The AI IntentGap Framework.
The AI Intent Gap is the distance between what a person wants accomplished and the instruction an AI system can actually act on. The framework makes that distance inspectable before the first response.
Download the framework PDFFour places intent breaks
A useful diagnosis, not a longer-prompt rule.
Outcome gap
The request names an activity but not the result, decision, or definition of done.
Context gap
The model lacks audience, background, source material, or assumptions that a colleague would already know.
Boundary gap
Constraints, exclusions, evidence rules, privacy limits, or approval requirements are missing.
Execution gap
The instruction does not say how the output will be used, checked, handed off, or acted upon.
Operational consequences
The cost is not merely a “bad answer.”
Rework
More correction loops, duplicated explanation, and slower decisions.
Inconsistency
Similar work produces different formats, assumptions, and quality levels.
Risk
Sensitive data, unsupported claims, or missing approvals reach the wrong execution path.
Weak learning
Teams see model output but cannot tell which instruction habit caused the failure.
Maturity model
From correction loops to governed instruction assets.
- 1 · Reactive
Teams correct weak AI output after it arrives. Quality depends on individual prompting habits.
- 2 · Structured
Common work uses clearer outcomes, named inputs, constraints, and output contracts.
- 3 · Reusable
Good instructions become maintained assets with owners, context, and revision history.
- 4 · Governed
Privacy boundaries, policy checks, human approvals, and aggregate quality signals are explicit.
- 5 · Adaptive
Instruction quality improves from measured failure patterns without treating raw conversations as default telemetry.
Next evidence