Self-Ask Prompting
A prompting technique where the AI explicitly asks and answers its own sub-questions before producing a final answer.
Also known as: Self-Questioning Prompting, Follow-Up Question Prompting
Category: Techniques
Tags: ai, prompting, llm-techniques, reasoning, decision-making
Explanation
Self-ask prompting is a structured reasoning technique in which the model is instructed to decompose a question into smaller follow-up questions, answer each one, and then combine the answers into a final response. It makes the decomposition step explicit rather than hoping the model does it implicitly.
A typical self-ask flow looks like this:
```
Question: Who was president of the US when the inventor of the telephone was born?
Are follow-up questions needed here: Yes.
Follow up: Who invented the telephone?
Intermediate answer: Alexander Graham Bell.
Follow up: When was Alexander Graham Bell born?
Intermediate answer: 1847.
Follow up: Who was US president in 1847?
Intermediate answer: James K. Polk.
So the final answer is: James K. Polk.
```
Why it works:
- **Forces decomposition**: Multi-hop questions are broken into single-hop lookups the model handles more reliably.
- **Reveals reasoning gaps**: Wrong intermediate answers are easier to spot and correct than a wrong final answer.
- **Enables tool use**: Each follow-up question can be routed to a search engine, database, or other tool instead of relying on the model's memory.
- **Improves factuality**: Particularly strong on multi-hop question answering where chain-of-thought alone underperforms.
Self-ask is closely related to chain-of-thought prompting but makes the sub-question structure explicit, which is especially useful when integrating retrieval or tool calls. Modern agentic systems often use self-ask as the skeleton for a plan-and-execute loop.
When to use: multi-hop factual questions, research tasks, any problem where decomposing into discrete look-ups improves accuracy. When not to use: simple one-step questions where the overhead outweighs the benefit, or tasks where reasoning is inherently holistic.
Related Concepts
← Back to all concepts