In addition to their intended purpose, graded reader texts can be made into a corpus appropriate for use with lower-level learners. Here I consider using such a corpus for data-driven learning (DDL), to make this approach more accessible to intermediate level students. However, how far does grading the corpus in this way compromise the authenticity of the language learners are exposed to? The simplified nature of such corpora may limit learners' exposure to lexical chunks, which are fundamental to the acquisition of natural and fluent language. This paper compares lexical chunks in graded corpora and the British National Corpus, examining frequency, type, and composition, to evaluate the ‘authenticity’ of graded input. Despite some differences, it is argued that the scale and type of lexical chunks are sufficient to provide input that reflects authentic language, suggesting that graded readers may offer an acceptable balance of accessibility and authenticity.