A trial that gives everyone the treatment and reports how many improved cannot establish that the treatment did anything. The comparison group exists to answer a question the treated group cannot.
Conditions change without any intervention
Many symptoms fluctuate, and people tend to seek care when they feel worst, which means the following measurement is likely to be better whatever is done in between.
That drift towards the average is a property of measurement rather than of medicine, and it produces apparent improvement in any group observed after a bad period.
A control group experiences the same drift, so comparing the two separates the change caused by the treatment from the change that would have occurred anyway.
Expectation alters what is reported
People who believe they are being treated report outcomes differently, particularly for anything assessed by the person experiencing it, such as pain, fatigue or mood.
The effect is real in the sense that the reports change, but it attaches to the act of being treated rather than to the specific intervention under study.
An inactive comparator that looks and is administered like the real thing holds that expectation constant across both groups so it cancels out of the comparison.
Blinding protects the comparison from the observers
If investigators know who received what, their assessments can shift without any intention to mislead, especially where the outcome involves interpretation or judgement.
Concealing the assignment from participants and from those measuring outcomes is what the term blinding describes, and it is separate from the use of a placebo itself.
Trials record how often participants correctly guessed their allocation, because a treatment with distinctive effects can break the blind and reintroduce the problem.
Randomisation balances what nobody measured
Allocating participants by chance means the two groups are similar not only in recorded characteristics but in ones that were never thought of or could not be measured.
That is the property no amount of statistical adjustment afterwards can reproduce, because adjustment can only correct for variables that were actually collected.
The larger the trial, the more reliably chance produces balance, which is one reason small studies remain unconvincing even when their results look striking.
A placebo is not always the right comparator
Where an effective treatment already exists, withholding it to create an untreated group would be unacceptable, so the new option is compared against current practice instead.
Such trials answer a different question. They ask whether the new approach is better, or no worse, rather than whether it works at all.
Reading any result therefore requires knowing what the comparison was, because the same reported improvement means very different things against nothing and against an established alternative. Decisions about care belong with a qualified clinician.