin appeared with the placebo. Half of them restarted statins after seeing their own data.⁴

The gluten story is the same shape. Many people who feel better off gluten and don't have celiac disease react to the expectation of gluten as much as to the protein: in a blinded crossover, people with self-reported gluten sensitivity got worse on every diet, including the gluten-free one.²⁴ I wrote about that in Gluten is (usually) not the enemy. Switching from a branded pill to a generic one that looks different brings a wave of new side effects and a dip in benefit, with no change in the molecule.²⁵ And in people who stop medicines early for side effects, the symptoms that drove them out are, over and over, the ones a sugar pill produces: headache, fatigue, dizziness, nausea, trouble sleeping, feeling "off."

Part 3Words that cause side effects

Nocebo is unusually easy to switch on with language, which means it's unusually easy to switch off.

None of these studies withheld important information. The finasteride men who weren't told about sexual side effects were still consented, and the women getting epidurals were still warned there'd be an injection. What changed was emphasis and vividness. Expectation follows the most vivid thing you were told, and it attaches to whatever your body does next.

Framing works in the helpful direction too. Children in a peanut immunotherapy trial whose families were taught that mild symptoms (an itchy mouth, a stomach ache) can be signs the treatment is doing its job reported fewer symptoms and less anxiety, and had better immune markers at the end.⁷ Adults starting methotrexate who watched a seven-minute video making the same point found their symptoms less burdensome, and at twelve weeks 8% had stopped the drug compared with 35% of those who got standard information.⁸ Nothing about the medicine changed. The story around it did.

Part 4Two rooms: how a trial works

Everything above explains why we can't judge a medicine by watching people take it. People who take a medicine improve for many reasons, and they get side effects for many reasons. A randomised placebo-controlled trial is the one tool that separates them. Here's the whole logic in one picture.

Three things to notice. First, the placebo room improves. Always. That's natural course, regression to the mean, attention and expectation, all together. Second, the drug room gets all of that too, because the rooms are identical except for the molecule. So the drug's real effect is the gap between the rooms, not the improvement in the drug room. A medicine that produces a lot of improvement can still be useless if the placebo room improved just as much. Third, if you unblind the rooms, the gap changes for reasons that have nothing to do with pharmacology: the drug room expects more, the placebo room expects less, and the raters nudge their scores. That is why blinding matters, and why losing it is such a problem later in this post.

Two more facts about the rooms. Placebo-room improvement in antidepressant trials has roughly doubled since the 1980s, which is partly why so many trials of effective drugs "fail": the bar moved.^2,26 And the gap for antidepressants, averaged over twenty-one drugs and more than 500 trials, is real but modest, larger in severe depression than in mild.^27,28 That's not an argument against the medicines. It's an argument for being clear about what each room contributes, and for expecting a lot of what a patient feels after starting one to be the placebo room's share.

Part 5Reading a side-effect list

The package insert and the pharmacy printout list every symptom reported more often on drug than placebo, and usually only the drug column makes it onto the page. The placebo column is the one that tells you whether a symptom is likely to be the pill. The tool below uses the escitalopram label because it's typical; most modern labels have the same two columns somewhere in section 6.

A rule of thumb that falls out of the table: when the two columns are close (dry mouth, dizziness, diarrhoea), a symptom you notice in week one is more likely background than drug, and worth waiting out. When they're far apart (ejaculation problems, nausea), the drug is the likelier cause, and the question becomes whether it fades. Most early antidepressant side effects fade within two to three weeks; sexual side effects usually don't, which is why they deserve a plan rather than patience.

Part 6Was it the pill? A worksheet

Clinicians have a formal version of this called the Naranjo scale, a ten-question score that estimates how likely a symptom is to be a drug reaction.²⁹ Here is a plain-language version. It stores nothing and sends nothing; it's for thinking.

The two most useful questions are the two people skip. Did you have this before? Almost everyone has some headache, fatigue, poor sleep or low-grade nausea in an ordinary fortnight, and nobody tracks it until a new pill gives it meaning. What happens if you stop and restart? A symptom that goes when the drug stops and returns when it restarts is the strongest evidence you can get outside a trial. Many "intolerances" don't survive a rechallenge, and many real reactions are confirmed by one.

Part 7Take your baseline first

The single most protective thing you can do before starting any medicine takes two minutes: write down how you already feel. Then, when something shows up in week one, you can check whether it's new.

A few more habits that help you see clearly:

• Decide the trial length in advance, with your prescriber. For most antidepressants it's four to six weeks at a working dose; the first two weeks are the noisiest and least informative. Stopping on day nine because of a headache is a decision made in the noise.
• Read the side-effect list once, with the placebo column, then put it away. Rereading it daily is a nocebo machine. If the leaflet lists only the drug column, ask the pharmacist or look up section 6.1 of the prescribing information.
• Change one thing at a time. Starting a medicine the same week you start a diet, stop coffee and change your sleep makes attribution impossible.
• Rate the symptom, not the verdict. "Headache 4/10 at 3 pm" is data. "The pill is giving me headaches" is a conclusion that will shape what you notice tomorrow.
• Know the ones that shouldn't wait. Rash with fever or mouth sores, allergic swelling, chest pain, fainting, new or worsening thoughts of suicide, severe agitation, confusion, or a high fever with stiffness and sweating: these are calls today, not experiments.

