A systematic review identifies all studies that address a defined question; a meta-analysis goes further and combines their results into a single summary estimate. The gain is precision: each trial estimates the effect with its own sampling error, and averaging across trials narrows the confidence interval around the summary effect. Pooling can also detect a modest but real effect that no individual trial was large enough to establish on its own.
The gain is only real if the studies are similar enough that one summary effect means something. The studies must share a population, an intervention, a comparator, and an outcome that are close enough to be considered the same question. When they do not — for example, trials of the same drug at very different doses, or in populations with very different baseline risks — a single pooled number averages across genuinely different effects and describes none of them accurately. In that situation the honest output is a structured narrative summary of the separate results, not a single combined estimate.
A useful way to hold this: pooling is a claim that the trials are exchangeable estimates of one underlying effect. If that claim is false, the summary is precise but wrong.