Authorities and managers often rely on individuals and businesses' self-reports and employ various forms of honesty declarations to ensure that those individuals and businesses do not over-claim payments, benefits, or other resources. While previous work has found that honesty pledges have the potential to decrease dishonesty, effects have been mixed. We argue that understanding and predicting when honesty pledges are effective has been obstructed due to variations in experimental designs and operationalizations of honesty pledges in previous research. Specifically, we focus on the role of whether and how an ex-ante honesty pledge asks individuals to identify (by ID, name, initials) and how much involvement the pledge requires from the individual (low: just reading vs. high: re-typing the text of the pledge). In four pre-registered online studies (N > 5000), we systematically examine these two dimensions of a pledge to find that involvement is often more effective than identification. In addition, low involvement pledges, without any identification, are mostly ineffective. Finally, we find that the effect of a high (vs. low) involvement pledge is relatively more persistent across tasks. Yet, repeating a low involvement pledge across tasks increases its effectiveness and compensates for the lower persistency across tasks. Taken together, these results contribute both to theory by comparing some of the mechanisms possibly underlying honesty pledges as well as to practice by providing guidance to managers and policymakers on how to effectively design pledges to prevent or reduce dishonesty in self-reports.