It has generally been assumed that after nonlow vowels in English, hiatus is resolved by inserting a homorganic glide (e.g. seeing [sijɪŋ], Itô & Mester 2009). However, despite suspicions that inserted glides may be fundamentally different from lexical glides (e.g. Cruttenden 2008), a systematic phonetic investigation of the purported glide is lacking. We examine the nature of hiatus resolution by comparing three environments: (i) vowel-vowel sequences within words (VV: kiosk), (ii) vowel-vowel sequences across word boundaries (VBV: see otters), and (iii) vowel-glide-vowel sequences across word boundaries (VGV: see yachts). The first finding is that a glottal stop produced between the vowels accounts for nearly half of the responses for VBV phrases, whereas glottal stops are present in less than 5% of the VV and the VGV conditions. Second, an acoustic comparison of VV, VBV, and VGV phrases not produced with glottal stops shows significant differences between the vowel-glide-vowel and the vowel-vowel sequences on all measures, including duration, intensity, and formants. These results indicate that American English speakers tend to resolve hiatus at word boundaries with glottal stop insertion, whereas there is no hiatus resolution at all within words. A brief optimality-theoretic analysis sketches out the phonological differences between hiatus at word boundaries and word-medially.