Whether congressional primaries are a source of polarization in the U.S. Congress remains contested. Central to claims that primaries polarize is the strategic positioning dilemma, which contends that candidates must satisfy ideologically extreme partisans in the primary before facing a comparatively moderate general electorate. We test whether candidates position themselves differently across primary and general elections using a unique dataset of positions over time. We apply an original scaling method based on candidates’ language on Twitter to estimate the positions of over 800 candidates in the 2020 primaries for the US House of Representatives and Senate over time. We show the validity of these positions by correlating them with NOMINATE (r > 0.9) where possible. We use a interrupted time series design on dynamic ideal points to show that the presence of a primary election caused artificial partisanship among Democrats. Democrats who did not become the party’s nominee moderated significantly following the primary, indicating strategic positioning for perceived electoral benefit during the nomination process. For Republicans, we observe no centrifugal effect of primaries, nor moderation among nominees or losing candidates after the nomination process.