Learner and L2 user corpora are increasingly valued in language testing and assessment as they can inform test design, revision, and validation. This paper illustrates the benefits of using an L2 corpus to explore patterns of epistemic stance marking in computer-mediated speaking tests with no live human interlocutor. Drawing on the British Council-Lancaster Aptis Corpus – comprising over 630,000 words of L2 speech – we explored the frequency of epistemic stance markers (adverbial, adjectival and verbal) across proficiency levels and speaking task types. The analysis revealed that epistemic stance was prevalent in test-taker discourse and that frequency was influenced by L2 proficiency and task type. The findings demonstrate that computer-mediated speaking tests can elicit expressions of epistemic stance in a comparable way to tests which involve human-human interaction. Implications are drawn for examiner training, test preparation, and an enriched understanding of the elements of pragmatic competence that can be elicited in computer-mediated speaking assessment.