FATA Framework

First Ask Then Answer

Overview

A two-stage interaction paradigm where AI systems proactively generate supplementary questions before answering, rather than reactively clarifying only when ambiguity is detected.

The core move: use the model’s reasoning to scaffold user expression, helping non-experts provide the information that experts would know to ask for.

Components

Stage 1 — Systematic Question Generation: The system generates multi-dimensional questions covering contextual background, constraints, preferences, environmental factors, and historical context. All questions are produced at once (single-turn) rather than through iterative back-and-forth.

Stage 2 — Context-Enhanced Response: The system integrates original query with collected supplementary information to produce personalized, expert-level responses. The supplementary information creates intersecting constraints that narrow the solution space.

When to Use

  • When users are non-experts in the problem domain
  • When personalization significantly improves response quality
  • When the cost of a wrong answer is high enough to justify the upfront investment
  • When the interaction can tolerate a structured questioning phase

Limitations

  • Adds friction to the interaction (questions before answers)
  • Single-turn questioning may miss complex interdependencies
  • Effectiveness depends on users actually answering the questions well
  • Not appropriate for simple factual queries
  • Quality of questions depends on the underlying model

Practical Considerations

The framework is prompt-only, no fine-tuning required. It can be integrated with existing tool-calling, RAG, or agent architectures as a preprocessing step.

Built-in constraints should prevent over-questioning and protect privacy. The questioning phase needs to feel like helpful scaffolding, not interrogation.

Related: 07-molecule—expert-consultation-externalization, 01-atom—information-scaffolding, 01-atom—expertise-information-gap