Can social media sites use LLMs to make bots that simulate their users?

For example, would it be feasible to augment GPT4 by additionally training it on a particular user's social media comments to create a bot that can simulate that user (but with all of GPT4's capabilities still intact)?