What does peer review actually do?
Peer review is the process by which a journal asks other researchers in a field to assess a submitted paper before publication — and understanding what it does and does not check is essential to reading science reporting sensibly.
The process. An author submits to a journal. An editor screens it for scope and basic quality, and either rejects it outright — a large proportion are rejected at this stage — or sends it to two or three reviewers with relevant expertise. Reviewers, working unpaid, produce a report recommending acceptance, revision or rejection. Most accepted papers go through at least one round of revision. The editor decides.
What peer review checks:
Whether the methods are appropriate for the question asked.
Whether the conclusions follow from the results presented.
Whether the work engages with existing literature and whether the claim is novel.
Whether the statistics are defensible and the reporting complete.
What peer review does not do, and this is the crucial part:
It does not verify that the data are real. Reviewers see what the authors report. Fabrication is generally caught later, if at all.
It does not replicate the experiment. Nobody repeats the work.
It does not guarantee the finding is true. Plenty of peer-reviewed findings have failed to replicate, which is the substance of the replication crisis.
It is not consistent. Reviewer agreement on the same paper is frequently poor.
Variants: single-blind (reviewers anonymous), double-blind (both), open (neither, reports published), and post-publication review.
Preprints — papers posted publicly before review — accelerated dramatically during the pandemic. They are not worthless, but a preprint has passed no filter at all.
Predatory journals publish for a fee with token or no review, which is why the mere phrase "peer-reviewed" is not sufficient.