Context. The fate of stars largely depends on the amount of mass lost during the end stages of evolution. For single stars with an initial mass between ∼8–30 M⊙, most mass is lost during the red supergiant (RSG) phase, when stellar winds deplete the H-rich envelope. However, the RSG mass-loss rate (Ṁ) is poorly understood theoretically, and so stellar evolution models rely on empirically derived mass-loss rate prescriptions. However, it has been shown that these empirical relations differ largely, with differences up to 2 orders of magnitude. Aims. We aim to derive a new mass-loss rate prescription for RSGs that is not afflicted with some uncertainties inherent in preceding studies. Methods. We have observed CO rotational line emission towards a sample of RSGs in the open cluster RSGC1 that all are of a similar initial mass. The ALMA CO(2–1) line detections allowed us to retrieve the gas mass-loss rates (ṀCO). In contrast to mass-loss rates derived from the analysis of dust spectral features (ṀSED), the data allowed us a direct determination of the wind velocity and no uncertain dust-to-gas correction factor was needed. Results. Five RSGs in RSGC1 have been detected in CO(2–1). The retrieved ṀCO values are systematically lower than ṀSED. Although only five RSGs in RSGC1 have been detected, the data allow us to propose a new mass-loss rate relation for M-type red supergiants with effective temperatures between ∼3200 and 3800 K that is dependent on the luminosity and initial mass, and that is valid during the phase where nuclear burning determines the evolution along the RSG branch. The new mass-loss rate relation is based on the new ṀCO values for the RSGs in RSGC1 and on prior ṀSED values for RSGs in four clusters, including RSGC1. The new Ṁ-prescription yields a good prediction for the mass-loss rate of some well-known Galactic RSGs that are observed in multiple CO rotational lines, including α Ori, μ Cep and VX Sgr. Moreover, there are indications that a stronger, potentially eruptive, mass-loss process is occurring during some fraction of the RSG lifetime, suggesting that RSGs might experience a phase change in mass loss leading to the wind mass-loss rate dominating the RSG evolution at that stage. Conclusions. Implementing a lower mass-loss rate in evolution codes for massive stars has important consequences as to the nature of their end-state. A reduction of the RSG mass-loss rate implies that quiescent RSG mass loss is not enough to strip a single star’s hydrogen-rich envelope. Upon core collapse such single stars would explode as RSGs. Mass-loss rates of order ∼6 times higher would be needed to strip the H-rich envelope and produce a Wolf-Rayet star while evolving back to the blue side of the Hertzsprung–Russell diagram. Future observations of a larger sample of RSGs in open clusters should allow a more stringent determination of the ṀCO–luminosity relation and a sharper diagnostic as to when the phase change in mass loss is occurring.