Role and Persona Prompting Quiz

Answers and explanations appear as you take the quiz.

1. What does persona prompting most reliably change in a model's output?
   • The factual knowledge the model can draw on
   • Tone, vocabulary and framing
   • The data the model was trained on
   • The price per token of the request
2. What did studies such as the one by Zheng and colleagues find when roles were added to system prompts for factual question answering?
   • Large, consistent accuracy gains for every role
   • Accuracy always dropped
   • Small, inconsistent effects that were hard to predict
   • Perfect accuracy once an expert role was used
3. Why does a persona like "a security reviewer" tend to be more useful than "the world's best assistant"?
   • It implies a specific checklist of standards to apply
   • It unlocks hidden security training in the model
   • It makes the model respond faster
   • It prevents all prompt injection attacks
4. Where does Anthropic's documentation recommend setting a model's role?
   • In the final line of every user message
   • In a separate file uploaded with the request
   • Inside the model's output
   • In the system prompt
5. Which of these is described as a misconception about persona prompting?
   • Personas can change the reading level of a response
   • Flattering labels like "world's best" raise output quality
   • A persona works better when its standards are spelled out
   • Personas are useful for setting audience
6. What risk comes with giving a model a confident expert persona?
   • It always refuses to answer
   • It doubles the token cost
   • It can make wrong answers more persuasive
   • It erases the system prompt
7. Which set lists the four parts of a useful persona given in the guide?
   • Name, age, hobby, location
   • Model, temperature, tokens, price
   • Tone, emoji, length, font
   • Identity, audience, standards, boundaries
8. What is the recommended way to find out whether a persona actually helps your task?
   • Run the same inputs with and without the role and score both against a rubric
   • Ask the model whether it likes the persona
   • Try it once and keep it if the answer looks good
   • Use the longest persona you can write