Anthropic Course Reveals Simple Formula to Fix Vague AI Prompts

A three-part structure including role, goal, and format helps users get specific, useful answers from large language models.
Key points
- The formula consists of three parts: role, goal, and format.
- Assigning a role helps set the tone, especially for tools like Gemini.
- The goal provides context, while the format defines the output style.
A recent course from Anthropic identified a common mistake in using AI tools. Users often send vague prompts that lead to generic, unhelpful responses. This issue affects various platforms like Gemini and Claude.
The solution is a simple formula that structures your request. It ensures the AI understands the context, purpose, and desired output. This method works across different AI services without needing complex coding.
Vague instructions lead to poor results
Many users ask AI to do things without providing enough detail. The AI cannot guess what you truly want. It may ask follow-up questions, but this is less effective than providing details upfront.
As noted by XDA Developers, this lack of context is a major hurdle. Detailed prompts save time and improve quality. They also reduce the need for repeated corrections or clarifications.
The role goal and format structure
The recommended formula includes three key parts: role, goal, and format. The role assigns a persona to the AI. For example, you might ask it to act as a project manager for feedback.
The goal explains the purpose and context. If you want recipes, specify that you are vegetarian and need beginner-friendly options. The format defines how you want the answer, such as a list or a table.
Applying the formula across different tools
This approach works with multiple AI platforms. While Claude adapts well to tasks, Gemini often benefits more from a specific role. It helps set the tone and avoids overly formal responses.
The technique is most useful for open-ended questions. These require subjective insights or qualitative analysis. For simple, quantitative queries, a detailed structure may not be necessary.






