BananaNL Blog

ChatGPT / 5 min

ChatGPT requirements prompts: separate goals, scope, assumptions, and acceptance criteria

A practical ChatGPT requirements prompt guide for turning notes and ideas into a draft that separates goals, users, scope, functional and non-functional requirements, open questions, and acceptance criteria.

BananaNL in 1 minute

A quick look at how selected prompts move into NotebookLM and AI Chat input fields.

Watch on YouTube

“Define the requirements” can blend decisions, assumptions, and gaps

ChatGPT can help turn discovery notes, proposals, existing specifications, and stakeholder requests into a draft requirements table. But a bare request to “define the requirements” can turn undecided assumptions into apparent facts, or list features without giving reviewers enough context to decide what belongs in scope.

OpenAI’s prompt guidance recommends clear and specific instructions, necessary context, breaking complex work into focused requests, and iterating on the output. For requirements work, define the business goal, users, scope, constraints, and output fields before asking for a draft so AI suggestions do not become confused with agreed facts.

Requirements guidance commonly separates business needs, functional requirements, non-functional requirements, operating conditions, and acceptance criteria. Use ChatGPT for question discovery, structure, and review angles—not as the final authority for scope, feasibility, security, legal, or stakeholder decisions.

Abstract image of a ChatGPT requirements prompt separating goals, scope, functional and non-functional requirements, open questions, and acceptance criteria

Five fields to define before using a requirements prompt

  1. Set the goal and success condition. State what problem should improve, for whom, and what outcome matters; mark a metric or decision as open when it is not agreed.
  2. Separate people and scope. List users, administrators, external parties, affected workflows, and out-of-scope work so feature decisions remain traceable.
  3. Keep business, functional, and non-functional requirements in different fields. Include performance, availability, permissions, data retention, operations, and accessibility—not only screens and clicks.
  4. Separate facts, assumptions, questions, and pending decisions. Tell the model not to fill gaps with guesses, so unsupported ideas can go back to stakeholders as assumptions.
  5. Add acceptance criteria and named reviewers. Humans should agree what proves each requirement is met; involve the appropriate specialists for legal, security, and personal-data decisions.

Four inputs to prepare before you prompt

Background, the problem to solve, and the intended outcomeUsers, workflow, usage situation, and what is out of scopeExisting systems, deadline, budget, integrations, and constraintsOpen questions, stakeholders to consult, and draft acceptance criteria

What to emphasize for each use case

New web serviceStart with users, problem, main flow, priority, and exclusions. Confirm the intended outcome before proposing screens.
Existing workflow improvementList the current flow, pain points, exceptions, manual work, and fixed constraints. Keep the current state separate from the desired change.
External API or tool integrationTrack data exchanged, timing, failure handling, permissions, and responsibility boundaries. Do not finalize technical specifications from model guesses.
Requirements review tableKeep requirement, evidence, priority, assumption, open decision, acceptance criterion, and reviewer together so the meeting can focus on decisions.

Requirements prompts to try in ChatGPT

Turn discovery notes into a draft requirements table

Organize the following discovery notes into a draft requirements document. Do not add facts that are not in the source. Return sections for “Goal and success condition”, “Users and scope”, “Business requirements”, “Functional requirements”, “Non-functional requirements”, “Constraints”, “Open questions”, and “Questions for stakeholders”. For every item, include the supporting source point. Notes: """(paste notes here)"""

It keeps confirmed information and questions that still need a decision in the same reviewable document.

Make acceptance criteria testable

Create a draft acceptance criterion for each feature below. Split each criterion into “Precondition”, “User action”, “Expected result”, and “Exceptions or items to confirm”. If numbers, legal requirements, security policy, or external specifications are missing, do not guess; write “Needs confirmation”. Feature draft: """(paste here)"""

The team can discuss evidence of completion, rather than only describing implementation.

Review a requirements table for gaps and conflicts

Review the requirements table below. Do not invent new facts. Classify every row as “Confirmed”, “Assumption”, “Undecided”, or “Possible conflict or gap”. Then list prioritized questions across business, function, non-functional needs, operations, data, permissions, exception handling, and acceptance criteria. Table: """(paste here)"""

It turns model suggestions into review topics instead of silently accepting them as specification.

Save reusable requirements prompts in BananaNL

Requirements work repeatedly needs the same structure: goal, scope, constraints, open questions, reviewer questions, and acceptance criteria. Save templates for new initiatives, workflow improvements, and integration reviews in BananaNL, then replace only the project context for each new task.

BananaNL is a Chrome extension that inserts saved prompts into the input field of AI Chat tools such as ChatGPT, Gemini, and Grok. It never auto-sends, so you can review confidential information, evidence, and agreement status before sending. Under the current pricing boundary, NotebookLM can be started for free; AI Chat prompt viewing, insertion, and saving are paid features; Image Collections, image saving, image editing, and video conversion are free features.

Abstract image of calling a saved requirements prompt template from BananaNL

FAQ

Can ChatGPT finalize requirements by itself?

No. It can help discover questions and structure a draft, but stakeholders and responsible reviewers must confirm facts, priority, feasibility, legal and security implications, and acceptance criteria.

What should I provide first?

Provide the background, problem, users, current workflow, scope, constraints, deadline, decisions already made, and open points. For long source material, structure it by topic or meeting before combining it.

Can I paste customer or internal material?

Check the AI service settings and your company rules first. Remove or anonymize personal data, contract terms, unreleased information, and credentials before you paste anything.

If searching for prompts is the hard part, use BananaNL

Prompts become useful when they are close to the input field. Use BananaNL to carry them there, then adjust before sending.