How to write acceptance criteria that people can verify

Dylan de Heer
To write useful acceptance criteria, describe the observable conditions a piece of work must satisfy for the appropriate person to accept it. Start from the user's intended outcome, make each condition verifiable, discuss important exceptions and agree how the result will be checked.
“Make the export useful” is a request for an outcome. “The exported file includes the records selected by the active filter and contains the agreed columns” is closer to a condition someone can inspect. The remaining details still need agreement: which filter, which columns, which permissions and what should happen when there are no records?
This guide shows how to turn that discussion into reviewable criteria. It uses a fictional product example, but the method also helps with internal services and other bounded deliverables. It does not require a particular template or tool.
What acceptance criteria are for
Acceptance criteria connect an intended outcome to a decision about the completed work. They clarify what must be true, rather than listing every implementation task needed to make it true.
Scrum Alliance describes acceptance criteria as clear, concise, verifiable conditions and distinguishes them from the Scrum framework itself: teams may use them, but they are not a required element of the Scrum Guide. Scrum Alliance: acceptance criteria.
A condition can be met or unmet for a particular check. If the work meets three of four required conditions, record that detail; do not call the fourth condition “mostly passed”. The team can then decide what the remaining issue means for acceptance.
Criteria are not evidence that something works. That evidence comes from an appropriate test, inspection, demonstration or review. Writing a clear criterion and verifying it are separate pieces of work.
Before you start: identify the request and the decision owner
Gather the request, the relevant user or stakeholder context, known constraints and the person or role who can confirm the intended outcome. Include someone who understands delivery and someone who can challenge how the result will be checked. In a small team, one person may hold several roles.
You do not need a polished user story before discussing criteria. You do need a bounded piece of work and an explanation of why it matters. If the underlying need is still unclear, begin with requirements elicitation rather than polishing uncertain conditions into authoritative prose.
Keep unknowns visible. A missing business rule is a question for its owner, not an invitation to choose a plausible threshold. Note the relevant source, version or discussion so reviewers can return to the reason behind a condition.
1. State the outcome in plain language
Write one short description of what the user should be able to accomplish and why. Avoid starting with a list of buttons, database tables or development tasks.
For our fictional example: “An account coordinator needs to export the requests currently included in a filtered view so they can review that same set of requests outside the system.”
That statement creates a useful boundary. The work concerns exporting a selected set, not building an entire reporting service. It does not yet decide the file type, columns or permission rules. Those are details to clarify with the people responsible.
Ask what would make the result useful and what would make it misleading. In this example, an export containing hidden records or missing the chosen date range could undermine the review even if the file downloads successfully.
2. Turn expectations into observable conditions
Take each important expectation and describe what someone can inspect. Replace an adjective with a meaningful condition or a question that still needs answering.
Initial wording | Better direction | What still needs agreement? |
|---|---|---|
The export must be accurate | Exported records match the agreed filtered set | Which filters apply and how is the expected set established? |
It should be easy to read | The file includes the agreed fields with recognisable column headings | Which fields, headings and intended reader? |
It must be secure | An unauthorised user cannot export records outside their permitted scope | Which roles and rules determine access? |
It should be fast | Complete the export within an agreed limit under stated conditions | Which limit, data size and environment are appropriate? |
The table is a writing exercise, not a specification for an existing product. Do not copy its security or performance conditions into a real backlog without the responsible review. A precise invented number is still an invented requirement.
Atlassian's guidance similarly emphasises outcomes, clarity and testability. Use those as review questions rather than treating a neat sentence as proof that every stakeholder means the same thing. Atlassian: writing acceptance criteria.
3. Check the ordinary case and meaningful exceptions
Walk through an ordinary successful case, then the situations most likely to change the result or expose an important risk.
For the export, useful questions include: what happens with no matching records, who may export, whether hidden fields are included and whether records can change between the displayed view and the export. Ask which of these matters within the agreed scope. The answers may reveal a separate piece of work rather than another condition to append.
You cannot document every possible future situation before starting. Prioritise the cases that affect the intended outcome, known constraints or a consequential failure. Leave a clear route for discussing newly discovered cases. Acceptance criteria should support shared understanding, not become a substitute for collaboration or engineering judgement.
If two stakeholders disagree, preserve the competing interpretations and ask the appropriate owner to resolve them. A draft that conceals disagreement is harder to review than one that marks the decision as open.
4. Choose the simplest format that makes the check clear
A short list works well for independent conditions. Scenario wording is useful when behaviour depends on a particular context and action. Scrum Alliance describes both list-based approaches and the familiar Given–When–Then structure; neither format makes unclear content correct. Scrum Alliance: formats and examples.
For the fictional export, a scenario might read:
Given an authorised coordinator is viewing the agreed filtered set of requests, when they export that set, then the file contains those records and the agreed columns.
This is a useful starting point, but it may combine several checks. Review whether record selection, field inclusion and file readability should be separate criteria so that a failed check is easy to diagnose.
Use the words only when the structure helps. A clear sentence such as “The export contains the agreed columns in the agreed order” can be easier to review than a forced three-part scenario.
5. Work through one complete example
Assume the fictional team has now agreed that the first release exports a CSV for authorised coordinators. It has also agreed the columns and the rule that only the records in the active filtered set should be included. These are invented example decisions, not Weeve capabilities or universal defaults.
A bounded set of draft criteria could be:
For an authorised coordinator, exporting the agreed filtered view produces a readable CSV containing the same request identifiers as the expected filtered set.
The file contains the agreed columns: request identifier, request title, status and assigned owner. It does not include the internal-comment field.
When the filtered set is empty, the coordinator sees the agreed empty-state message and no misleading populated file is produced.
A user without the agreed export permission cannot initiate the export through the supported interface.
Now discuss how to check them. Establish a small, authorised example set with known identifiers and values. Compare the file with that expected set. Inspect the columns, exercise the empty case and check the permission behaviour in the appropriate environment.
Do not treat a disabled button as proof that access control is complete. The team's security and technical reviews may require other checks. Similarly, the example's “same set” wording needs a decision about changing data if that situation matters to the intended workflow. Record the decision before presenting the criterion as settled.
The useful output is the criteria plus their agreed interpretation and review method. A longer list is not automatically a better one.