Part 8For clinicians: words that help

An expert consensus on placebo and nocebo in practice boils down to a few habits, all of them evidence-based and none of them deceptive.^30,31

• Lead with tolerability, then specifics. "Most people take this without trouble. A minority get nausea in the first week or two, which usually passes," is accurate and produces fewer nocebo symptoms than a list read aloud.
• Frame early symptoms as what they often are. Transient side effects frequently mean the drug is reaching its target. Said in advance, that framing cuts symptom burden and discontinuation.^7,8
• Offer contextualised consent. Ask what level of detail the person wants about mild, self-limiting effects, while always disclosing the serious and the irreversible. Ethicists have argued this is more respectful of autonomy than a recited list, not less.³²
• Set the expectation of benefit honestly. Expectation is part of the medicine's effect; a neutral shrug when prescribing an antidepressant costs the patient some of its efficacy.^26,33
• Baseline before, rechallenge after. Get the two-week symptom baseline at the prescribing visit. Treat a stopped medicine as a hypothesis, and when it's safe, test it with a rechallenge.
• Keep the ritual stable. Avoid unnecessary switches between manufacturers and appearances, or warn about them in advance; a differently coloured pill is a nocebo trigger.²⁵
• Record what the patient said. In trials and in clinics, the verbatim matters more than the code. The life of an adverse event follows what happens to those words.

Part 9The psychedelic placebo problem

Now take the two rooms and put a psychedelic in one of them. Within an hour, almost everyone in both rooms knows which room they're in. The participant knows. The therapists sitting with them know. Often the rater who scores their depression next week can tell from the conversation. The blind, which is what made the gap between the rooms meaningful, is gone.

This matters for both directions of the effect. Participants who realise they got placebo may feel let down, and their placebo-room improvement shrinks; participants who realise they got the drug bring every hope they arrived with. In a review of the problem, Muthukumaraswamy and colleagues showed how this can inflate the apparent gap even when the molecule does nothing extra, and noted that most trials hadn't even measured it.⁹ A 2026 trial that gave healthy volunteers psilocybin, MDMA, methylphenidate or placebo and then asked everyone to guess found that participants and investigators identified the psychedelic conditions far above chance.³⁴ A microdosing study made the point from the other side: participants who believed they were on the active dose improved more, whether or not they were.³⁵

There is encouraging evidence on the other side of the ledger. In the trial comparing psilocybin with escitalopram, people expected far more of psilocybin, but expectancy predicted outcome only in the escitalopram arm; in the psilocybin arm it didn't.³⁵ And the dose-response pattern in the lysergide phase 2b trial I worked on, where 100 micrograms beat 25 and 50, is harder to explain by unblinding alone, since all three doses were noticeable.³⁶ Neither settles the question. They are reasons to keep designing better trials rather than to give up on the question.

The toolkit the field has converged on, and which the FDA's 2026 guidance now asks for, looks like this:^10,37

Active comparators

A placebo that does something: niacin (which causes flushing), methylphenidate, a very low dose of the psychedelic itself. Hopkins' first modern psilocybin study used methylphenidate so that both rooms felt a drug.³⁸ Low-dose arms in dose-finding trials serve the same purpose.

Independent raters

The person scoring depression should not be the person who sat in the dosing room, should not know what happened in it, and ideally is off-site or central, scoring from a structured interview.

Ask everyone to guess

Participants, therapists and raters all record which arm they think the person was in, and how confident they are. The blind is then a measured quantity, not an assumption.

Measure expectancy

Before dosing, record how much benefit each participant expects. Then the analysis can test whether the drug's effect survives adjustment for it.

Standardise the context

Same room, same music, same preparation, same amount of therapist time in every arm, so that the only difference is the capsule.

Long follow-up and objective anchors

Effects that last months after a single dose are harder to produce by expectancy alone; functional and clinician-independent outcomes help too.

Part 10What we do about it at ISR

At Inner Space Research we run industry-sponsored trials of psychedelic and conventional psychiatric medicines in Utah. Sponsors design the studies; a site's job is to run them so that the two rooms stay as identical as the protocol intends. In practice, the placebo problem shapes our days in a few specific ways.

• Separation of roles. The team in the dosing room is not the team that rates outcomes. Rating is done by trained, calibrated raters working from structured interviews, and in many of our studies a central or independent rater scores from a recording, blind to everything but the interview.^39,40
• Neutral language, from the first phone call. Recruitment, consent and preparation describe the study honestly without promising transformation. Expectation is part of what a trial measures, so we try not to add to it, and we record it before dosing when the protocol asks.
• The same session for everyone. Preparation, the dosing-day setting, the music, the time with facilitators and the integration visits are identical across arms, so that a person on placebo gets the same care as a person on drug.
• Blinding checks. Where protocols include them, participants, facilitators and raters record their guess about assignment, and we treat a protocol that doesn't as one that should.
• Careful adverse-event recording. We write down what the participant said before anyone codes it, and we ask at follow-up rather than waiting to be told, because a sugar pill produces headaches and nausea too, and the only way to know what the drug adds is to count both rooms the same way.
• Honesty with participants about the design. People join psychedelic trials hoping for the active drug. We tell them plainly that the placebo room is where the knowledge comes from, that the care is the same in both rooms, and that many protocols offer the active drug afterwards.

None of this makes the problem disappear. A trial can be run perfectly and still be unblinded by a drug that announces itself. What it does is make the size of the problem measurable, which is the only way the field will find out how much of what psychedelics do is the molecule. If you want the longer argument about that question, it's in How much of the change is the molecule?

This post is educational and isn't medical advice. Don't stop or start a prescribed medicine on the strength of a worksheet; use it to have a better conversation with the person who prescribed it. If you're struggling, in the US you can call or text 988 any time.