Discover
Widen the map before deciding what to produce.Ask AI to identify the structure of the topic, key issues, risks, options, unknowns, and useful questions that may otherwise be missed.
Start HereNEXIUS RESEARCH PAPER / AI LITERACY + FLUENCY
Validating the DUO Framework.
A conversation-based approach to effective generative AI use: Discover, Understand, then Output.
DUO DISCOVER / UNDERSTAND / OUTPUT
EVIDENCE STATUS
Theoretically grounded.
Not yet empirically proven.
The paper validates the logic behind DUO against adjacent research. It does not claim that the named three-stage framework has already been tested directly. This distinction defines the next research step.
ABSTRACT
Better output begins
before the output.
This paper examines the DUO framework - Discover, Understand, Output - as a practical model for improving how professionals use generative AI tools such as ChatGPT. DUO asks users to discover the structure of a topic, deepen understanding through questioning and refinement, and only then request the final output. The named framework has not yet been empirically validated, but its logic is supported by research on scaffolding, metacognition, iterative prompting, human-AI co-creation, and writing process theory.
THE THREE-STAGE MODEL
A simple structure for
a better AI conversation.
Understand
Turn broad information into contextual judgement.Challenge assumptions, request examples, compare options, test what is missing, and apply the emerging understanding to the user's real situation.
Output
Produce only after the thinking has been refined.Create the memo, report, presentation, proposal, checklist, email, or other deliverable using the clarified audience, purpose, priorities, and constraints.
THEORETICAL FOUNDATIONS
Five bodies of evidence
support the logic.
DUO is useful as an instructional framework because it translates established learning and collaboration principles into a sequence that non-specialists can remember and apply.
- 01
Scaffolding
AI provides structured support while the user remains responsible for judgement and gradually builds competence.
- 02
Metacognition
The sequence encourages users to plan, monitor, question, evaluate, and adjust their own approach.
- 03
Iterative prompting
Context and instructions improve over multiple turns instead of depending on one supposedly perfect prompt.
- 04
Human-AI co-creation
AI expands and refines possibilities while the human evaluates, directs, and makes the final decision.
- 05
Writing process theory
Discover resembles prewriting, Understand supports organisation and critical development, and Output supports drafting and polishing.
PRACTICAL VALUE
What DUO changes
in professional work.
- Reduces generic output by building context before production.
- Improves judgement by requiring users to question assumptions and compare options.
- Keeps the user accountable instead of treating AI output as a finished answer.
- Provides a simple teaching model that is easier to remember than a long prompt formula.
- Improves the likelihood that final deliverables are relevant, structured, and fit for purpose.
LIMITATIONS
Use the framework
without overstating it.
DUO depends on active participation. Its benefits weaken when users skip the Understand phase, accept responses without review, or fail to verify consequential claims.
It does not remove the need for factual verification in legal, financial, medical, tax, regulatory, or technical work. It may also be unnecessary for very simple tasks such as basic rewording or format conversion.
NEXT RESEARCH STEP
Move from theoretical support
to measurable validation.
Compare prompting approaches
Compare one-shot prompting, a conventional prompt template, and DUO on relevance, clarity, structure, contextual fit, originality, accuracy, confidence, and time.
Measure human learning
Compare whether DUO users can understand and complete a related task without AI more effectively than users who learned through one-shot prompting.
REFERENCES NAMED IN THE PAPER
Starting points for
further investigation.
Don-Yehiya, S., Choshen, L., & Abend, O. (2023). Human learning by model feedback: The dynamics of iterative prompting with Midjourney.
Li, H., Leung, J., & Shen, Z. (2024). Towards goal-oriented prompt engineering for large language models: A survey.
The paper also draws on the wider literature concerning metacognitive scaffolding, reflective prompting, human-AI co-creation, and process-based writing.
CONTINUE FROM RESEARCH TO PRACTICE
Build fluency first.
Then redesign the work.
Connect DUO to the wider Nexius approach for AI-capable workforces and governed agentic operations.