The advance estimate lands about a month after the quarter ends and relies on assumptions for months the statisticians do not yet have data for, especially trade and inventories. The second and third estimates replace those assumptions with real numbers, and annual benchmark revisions later rewrite years of history.
The lesson for traders is that GDP is a slow, heavily revised summary of things already reported. Monthly nowcasting models that aggregate the same source data usually have a good estimate weeks before the advance print.
Example: the advance estimate is 2.8% annualised. The second estimate revises it to 2.5% on softer inventories and the third to 2.4%. Markets moved on the 2.8% and ignored both revisions.
Related: gdp, real-gdp, data-revision, nowcasting, annualised-rate