Usage-based theories of language learning suggest that native speakers of a language are acutely aware of formulaic language due in large part to frequency effects. Corpora and data-driven learning can offer useful insights into frequent patterns of naturally occurring language to second/foreign language learners who, unlike native speakers, are not privy to a lifetime of input and fine-tuning. Recently, the use of the web in combination with the Google search engine as an accessible corpus and concordancer has received much attention. This article describes an experiment which tests the hypothesis that native speakers of English perceive learner-generated phrases to be more natural after learners have searched the phrases on Google and modified them in light of the frequency of search results. The findings indicate that native speakers perceive phrases that generated more results in Google searches to be more natural. (Contains 3 tables, 2 figures and 2 notes.)