6. Review the wording with the people who will use it
Ask a stakeholder to explain what a successful result looks like, a delivery colleague to describe any ambiguity and a reviewer to explain how they would verify the condition. Different answers reveal where the wording still needs work.
Check whether each condition:
Supports the stated outcome and belongs within this piece of work.
Describes a result rather than an unnecessary implementation choice.
Can be verified using known inputs and an agreed expected result.
Identifies a consequential exception without hiding an unresolved rule.
Can be traced to an agreed need, constraint or explicit decision.
If a detail is a real constraint, keep it even when it affects implementation. For example, an agreed interoperability requirement may legitimately specify a file format. The point is to distinguish a necessary condition from an arbitrary design choice.
Acceptance criteria, Definition of Done and test cases
These concepts answer related but different questions.
Acceptance criteria describe conditions specific to the work being discussed. Test cases describe concrete checks, inputs and expected results used to verify behaviour. One criterion may need several test cases.
In Scrum, the Definition of Done describes the quality state required for an Increment. It creates a shared understanding of when work meets those quality measures; it is not simply a replacement name for the conditions on one backlog item. The Scrum Guide: Definition of Done.
For the fictional export, matching the selected records is an item-specific acceptance condition. Shared quality requirements may also apply to the change. Passing one export demonstration does not establish that every applicable quality requirement has been met.

Recover the context from conversations without inventing agreement
Important details often sit in discovery or review conversations: a stakeholder's reason for wanting a field, a permission boundary or an example of an unwanted result. Revisit those sources when the written request leaves uncertainty.
In Weeve, select the relevant saved recordings before asking for proposed criteria. A focused request is: “Identify the outcomes and constraints discussed in these recordings. Propose verifiable acceptance criteria with source passages. Separate explicit agreements, suggestions, contradictions and unanswered questions.” Chat with recordings explains the supported workflow.
Review the cited material yourself and confirm the wording with the responsible people. A transcript may be wrong, a summary may omit context and a suggested criterion may sound firmer than the conversation supports. Weeve can help prepare a draft; it does not run the checks or decide whether the product should be accepted. Use the appropriate local or optional external processing mode for your authorised source material.
Keep a route for clarification and change
Save the agreed wording beside the work it describes, along with unresolved questions and the basis for important decisions. When a new case appears, establish whether it clarifies an existing condition or changes the requested outcome. Discuss consequences for scope and delivery before silently adding it to the acceptance bar. Our scope-creep guide covers that wider conversation.
Start by taking one vague request and asking three questions: what should be observable, how will we check it, and who can confirm that this is the right expectation? Write the first criteria, discuss the uncertainties and verify the result separately from the wording.



