Self-Verification Prompting: Asking AI to Check Its Own Work

Writing prompts is like riding a bike. You do not need a higher education to get started. But with the right technique you can go faster and further.

A common way for prompting is back-and-forth interaction with AI system. User will critique the output and request AI to improve on it and in return AI will regenerate, hopefully, a better result.

Traditional prompting is slow and inconsistent

This approach can eventually produce high quality output, but at the cost of time invested. Now, since the result was not generated on a single prompt, but rather on a multiple iterations, replicating similar result will not be easy.

A simple fix is the ‘Self-Verification Prompt’ technique that unlocks LLM’s potential for better, more consistent results and removes the need for back-and-forth refinement.

What is Self-Verification Prompting?

It is a rather simple concept that tells AI model to acts as its own quality controller by checking its own work before moving to the next step and delivering the final output.

Below is a three-step internal cycle example that happens before AI model generates the final result:

1. Generation: AI model generates ‘draft’ answer.
2. Verification: AI model tests that draft against the set of rules provided.
3. Refinement: If the test fails, AI model interprets what went wrong and goes back to step 1 to adjust/improve the draft.

Why Self-Verification Prompt technique works?

1. It allows for a fewer hallucinations and overconfident errors, especially when working with details and numbers
2. Outputs will be more consistent since the model follows the same review checklist each time
3. Better decision support: you get clearer assumptions and cleaner reasoning, not just a conclusion

Below we have an example of a standard prompt and a self-verified prompt.

😎 All of this is done with a single prompt!

👉 Transform Your AI Prompting from Average to Exceptional with those 4 simple steps!