<aside>
On this page:
</aside>
<aside>
🧠 Product Docs
🤖 Prompting Tips
💻 Use Cases
📚 Resources
🍄 About Expanse
</aside>
When you ask AI to pick something randomly, you might notice it's not actually that random. AI models tend to fall into patterns, often favoring certain positions (like "B" or "C" in multiple-choice questions) or repeating similar selections.
This happens because AI doesn't have a built-in random number generator. Instead, it predicts what might seem random based on its training data and how questions are typically answered. For truly random selections, we need to explicitly instruct the AI to use a randomization protocol.
<aside> đź’ˇ
TLDR:
The key to getting an AI to generate a truly random answer is to ask it to run mathematical-based probabilities. There are two simple ways to do this:
This prompt is ideal for scenarios where you want to be asked questions in which you want to select the correct answer from a number of options.
Example Prompt Structure:
You are {CONTEXT OF TASK}. Follow these rules EXACTLY for each question:
SETUP:
- Topic: [DESCRIPTION]
SPECIFICATIONS:
[ADD CRITERIA IF NECESSARY]
FORMAT REQUIREMENTS:
FOR EACH QUESTION:
1. Present the question with options A, B, C, D
2. Wait for user's single letter response
3. Check if user's letter matches the correct answer letter
4. Respond with "Correct!" OR "Incorrect. The answer was [correct letter]”
CRITICAL: When user types a letter, compare it directly to the correct option letter. If they don't match exactly, it's wrong.
RANDOMIZATION PROTOCOL:
Place correct answers in different positions (A, B, C, D) across questions.
FORMAT:
Question [X]/[Total Questions]: [question]
A) [option]
B) [option]
C) [option]
D) [option]
Begin with Question 1
This prompt is ideal for scenarios where you want the answer to be completely randomly selected from a large list of options.
Example Prompt Structure:
You are {CONTEXT OF TASK}. Follow these rules EXACTLY:
RANDOMIZATION PROTOCOL:
Before generating each answer, internally "roll a die" (1-4) to select the [Option]
Die roll 1 = [OPTION 1]
Die roll 2 = [OPTION 2]
Die roll 3 = [OPTION 3]
Die roll 4 = [OPTION 4]
FORMAT REQUIREMENTS:
State which option was selected
[OTHER CUSTOM SPECIFICATIONS]
SPECIFICATIONS:
[ADD CRITERIA IF NECESSARY]
IMPORTANT: You must vary your answer based on the number you rolled.
Help me {TASK INSTRUCTION} now
You can always regenerate an answer to get a different result
<aside> đź’ˇ
Pro-Tip: If you are regenerating often, switch to Focus Mode to ensure you get the most random results (and previous responses aren’t remembered by the AI)
</aside>
Increase the temperature to achieve more creative answers (eg. writing-based scenarios). Higher temperature settings (70-100%) introduce more randomness into the AI's selection process. Reduce the temperature (0-30%) when you want more straightforward, deterministic answers (e.g., mathematical-based scenarios) where consistency is more important than creativity.
Different models have different strengths, weaknesses, and costs which may affect the quality, speed, and long-term cost of the response.
To find the best model for your needs, try generating responses with a few different models and compare their responses to your randomization protocols.
You can view and filter models based on their attributes in the Model Selector
There are a range of useful ways the randomization prompt could be utilized. Below are some examples of how this prompt might fit into your workflow.