The impressive success of recent AI-based text generation technologies fits perfectly with recent linguistic theorising that focuses on the patternedness of language and language use. Large language models work so well because they reproduce the highly complex patternedness of language in an exceptionally finegrained way. This contribution traces linguistic notions of patternedness in corpus linguistics and beyond and applies them to the domain of academic writing which is highly standardised and thus – at least in some parts – uncreative. Since generated texts, unlike human texts, do not make any validity claims, but at most claims of fit, their use should neither be overestimated as co-authorship nor concealed. On the contrary, their use as a supporting tool should be made explicit and transparent.