The collaborative research center “Co-constructing explainability” (TRR 318) deals with creating everyday explanations of decisions made by black-box AI algorithms. Even if an explanation itself was correct and coherent, there may be biases that hamper proper understanding or worse, lead to misconceptions. One way to mitigate this issue is to guide attention. Directing attention in a live interaction between a human and an agent requires repeatedly assessing attention within a reasonable timeframe. Temporal-order judgments in conjunction with Bundesen’s theory of visual attention (TVA) have been a tool to assess visual attention. Current Bayesian approaches estimating attentional parameters happen only after conclusion of the experiment. As a first step to real-time attention assessment, we will evaluate the precision of parameter estimations under typical experimental settings and constraints, which entail gradual data accumulation and varying data quality. Considering expected effect sizes, we will discuss the practical feasibility and utility of the presented approach.