TL;DR — GROW (Goal-Reality-Options-Will) is the most common coaching scaffold in AI products. CLEAR, OSKAR and IGROW are alternatives that fit different conversation styles. The choice should follow the product goal — outcome-focused, solution-focused, or change-focused.
We built Blink AI around GROW because it is well-established and easy to operationalize as a state machine. It is not the only option. Different methodologies map to different product goals, and choosing one based on what your product actually does is more important than picking the most famous one.
GROW
Goal → Reality → Options → Will. Outcome-focused. The model is a state machine that progresses through phases. Easy to implement, easy to instrument. Default choice.
CLEAR
Contracting → Listening → Exploring → Action → Review. Listening-heavy, slower-paced. Fits longer-form coaching where the relationship matters. Harder to instrument because “listening” turns do not have clear exit criteria.
OSKAR
Outcome → Scaling → Know-how → Affirm + Action → Review. Solution-focused, optimistic, fast. Works well in skill-development products. The scaling step (“on a scale of 1-10, where are you now?”) is unusually easy to render in UI.
IGROW
Identify (the issue) added to the front of GROW. Useful when users arrive without a clear goal. Effectively turns the first phase into an extended diagnostic.
How to pick
- Outcome-focused product — GROW or OSKAR
- Relationship / long-term coaching — CLEAR
- Users arrive with vague concerns — IGROW
- Single-issue, fast resolution — OSKAR
The shared engineering reality
Whatever methodology you pick, the engineering is the same: structured phase state, phase-specific system prompts, exit criteria the orchestrator can check, an instrumentation layer that tracks phase progression. The methodology is the product; the architecture is the substrate.
Frequently asked questions
Should we pick one methodology or mix?
Pick one as the spine. Mix small pieces where they help. A session structured by GROW with CLEAR-style listening prompts is fine. A session that switches methodologies mid-flight confuses the model and the user.
Is methodology required at all?
For coaching products, yes. Without one, the AI defaults to advice-giving. The methodology is what makes the product distinct from “ChatGPT, but it asks questions sometimes”.
Working on something similar?
T-Square architects, builds and operates production systems for learning, AI and custom software products. Talk to a senior engineer for a second opinion.