AI is most useful when it has a clear job, enough context, and a person who can judge the result. It is weakest when the task depends on hidden facts, unsupported claims, or decisions that need accountability. Before writing a prompt, decide whether the task belongs in an AI-assisted workflow at all.
This lesson gives you a quick fit test. It is intentionally simple. You can run it in less than a minute before asking for a draft, summary, comparison, checklist, or plan.
Fit Is A Decision, Not A Default
The wrong question is, “Can AI do this?” The better question is, “What role could AI safely play in this workflow?”
Many tasks contain several smaller jobs. AI might be useful for one part and inappropriate for another. For example, it may help turn meeting notes into a first-draft action list, but a person still needs to confirm owners, dates, commitments, and anything that affects another team. It may help compare two policy drafts, but it should not invent the policy rationale or approve the final interpretation.
Treat AI as a support role. It can draft, organize, compare, summarize, brainstorm, and point out questions. It should not quietly become the owner of the work.
A Fit Test You Can Run In 30 Seconds
Ask five questions before using AI:
- Is the task easy to describe?
- Can I provide the source material or context the model needs?
- Can I review the output for correctness?
- Is the expected output a draft, structure, summary, or recommendation rather than a final action?
- Would a wrong answer be low consequence or easy to catch before it causes harm?
If most answers are yes, the task is probably a strong fit. If some answers are yes, it may be a partial fit: use AI for the narrow draft or analysis step, then keep review tight. If several answers are no, do not force it. Improve the context, narrow the task, or keep the work human-owned.
The most important question is reviewability. If you cannot tell whether the output is right, the task is not ready for AI assistance. A confident answer you cannot check is not leverage. It is risk with nice formatting.
Strong-Fit Task Types
Strong-fit tasks usually have clear input and a reviewable output. Examples include:
- turning rough notes into a cleaner outline
- summarizing a document when the source is provided
- comparing two drafts against a stated checklist
- generating first-pass questions for a planning session
- rewriting text for a known audience while preserving facts
- extracting action items from a transcript that a person will review
- producing a checklist from a policy or procedure
- proposing test cases from acceptance criteria
These tasks work because the model has something to work from and the human reviewer has something to check against.
Weak-Fit Task Types
Weak-fit tasks usually ask the model to fill in facts it does not have, make decisions it should not own, or produce output that cannot be verified. Be cautious with tasks such as:
- deciding what another person or organization should be told without human approval
- producing factual claims without source material
- handling secrets, credentials, regulated records, or sensitive personal details
- making final legal, financial, medical, security, or employment judgments
- updating systems, sending messages, or committing resources automatically
- summarizing material so long or messy that important context may be missing
- ranking people or opportunities when the criteria are unclear or unfair
A weak-fit task can sometimes become usable when you narrow it. Instead of asking AI to make the decision, ask it to list factors a person should consider. Instead of asking it to write the final message, ask it for a draft that must be reviewed. Instead of asking it to reason from memory, give it the source material and require uncertainty labels.
Choosing The AI’s Role
Once the task passes the fit test, choose the role before writing the prompt.
Use drafter when you need a first version that a person will revise. Good for outlines, emails, plans, summaries, and explanations.
Use organizer when you need messy material structured. Good for grouping notes, turning bullets into sections, or converting raw observations into a checklist.
Use reviewer when you need a second pass against clear criteria. Good for checking whether a draft follows requirements, misses obvious risks, or uses the wrong tone.
Use sparring partner when you need alternatives or questions. Good for planning, tradeoff analysis, and finding blind spots before a meeting.
Use extractor when you need facts pulled from provided material. Good for action items, terms, dates, owners, or decisions, as long as the source is included and the output can be checked.
If you cannot name the role, your prompt is probably too vague. Tighten the task first.
The Cost Of A Bad Fit
A bad fit is not neutral. It can create work that looks useful while quietly moving risk into the reviewer’s lap.
Common costs include:
- rework when a polished draft misses the actual need
- false confidence when unsupported claims sound complete
- review fatigue when a person has to check every sentence from scratch
- hidden risk when the output moves into a decision or workflow before anyone owns it
Some mistakes are annoying but recoverable. A weak meeting summary can be corrected. A rough outline can be rewritten. Other mistakes are expensive and hard to catch: a final message sent with the wrong commitment, a recommendation based on missing context, or a workflow action triggered from an unchecked assumption.
The fit test exists to catch that difference early.
Three Worked Fit Judgments
Task: Summarize a meeting transcript into decisions and open questions.
This is a strong fit if the transcript is provided and a person will review the summary. The AI role is extractor and organizer. The prompt should say not to invent owners, dates, or decisions that are not present.
Task: Decide whether a vendor proposal should be accepted.
This is not a good final-decision task. It may be a partial fit if AI compares the proposal against known criteria, highlights missing information, and lists review questions. The person or team still owns the decision.
Task: Rewrite a rough internal explanation for a non-technical audience.
This is usually a strong fit. The source text and audience are known, the output is reviewable, and the consequence of a weak draft is manageable. The prompt should require the model to preserve facts and flag anything unclear instead of filling gaps.
A Fourth Judgment: When The Answer Is “Not Yet”
Some tasks are not bad AI use cases. They are just not ready yet.
Task: Draft a customer-facing FAQ for a new policy.
This might become a strong fit later. AI can help turn approved answers into clear questions and responses. Right now, it is a weak fit if the policy answers are not agreed, the owner is unclear, or there is no review path for the final wording.
The better AI-assisted step is narrower:
List the questions this FAQ should answer. Mark each answer as provided, missing, or needs owner review. Do not draft customer-facing language yet.
That changes the role from final drafter to organizer. It also makes the missing inputs visible before the work becomes polished.
“Not yet” is a useful decision. It says the task may fit AI after the source material, owner, and review path are ready.
Signals You Chose The Wrong Fit
Recheck the fit if you notice any of these patterns:
- you spend more time correcting the output than the task would have taken
- you cannot tell whether the answer is right
- you keep adding more context because the model is guessing
- the task changed from draft support into a consequential decision
- the output needs facts, owners, or approval you do not have
- the model keeps producing confident language where uncertainty would be safer
When that happens, stop and narrow the role. Ask for questions, a checklist, a comparison, or an assumption list instead of a final answer.
Compact Exercise
Pick one task you were tempted to give to AI today. Run the five-question fit test. Label it strong, partial, or weak. Then name the AI role: drafter, organizer, reviewer, sparring partner, or extractor.
If the task is partial or weak, rewrite it into a narrower AI-assisted step that still leaves the final judgment with a person.
Reflection & Next Step
Before moving on, choose one real task and write a one-line fit decision:
AI role: [drafter / organizer / reviewer / sparring partner / extractor / not ready yet]
Reason: [fit reason]Then name the human owner for review. If you cannot name that owner, the task is not ready for an AI-assisted workflow.
Use the Use-Case Fit Worksheet to capture the decision, then continue to Prompt Structure That Survives Real Work once the task has a clear AI role.