How to Use AI to Write Acceptance Criteria for User Stories

You’ve written the user story, but now you need acceptance criteria that your team can actually test. Writing clear, specific criteria from scratch takes time—and the more stories in your backlog, the more tedious it becomes.

AI tools like ChatGPT, Claude, or Notion AI can draft acceptance criteria in seconds once you know how to prompt them. This guide shows you exactly how to do it, with tested prompts and tips to keep the output useful.

Why Use AI to Write Acceptance Criteria?

Acceptance criteria define what “done” looks like for a user story. They guide developers, testers, and stakeholders, so everyone knows when a feature is ready to ship.

Writing them manually isn’t hard, but it’s repetitive. AI speeds up the first draft, freeing you to focus on edge cases and business logic. You still review and refine—AI just handles the boilerplate.

If you’re already using AI to generate user stories, adding acceptance criteria to your workflow is a natural next step. The two tasks pair well because the AI already understands your feature context.

What Makes Good Acceptance Criteria?

Before you prompt an AI, know what you’re aiming for. Good acceptance criteria are:

• Testable – A QA tester or developer can confirm whether each criterion is met
• Specific – No vague terms like “fast” or “user-friendly” without defining them
• Independent – Each criterion can be checked on its own
• Focused on outcomes – They describe what happens, not how to build it

Most teams use either Given-When-Then format (from Behavior-Driven Development) or a simple checklist. Pick the format your team already uses, or try both and see which fits your workflow.

How to Prompt AI to Write Acceptance Criteria

A basic prompt works, but a structured one gives you better results. Include the user story, any relevant context, and the format you want.

Basic Prompt Template

Copy this prompt and replace the bracketed sections with your details:

Write acceptance criteria for this user story:
[Paste your user story here]

Context: [Brief description of the feature or product area]

Format: [Given-When-Then / Checklist]

Example:

Write acceptance criteria for this user story:
As a project manager, I want to filter tasks by assignee so I can see what each team member is working on.

Context: This is a task board in a project management tool. Users can assign tasks to team members, and tasks have statuses (To Do, In Progress, Done).

Format: Given-When-Then

The AI will generate criteria like:

• Given I am on the task board, When I select a team member from the filter dropdown, Then only tasks assigned to that person are displayed
• Given no filter is applied, When I view the task board, Then all tasks are visible
• Given I have applied a filter, When I clear the filter, Then all tasks are displayed again

Advanced Prompt: Add Constraints and Standards

If your team has conventions—like always including error states or accessibility checks—add them to the prompt.

Write acceptance criteria for this user story:
[User story]

Context: [Feature area]

Format: Checklist

Include:
– Happy path (normal use)
– Error handling
– Edge cases (e.g., empty states, maximum limits)
– Accessibility (keyboard navigation, screen reader support)

This prompts the AI to think beyond the obvious scenario. You’ll get criteria that cover what happens when a user enters invalid data, hits an API limit, or navigates without a mouse.

For more on structuring effective prompts, check out how to write better prompts for AI tools in business.

Step-by-Step: Using ChatGPT to Write Acceptance Criteria

Here’s a worked example using ChatGPT. The same approach works in Claude, Notion AI, or any conversational AI.

Step 1: Open ChatGPT and paste your user story. Include the context you’d normally give a teammate—product area, related features, or any constraints.

Step 2: Specify the format. Ask for Given-When-Then, a checklist, or both. If you’re unsure, start with a checklist—it’s easier to scan.

Step 3: Review the output for testability. Look for criteria that are too vague (“The UI should be intuitive”) or too technical (“The backend should use a Redis cache”). Rewrite or ask the AI to clarify.

Step 4: Ask follow-up questions to refine. If the AI missed edge cases, prompt it: “What about when the user has no tasks assigned?” or “Add criteria for mobile screen sizes.”

Step 5: Copy the final criteria into your backlog tool. Paste them into Jira, Linear, Notion, or wherever your team tracks stories. Tweak formatting if needed.

Example Prompts for Common User Story Types

Different story types need different criteria. Here are tested prompts for common scenarios.

Form Submission

Write acceptance criteria for a contact form user story. Include validation (required fields, email format), success and error messages, and what happens after submission. Format: Checklist.

