This study addresses the relationship between the reporting of speech and the reporting of thought based on a set of annotated data from several typologically diverse languages. The context-based approach we adopt allows us to explore the relationship between reported speech and reported thought without relying on specific lexical or constructional cues. We identify three types of relationship between reported speech and reported thought. First, in all our languages, reported speech constructions can be recruited for the expression of reported thought; such uses can only be identified based on context. Second, in some of the languages, we find expressions that are best described in terms of speech-to-thought coercion: in a construction normally associated with reported speech, a verb of thinking or another lexical marker triggers a reported thought interpretation. Such an interpretation is sometimes at odds with the construction’s original properties, since situations of thinking and speaking differ in a number of linguistically relevant ways. For example, situations of thinking do not involve an addressee, and that difference accounts for the seemingly superfluous use, in some languages, of expressions such as “think inside one’s head” or “think to oneself”, which serve to reconcile the construction’s argument properties (an implied addressee) with its coerced interpretation. Speech-to-thought coercion also explains why, despite the absence of an addressee in the verb’s argument structure, it is possible to refer to addressees within thought reports. Finally, in some languages, reported thought constructions are attested that have no equivalent among expressions of reported speech. We conclude, based on the structural diversity of reported thought expressions, that reported thought cannot be treated as a uniform cross-linguistic concept in a way similar to reported speech (Spronck & Nikitina 2019), and that conceptualization of thought processes varies across languages in ways that make direct comparison impossible.