Abstract Background When conducting COVID-19 contact tracing, pre-defined criteria allow differentiating high-risk contacts (HRC) from low-risk contacts (LRC). Our study aimed to evaluate whether contact tracers in Belgium followed these criteria in practice and whether their deviations improved the infection risk assessment. Method We conducted a retrospective cohort study in Belgium, through an anonymous online survey, sent to 111,763 workers by email. First, we evaluated the concordance between the guideline-based classification of HRC or LRC and the tracer’s classification. We computed positive and negative agreements between both. Second, we used a multivariate Poisson regression to calculate the risk ratio (RR) of testing positive depending on the risk classification by the contact tracer and by the guideline-based risk classification. Results For our first research question, we included 1105 participants. The positive agreement between the guideline-based classification in HRC or LRC and the tracer’s classification was 0.53 (95% CI 0.49–0.57) and the negative agreement 0.70 (95% CI: 0.67–0.72). The type of contact tracer (occupational doctors, internal tracer, general practitioner, other) did not significantly influence the results. For the second research question, we included 589 participants. The RR of testing positive after an HRC compared to an LRC was 3.10 (95% CI: 2.71–3.56) when classified by the contact tracer and 2.24 (95% CI: 1.94–2.60) when classified by the guideline-based criteria. Conclusion Our study indicates that contact tracers did not apply pre-defined criteria for classifying high and low risk contacts. Risk stratification by contact tracers predicts who is at risk of infection better than guidelines only. This result indicates that a knowledgeable tracer can target testing better than a general guideline, asking for a debate on how to adapt the guidelines.