Data Display or Dashboard

Write acceptance criteria for a dashboard that shows sales metrics. Include data accuracy, loading states, empty states, and refresh behaviour. Format: Given-When-Then.

User Authentication

Write acceptance criteria for a login feature. Cover successful login, incorrect password, account lockout after failed attempts, and password reset link. Format: Checklist.

API Integration

Write acceptance criteria for integrating a payment gateway. Include successful payment, failed payment, timeout handling, and confirmation email. Format: Given-When-Then.

These prompts give the AI enough structure to generate useful criteria without needing you to spell out every detail.

How to Review and Refine AI-Generated Criteria

AI output is a first draft, not the final version. Here’s what to check before adding criteria to your backlog.

Is Each Criterion Testable?

Ask yourself: can a tester confirm this is done? If a criterion says “The user has a good experience,” rewrite it: “The form submits within 2 seconds on a standard connection.”

Are There Missing Edge Cases?

AI often covers the happy path but misses unusual scenarios. Think about what happens with empty data, maximum values, or when external services fail. Add those manually or prompt the AI again.

Is the Language Clear?

Remove jargon or overly technical phrasing. Your criteria should make sense to developers, testers, and non-technical stakeholders. If you need to explain a term, simplify it.

Does It Match Your Team’s Standards?

Check that the format and level of detail match what your team normally writes. If your team always includes accessibility or performance benchmarks, make sure the AI did too. If not, add a note to your prompt template for next time.

Understanding when a user story is “done” helps you spot gaps in acceptance criteria—if your Definition of Done includes security checks or documentation, your criteria should reflect that.

Which AI Tools Work Best for Acceptance Criteria?

You don’t need a specialised tool. Most general-purpose AI assistants handle this well if you prompt them clearly.

ChatGPT (OpenAI)

Works well for structured tasks like acceptance criteria. The free tier is fine for occasional use; the paid plan (ChatGPT Plus) gives faster responses and access to GPT-4, which handles complex prompts better.

Claude (Anthropic)

Similar to ChatGPT, with a slightly different tone. Some users find Claude better at following multi-step instructions. Free tier available; paid tier increases usage limits.

Notion AI

Built into Notion, so you can generate criteria directly in your project database. Convenient if your backlog lives in Notion. Requires a Notion AI add-on subscription (pricing varies by plan).

If you’re deciding between Notion AI and ChatGPT for project management, the key difference is where you work: Notion AI integrates with your workspace, while ChatGPT requires copy-pasting.

AI Tools for Scrum Masters

If you’re a Scrum Master juggling multiple backlogs, consider tools that streamline your entire workflow. Our guide to AI tools for Scrum Masters covers options that go beyond just writing criteria.

How to Store and Reuse Your Best Prompts

Once you’ve refined a prompt that works, save it so your team can reuse it. Consistency matters—if everyone on your team prompts AI differently, you’ll get inconsistent results.

Create a prompt library. Store your tested prompts in a shared document, wiki, or database. Include the prompt text, example output, and notes on when to use it.

Use a Notion database for prompts. If your team uses Notion, you can build an AI prompt database where everyone adds prompts, tags them by use case, and rates their effectiveness.

Track what works. Note which prompts generate the best criteria, and refine them over time. If you find yourself always adding the same follow-up question, bake it into the original prompt.

If you want a ready-made system, the AI Project Command Center includes a prompt library with 20+ ready-to-use prompts for project management tasks, including user stories and acceptance criteria. It also helps you track which AI tools you’re paying for and whether they’re worth the cost.

Common Mistakes and How to Avoid Them

Accepting AI Output Without Review

The biggest mistake is trusting AI-generated criteria without checking them. Always read through and test for clarity, completeness, and testability. AI can miss context only you know.

Prompts That Are Too Vague

A prompt like “Write acceptance criteria” without context gives generic results. Always include the user story, feature area, and format. The more context, the better the output.

Overcomplicating Criteria

Sometimes AI generates overly detailed criteria that cover implementation details. Your acceptance criteria should focus on what the feature does, not how it’s built. Simplify anything too technical.

Not Involving the Team

Don’t let AI replace collaboration. Use it to draft criteria, then review them with your team—developers, testers, and product owners. They’ll spot gaps and add insights the AI can’t.

Integrating AI Criteria into Your Backlog Workflow

Once you’ve generated criteria, add them to your backlog tool. Here’s how to make that smooth.

