Many new users don’t understand how ChatGPT works or how to write effective prompts. They often expect the model to infer missing context or produce perfect answers from very little information. When that doesn’t happen, users may feel that ChatGPT ignored their request, guessed details, or gave unreliable information, reducing trust in the product.
I believe a significant percentage of the perception that “AI lies” comes from this misunderstanding. Large language models are probabilistic systems designed to generate the most appropriate continuation of a conversation. When a user’s request lacks important information, the model may still produce a plausible-looking answer instead of asking for clarification. Users often interpret this behavior as intentional deception, even though it is a limitation of how the model operates.
I suggest adding a very short onboarding tip for new users (or making it optional in Settings), along with a link to a more detailed guide.
For example:
Tip: The more context you provide—such as your goal, constraints, and preferred response format—the better ChatGPT can help. If important information is missing, ChatGPT may ask clarifying questions instead of guessing. Learn more.
The “Learn more” page could include:
How to write effective prompts.
Why context improves response quality.
Examples of weak vs. strong prompts.
When ChatGPT may need additional information.
The probabilistic nature and limitations of large language models.
This small addition could significantly improve the first-time user experience, reduce frustration, increase trust in ChatGPT, and help users get better results without requiring them to learn prompt engineering in depth.
The goal is not to teach prompt engineering as a technical subject, but to help users understand how to communicate effectively with the AI from their very first conversation, while also setting realistic expectations about the quality of its responses. Better input generally leads to better output.