Data synthesis has two common meanings. In research, it means combining findings from several studies or sources into one result, as in a systematic review or meta-analysis. In data engineering and AI, it means generating new data records that reproduce the statistical patterns, structure, and relationships of a source dataset without copying its original records.
The two senses share a goal: producing something usable that no single source provides on its own. For example, a team with only 40 recorded cases of a rare equipment fault may synthesize additional examples so a classifier sees enough of them during training. Synthesized data still needs checks. Teams compare it with the source on the task it will serve and confirm that it does not reproduce individual records. Whether it helps depends on that task, not on how realistic the records look.