The paper investigates the impact of large language model (LLM)-powered conversational search systems on information seeking and opinion polarization. Two experiments were conducted to explore whether such systems increase selective exposure and whether opinion biases in LLMs exacerbate this effect. The first experiment compared conventional web search with LLM-powered conversational search, finding that participants engaged in more biased information querying with LLM-powered systems, particularly when the system reinforced their existing views. The second experiment examined the effects of opinion-biased LLMs, both reinforcing and challenging users' views, on selective exposure and opinion polarization. Results showed that participants exhibited higher levels of confirmatory information querying and polarization when using LLM-powered systems, especially when the system reinforced their views. These findings highlight the need for careful development and regulation of LLM-powered conversational search systems to prevent the creation of echo chambers and promote diverse information exposure.The paper investigates the impact of large language model (LLM)-powered conversational search systems on information seeking and opinion polarization. Two experiments were conducted to explore whether such systems increase selective exposure and whether opinion biases in LLMs exacerbate this effect. The first experiment compared conventional web search with LLM-powered conversational search, finding that participants engaged in more biased information querying with LLM-powered systems, particularly when the system reinforced their existing views. The second experiment examined the effects of opinion-biased LLMs, both reinforcing and challenging users' views, on selective exposure and opinion polarization. Results showed that participants exhibited higher levels of confirmatory information querying and polarization when using LLM-powered systems, especially when the system reinforced their views. These findings highlight the need for careful development and regulation of LLM-powered conversational search systems to prevent the creation of echo chambers and promote diverse information exposure.