Asymmetric information is prevalent in many economic settings and often leads to issues such as market inefficiency and adverse selection. In this study, we investigate information transmission in a sender-receiver game with a partially informed third-party sender whose interests may align with or conflict with those of the receiver. The experiment compares three communication formats: 1) state only (Baseline), in which only the state is communicated; 2) probability addition, in which the sender also reports his/her subjective probability that the state is truthful; and 3) text addition, in which the baseline recommendation is supplemented by a structured text message. The findings demonstrate that message format significantly affects both lying behavior and persuasion outcomes. Lying is most prevalent under structured text messages, followed by probability messages, and least under simple state messages, with these treatment differences emerging after a brief learning phase. Text messages elicit the highest rate of receivers following favorable messages, while probability messages increase following rate only when reported probability is moderate rather than extreme. Moreover, allowing senders to self-select richer message formats in a later stage unexpectedly reduces lying in the Baseline, suggesting that the presence of alternative communication channels can function as a commitment device against self-serving dishonesty.