Guide
AI-drafted Gherkin scenarios from a Jira story: what is sent, and how to review
By Miftode Vasilica, maker of Gherkin Studio ·
An AI model can turn a Jira story into a first draft of Gherkin scenarios in a few seconds. Before you switch that on, two questions matter more than the quality of the draft: what exactly leaves Jira, and who reads the draft before it becomes the team's agreement. This page answers both for Gherkin Studio, and gives a short review routine that works with any tool.
The empty panel
A story has a title, two lines of description and no scenarios. Refinement is in an hour. The tester knows that an empty page gets filled with the happy path and nothing else, because the happy path is what everyone already has in mind.
A draft changes the job. The team stops inventing cases and starts crossing out the wrong ones. That is quicker, and it surfaces the cases nobody would have raised: the expired link, the link used twice, the user who is not allowed.
It is also how wrong scenarios get in. A draft that reads well is easy to accept without reading.
What is sent, exactly
In Gherkin Studio, AI drafts are off by default. Nothing is sent to anyone while the setting is off.
A Jira admin switches it on in Apps → Gherkin Studio and adds your organisation's own Anthropic API key. From then on, when someone clicks AI draft on a story, this is sent to Anthropic:
- the story's key, summary and description;
- the scenarios already written on that story, so the draft adds to them and does not repeat them;
- the wording of steps commonly used in the project, at most 60, so the draft reuses them.
Nothing else. The story's comments, attachments, reporter and assignee are not sent, and neither are other stories.
Very long descriptions are cut at 30,000 characters, and the panel tells you when that happened.
Because the key is yours, the usage is billed by Anthropic to your organisation, under your organisation's agreement with them. Ask your security team before you add the key. That conversation is easier with the list above in hand.
What comes back
The model is asked for the main path first, then the cases people forget: empty and maximum inputs, invalid data, permissions, duplicates, time limits, failures of a network or a service. Each scenario is tagged with the story's key.
The draft is checked with the same Gherkin parser as the editor before you see it.
The draft is shown first. You add it or discard it. Nothing is written to the story until a person presses the button. If the answer was cut off for length, the panel says so.
A five-minute review
Read the draft with the story open next to it.
- Check every number. "61 minutes", "5 attempts", "10 MB". The model picks plausible values. Plausible is not agreed. Each number is a question for the product owner.
- Delete what the story does not ask for. Drafts over-cover. A scenario for a rule nobody decided is a requirement that slipped in through the test.
- Look for the comment at the top. When a story is too vague to test, the model is asked to say what is unclear in a
#comment. That comment is often the most useful line in the draft: take it to refinement. - Compare step wording. If the draft says "I open the reset link" and your suite says "I follow the reset link", change it. One wording, one step definition.
- Say the scenarios out loud in refinement. If the product owner hesitates, the scenario is not agreed yet.
What you add after this is the team's scenario, not the model's.
What an AI draft is not
- Not a test. It is text. Somebody still automates the steps.
- Not a decision. The model does not know your rule for expired links. It guesses, and it is good at guessing.
- Not a replacement for the conversation. It gives the conversation something to disagree with.
Short answers
Is my Jira data used to train the model?
That is governed by your organisation's agreement with Anthropic, because the key is yours. Gherkin Studio sends only what is listed above and only on a click.
Can I use AI drafts without sending anything outside Atlassian?
No. Any AI draft means the story text goes to a model provider. If that is not acceptable, leave the setting off. Every other feature works without it.
Who can click AI draft?
Anyone who can edit the issue, once a Jira admin has switched the feature on.
Does it cost extra?
Gherkin Studio does not charge for it. Anthropic bills the usage to your own API key.
Where Gherkin Studio fits
Gherkin Studio is a BDD app for Jira Cloud, built on Atlassian Forge. It is on the Atlassian Marketplace, with a 30-day free trial. Read the AI drafts section of the docs and the privacy policy.