Practical guide
Context framing: define the work before prompting
Turn an ambiguous assignment into a usable brief, with a worked office-move exercise and a practical review checklist.
Context framing means deciding what problem the work must solve before asking AI to produce an answer. A useful frame names the decision, audience, constraints, available evidence and deliverable. This guide offers an original practice exercise for making those choices explicit; it does not treat a polished prompt as proof that the underlying task is well understood.
Separate the decision from the requested document
Start with the action somebody needs to take, not the file they asked you to create.
In a fictional exercise, an operations lead requests a presentation about office options for forty employees. The actual decision is whether to renew a lease, relocate or negotiate a temporary extension. A slide deck comparing attractive buildings could miss the decision entirely if the present lease expires before any alternative is available.
Write a one-sentence decision statement and identify its owner. Then list what that person can approve today. Distinguish a recommendation from permission to commit money. This changes the brief from a broad research assignment into a bounded choice with a deadline and an accountable reader.
Make constraints testable
A constraint should let a reviewer rule an option in or out without guessing what the author meant.
For the office exercise, separate the move date, approved budget ceiling and accessibility requirements from preferences such as decor or proximity to cafes. If accessibility requirements have not been confirmed, mark them as missing information rather than inventing a standard. Do not quietly turn an unknown into a flexible preference.
Ask AI to restate the brief as conditions an option must satisfy. Review each condition yourself against the assignment materials. A useful output includes the source of the condition, who can clarify it and the consequence of getting it wrong. These fields make disagreements visible before detailed research begins.
Expose assumptions before adding detail
Use an assumption log to distinguish supplied facts from choices made for convenience.
The exercise may state that forty people are employed, without saying how many desks are needed each day. Treating headcount as desk demand could distort every cost comparison. Record the assumed attendance pattern and ask whether the decision is sensitive to it. The point is to locate uncertainty, not to produce a longer prompt.
Try two plausible attendance arrangements and see whether the shortlist changes. If it does, ask for attendance evidence before recommending an option. If it does not, explain why the assumption is currently less consequential. Keep these trial arrangements visibly hypothetical so they cannot be mistaken for observed company behavior.
Test the frame with a deliberately unsuitable answer
A cheap counterexample can reveal an incomplete brief before the full assignment is attempted.
Give the reviewer an office that fits the budget but opens after the lease ends. Ask whether the brief clearly explains why this option fails. Next offer a short extension that is expensive but preserves continuity while missing requirements are resolved. A good frame should support comparison of that tradeoff without silently changing the objective.
Have a cold reader describe the expected deliverable from the brief alone. If they propose an attractive-property catalogue while the owner expects a decision memo, revise the brief. This is a practical comprehension check, not a numerical measurement of general ability.
Hand over a frame that can change
Finish with a compact brief and a rule for revisiting it when circumstances change.
The handoff should contain the decision statement, decision owner, deadline, mandatory conditions, preferences, evidence gaps and desired output. Add a short definition of a usable recommendation: for example, an option comparison with unresolved conditions and the next approval required. The format can be simple; the distinctions matter more than the template.
Now introduce a landlord extension offer. Ask which parts of the brief change and which remain fixed. A reviewer can examine whether the worker updates the decision horizon without losing the accessibility question. Keep conclusions tied to this exercise and its materials, rather than inferring a permanent trait from one completed brief.
Sources and scope
NIST addresses AI risk management. Skills England describes workplace AI foundations.
These sources provide background, not endorsement of this exercise. The worked example and suggested review method are original illustrative guidance. They are not customer results, validated benchmarks or evidence of a particular product capability. Adapt the exercise to the task and use qualified review where consequences require it.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England
Explore this topic
- How to develop and assess assumption logging in AI-enabled work
Assumption logging is the practice of recording an uncertain belief that shapes a task, identifying why it matters, and defining how it will be checked or revised. In AI-enabled work, it prevents plausible model output from silently becoming project fact and gives collaborators a visible route for challenge, evidence gathering, and decision updates.
- How to develop and assess audience definition in AI-enabled work
Audience definition is the ability to identify who must use a work product, what they need to do with it, and which differences in context could change its design. In AI-enabled work, it prevents a fluent generic output from standing in for evidence about actual readers, decision rights, accessibility needs, domain knowledge, and operating conditions.
- How to develop and assess constraint identification in AI-enabled work
Constraint identification is the ability to find, classify, and trace conditions that limit acceptable ways to complete a task. In AI-enabled work, it keeps an assistant's plausible plan from ignoring a deadline, approval right, data boundary, required format, technical dependency, or non-negotiable policy that changes what a valid answer looks like.
- How to develop and assess decision framing in AI-enabled work
Decision framing is the ability to define the choice a piece of work must support before analysis or drafting begins. It identifies the decision owner, objective, viable options, evidence threshold, constraints, timing, and consequences of delay. In AI-enabled work, a clear frame keeps a fluent answer from solving a different problem than the one the accountable person must decide.
- How to develop and assess problem decomposition in AI-enabled work
Problem decomposition is the ability to divide a complex task into connected pieces that can be investigated, produced, and verified without losing the purpose of the whole. In AI-enabled work, decomposition defines where AI can assist, which inputs each step requires, how outputs connect, and where human judgment or integration checks remain necessary.
- How to develop and assess scope control in AI-enabled work
Scope control is the ability to keep work aligned with an agreed outcome while evidence, requests, and AI-generated possibilities continue to expand. It requires a visible baseline, an explicit route for evaluating changes, and a record of what was accepted, deferred, or excluded. In AI-enabled work, this skill prevents useful exploration from silently turning into a different assignment.
- How to develop and assess success criteria in AI-enabled work
Success criteria turn a broad ambition into evidence that can be checked. They define what outcome matters, for whom, under which conditions, by when, and what would count as adequate, inadequate, or still unknown. In AI-enabled work, explicit criteria help a worker judge the whole result rather than rewarding fluency, speed, or tool use by themselves.
- How to develop and assess task interpretation in AI-enabled work
Task interpretation is the ability to turn an assignment, evidence pack, and working context into a faithful account of what must be produced. It identifies the intended user, required result, usable inputs, constraints, ambiguities, and completion conditions before execution begins. In AI-enabled work, this prevents a plausible output from answering a nearby but different request.