Text generated by language models is becoming common by the day because of easy access to free tools. Detection of AI-generated text is being sought as a basic requirement to avoid abuse of this technology. There are multiple ways proposed in the literature to detect AI-generated content. However, most of them require longer texts and fail to localize to particular boundaries within the text. Therefore, we turn to sentence-level detection of AI-generated text. In this study, we investigate the basic statistical method of perplexity and how it can be applied to the detection of AI-generated sentences in order to establish a baseline. We propose a sentence representation using interpolated perplexity vectors and conclude that perplexity based detectors are useful for detecting text generated by the same model and may be used for detecting text generated from models of their own family. Our work is available here .