Drift diffusion models with time-dependent drift rates have gained popularity because they allow the speed at which internal evidence accumulates to vary over time. Here, we aim to provide more direct evidence for the time course of evidence accumulation using participants’ metacognitive judgments: We model the evidence accumulation processes as a confidence-weighted majority vote — which is typically only used to study group decisions. In the same way as a group member adds a vote and a confidence rating to a group decision process, we model each time step of a visual evidence accumulation process to add a vote and confidence rating (corresponding to the noisy visual evidence accumulated at that time point). This allows us to disentangle the contribution of each time step to the overall task performance. In a simple psychophysical experiment, N=6 participants each measured in three 2 h long sessions discriminated masked horizontal/vertical stimuli at different presentation durations. Using log-odds transformation and calibration, we obtained estimates for the internal evidence distribution after different presentation durations. Despite substantial interindividual differences, participants showed a reliable acceleration in their evidence accumulation throughout the first 50 milliseconds of stimulus presentation. In conclusion, our approach replaces model assumptions by metacognitive measures to estimate time-dependent drift rates, which may give more empirical access to time-dependent drift rates but can also be biased by participants’ metacognitive abilities.