Copy criteria into your task management tool. Paste them into Jira, Linear, Asana, or wherever you track stories. Adjust formatting (bullets, numbering) to match your team’s style.

Link criteria to the story. Keep criteria visible to everyone—developers, testers, and stakeholders. Don’t hide them in comments or attachments.

Update criteria as the story evolves. If the team discovers new edge cases during development, add them. Acceptance criteria aren’t set in stone; they’re a living part of the story.

Review criteria in refinement or planning. Walk through the criteria with your team before committing to the story. This catches ambiguities early and ensures everyone agrees on what “done” means.

Combining AI with Your Existing Backlog Process

AI fits into your workflow without replacing it. Here’s how it complements the work you’re already doing.

If you’re already generating user stories with AI, adding acceptance criteria is the next logical step. The AI already understands your feature context, so it can draft criteria faster. Just make sure the story itself is clear—vague stories produce vague criteria.

You can also use AI to help with related tasks. For example, AI can help prioritise your task list using frameworks like MoSCoW or RICE, so you know which stories to write criteria for first.

If your team runs sprints, consider using AI tools for sprint retrospectives to identify patterns in which criteria were unclear or missed. That feedback helps you write better criteria next time.

When Not to Use AI for Acceptance Criteria

AI works well for straightforward stories, but some situations need human judgment.

Complex business logic. If your feature involves intricate rules, regulatory requirements, or domain-specific knowledge, AI won’t have the context. Draft the criteria yourself or with a subject-matter expert, then use AI to expand or format them.

High-risk features. For security-critical, compliance-heavy, or financial features, have a human write and review criteria carefully. AI can assist, but don’t rely on it alone.

Brand-new product areas. If you’re exploring a new feature with uncertain requirements, collaborate with your team first. AI works best when the direction is clear.

Tips for Teams Adopting AI for Acceptance Criteria

Start with one story type. Don’t overhaul your entire process at once. Try AI for form submissions or CRUD operations first, then expand to more complex stories.

Share successful prompts. When someone finds a prompt that works well, add it to your team’s shared library. This builds consistency and saves everyone time.

Set a review standard. Agree as a team that all AI-generated criteria must be reviewed by a human before going into the backlog. This keeps quality high and builds trust in the process.

Iterate on your prompts. Track which prompts produce the best criteria and refine them. If you notice the AI always misses accessibility checks, add that to your template.

Combine AI with team expertise. Use AI to draft criteria, then refine them in refinement sessions. The team adds context, edge cases, and business logic the AI can’t know.

For more ways to streamline your workflow, explore the full AI Tools category for tested guides on using AI in business.

Frequently asked questions

Can AI write acceptance criteria without a user story?

AI needs context to write useful criteria. If you don't have a formal user story, describe the feature, who will use it, and what problem it solves. The more detail you provide, the better the output. A vague prompt produces vague criteria.

Should I use Given-When-Then or checklist format for acceptance criteria?

Both work. Given-When-Then (from Behavior-Driven Development) is more structured and works well for complex logic. Checklists are faster to scan and suit simpler stories. Pick the format your team already uses, or try both and see which fits your workflow.

How do I stop AI from writing overly technical acceptance criteria?

Add a note to your prompt: 'Focus on what the feature does for the user, not how it's built.' If the output is still too technical, ask the AI to simplify or rewrite for a non-technical audience. Review and remove implementation details before adding criteria to your backlog.

Can I use AI to write acceptance criteria for non-functional requirements?

Yes, but you need to be specific. Instead of 'The page should be fast,' tell the AI: 'Write acceptance criteria for page load performance. Target: under 2 seconds on a standard connection.' Provide benchmarks, standards, or thresholds so the criteria are testable.

Do I need a paid AI tool to write acceptance criteria?

No. Free tiers of ChatGPT, Claude, and other AI assistants work fine for this task. Paid plans offer faster responses, higher usage limits, and access to more advanced models, but they're not required unless you're generating criteria at high volume.

How do I ensure acceptance criteria generated by AI match my team's Definition of Done?

Add your Definition of Done to the prompt. For example: 'Include accessibility checks and mobile responsiveness in every set of criteria.' Review the output to confirm it covers all required areas. If the AI misses something, add it manually and update your prompt template for next time.