The skill of proxy-based reconstructions of Northern hemisphere temperature is reassessed. Using a rigorous verification method, we show that previous estimates of skill exceeding 50% mainly reflect a sampling bias, and that more realistic values vary about 25%. The bias results from the strong trends in the instrumental period, together with the special partitioning into calibration and validation parts. This setting is characterized by very few degrees of freedom and leaves the regression susceptible to nonsense predictors. Basing the new estimates on 100 random resamplings of the instrumental period we avoid the problem of a priori different calibration and validation statistics and obtain robust estimates plus uncertainty. The low verification scores apply to an entire suite of multiproxy regression-based models, including the most recent variants. It is doubtful whether the estimated levels of verifiable predictive power are strong enough to resolve the current debate on the millennial climate.