We investigated how deictic shifts and memory representation influence production difficulties in reported speech. In Experiment 1, participants read short stories, including a conversation between two protagonists. Participants then recalled the last sentence from the conversation in either direct or indirect speech. When participants had verbatim memory of the to-be-reported utterance, direct speech production was faster than indirect speech if indirect speech production required a deictic shift. If participants had verbatim memory and indirect speech production did not require a deictic shift, direct and indirect speech’s speech latencies were similar. In Experiment 2, we investigated the production difficulties of reported speech when verbatim memory was unavailable to participants. First, participants read short stories. Next, half of the participants completed an intervening task, whereas the other half did not. Last, participants undertook a sentence completion task by filling in a missing word in direct or indirect speech. When participants did not have verbatim memory of to-be-reported utterances, direct speech production was faster than indirect speech if a deictic shift was needed to produce direct speech. These results suggest that production difficulties in reported speech depend on how to-be-reported utterances are represented in memory and whether deictic shifts are required.