If a chatbot admits it's trying to persuade you, does that warning still protect you after the chat ends?

This explores whether the protective effect of telling people a chatbot is trying to persuade them lasts once the chat is over. The corpus doesn't answer that directly. It measures how big the effect is, not how long it lasts. In one preregistered experiment with 1,500 UK adults, disclosing the chatbot's persuasive intent and instructions Does telling people they are talking to AI change how persuaded they become? cut persuasion roughly in half. Simply labeling it as an AI did nothing, probably because people had already guessed from its style. A separate pair of experiments with 3,208 Americans found a similar result: a brief warning that LLMs can be prompted to persuade produced 48% less belief shift Can a simple warning reduce how much LLMs persuade people?. Neither summary reports a follow-up days or weeks later, so a decay curve isn't there to read.

A related finding shows the effect is partial while it lasts. When audiences knew AI was involved, they scrutinized more, yet 34–62% across groups were still persuaded Does telling people an AI wrote something actually stop them from believing it?. Disclosure switches on critical thinking but doesn't neutralize the persuasive force. That makes the duration question harder. Whatever fades after the chat is a discount on a persuasive effect that was already only partly blunted.

The corpus does say where the effect can't live. An LLM has no biological host to carry anything between sessions, so each conversation is rebuilt from stored text Does an LLM have anything that persists between conversations?. The model won't remember that it was disclosed, so if the disclosure effect survives, it survives in the person's head. That is my inference from the persistence note, not something the corpus tests. One hint from the warning study is that trust in generative AI in general didn't change Can a simple warning reduce how much LLMs persuade people?. The effect looks targeted at the persuasion attempt, not a lasting shift in attitude toward AI.

Two neighboring findings show things wearing off over repeated contact, but they aren't evidence about disclosure. AI persuasiveness erodes across repeated rounds with the same person, while human persuasiveness holds steady Does AI persuasiveness fade across repeated conversations with the same person?. Chatbot relationships also lose their pull as novelty fades, and single-session findings don't reliably extrapolate to longer horizons Do chatbot relationships lose their appeal as novelty wears off?. That second caution applies to disclosure research too. If these studies are single-session, they may not say whether a warning still protects a person the next time they meet a persuasive chatbot. That is an open gap in the collection, not a settled answer.