Agencies fit a moving seasonal factor to several years of history. That works when seasonality is stable and fails when it is not, which is why the pandemic years left residual seasonality in series from initial-jobless-claims to cpi for years afterwards.
Two habits help. Compare the non-adjusted series to the same month in prior years to see whether the adjusted surprise is real. And be sceptical of January and of any month where the seasonal factor is large relative to the change being measured.
Example: unadjusted retail sales fall 18% from December to January. The seasonal factor expects a 19% fall, so the adjusted figure is reported as plus 1.0% growth. A one point error in the factor produces the entire headline.
Related: data-revision, annualised-rate, initial-jobless-claims, cpi, base-effects