I spent a while reading the actual literature on meditation rather than the summaries of it, and the picture is more mixed and more interesting than either side of the argument suggests.
The methodological problem
Start here, because it affects everything downstream.
You cannot blind a meditation trial. Participants know whether they are meditating.
Which means expectancy effects — improvement because you expect improvement — cannot be separated out the way they can in a drug trial.
The better studies use active control groups, where the comparison group does something equally plausible and equally time-consuming.
Studies comparing meditation against a waiting list consistently find larger effects than those comparing against an active control, and that gap is the size of the expectancy problem.
Where the evidence is reasonably strong
Structured eight-week programmes for anxiety and depression symptoms have a substantial body of trials behind them.
Effect sizes are generally moderate, comparable to other psychological interventions, and they hold up in reviews that restrict to higher-quality studies.
Relapse prevention in recurrent depression, using a specific structured programme, has the strongest evidence in the whole field, to the point that several national clinical guidelines recommend it.
That is a genuine, replicated, clinically meaningful finding and it is worth stating clearly.
Where it is weaker
Claims about attention, cognitive performance and memory have produced inconsistent results.
Some studies find improvements, others find none, and the field has a publication bias problem that has been documented directly.
Physical health outcomes — blood pressure, immune markers, inflammation — show small effects in some studies and are frequently reported with more confidence than the data supports.
The brain imaging question
Studies showing structural brain differences in long-term meditators are widely cited and frequently overinterpreted.
Most are cross-sectional, comparing meditators with non-meditators, which cannot establish that meditation caused the difference.
People who meditate for decades differ from people who do not in many ways, and self-selection is the obvious explanation for at least part of what is observed.
Longitudinal studies exist and are smaller, and their findings are more modest.
Adverse effects
Underreported for a long time and now receiving proper attention.
A minority of practitioners experience anxiety, dissociation, disturbing memories or worsened mood, particularly during intensive practice.
Research surveying practitioners has found rates that are not trivial.
Which does not make meditation dangerous and does mean the framing of it as universally beneficial with no downside is inaccurate.
Intensive retreat practice appears higher risk than short daily practice, and people with trauma histories are more likely to be affected.
What is being measured
Worth noting because it constrains interpretation.
Most outcome measures are self-report questionnaires, completed by people who know which group they are in and have invested effort.
Which is not worthless and is a softer form of evidence than objective measures.
The studies using physiological or behavioural outcomes generally find smaller effects than the questionnaire studies do.
What I take from it
For anxiety, low mood and stress, there is reasonable evidence that structured practice helps a meaningful proportion of people.
For most other claims, the evidence does not currently support the confidence with which they are stated.
Which is not a reason not to meditate. It is a reason to have accurate expectations, since inflated expectations produce disappointment and abandonment.
The obvious caveat
I am not a clinician and this is a summary of published research rather than clinical guidance.
Anyone experiencing significant anxiety or depression should speak to a doctor, since there are treatments with stronger evidence and a practice that helps some people is not a substitute for assessment.
The app studies
A distinct literature worth separating from the research on structured programmes.
Trials of meditation applications have generally found smaller effects than trials of in-person eight-week courses.
Which is unsurprising, given that the app removes the group, the teacher, the commitment and the structured curriculum, all of which are plausible active ingredients.
Attrition in app trials is also very high — a large proportion of participants stop using them within weeks — which affects both the results and their real-world relevance.
Several app studies have been conducted or funded by the companies involved, which is disclosed and worth noting.
Duration and frequency
What the research says about how much practice is needed is thinner than people expect.
Structured programmes typically prescribe substantial daily practice, and studies rarely test shorter durations against longer ones directly.
Which means the standard recommendations reflect programme design rather than an established dose-response relationship.
The studies that have examined practice time have generally found weak correlations between minutes practised and outcomes, which is either evidence that less is sufficient or evidence that self-reported practice time is unreliable.
Both interpretations are plausible and the field has not settled it.
Where to be sceptical of a claim
A quick filter I use on anything in this area.
If the study compared against a waiting list, expect the effect to shrink against an active control.
If it measured brain structure cross-sectionally, it cannot establish causation.
And if the sample was people already committed to the practice, self-selection explains a great deal.