Prompt details

Tested on

Claude Opus 5.5

Length

479 words · 3 fill-ins

Example input

[Requirements or user stories]: Checkout for the Fernway Outdoor online store, release 4.2. US-1 Guest checkout: a shopper can check out without an account, giving email, shipping address and payment. After the order, they are offered the option to create an account. US-2 Saved cards (new): a signed-in customer can save a card at checkout and choose a saved card next time. They can remove a saved card under Account. Expired saved cards are shown but cannot be selected. US-3 Coupons: one coupon code field, one coupon per order. A coupon is either percent off or a fixed amount off. A coupon can have a minimum order subtotal, an expiry date, and can exclude sale items. A fixed-amount coupon cannot take the order total below $0.00. US-4 Order total: subtotal minus coupon discount, plus shipping, plus tax. Shipping is free on orders over $75 after the discount. Tax comes from the tax service, based on the shipping address. US-5 Payment: cards through our payment provider, with 3-D Secure when the card's bank asks for it. On a decline, show a clear error and keep the cart. US-6 Confirmation: an order confirmation page, and a confirmation email within 5 minutes.

[Platforms and environments]: Web: latest Chrome, Safari, Firefox and Edge on desktop; Safari on iPhone and Chrome on Android for mobile web. Apps: iOS 17 and later, Android 12 and later (app version 4.2). Staging environment with the payment provider's sandbox and the tax service in test mode.

[Risky areas and past bugs]: In release 4.1, 6 customers were charged twice after tapping Place order twice on a slow mobile connection. In February the "exclude sale items" rule broke and coupons were applied to sale items. Saved cards are new in this release. Guest checkout screens were redesigned in this release. Apple Pay and Google Pay are out of scope for 4.2.

Example output

1. Scope and risks

In scope

US-1 to US-6 on staging, using the payment provider's sandbox and the tax service in test mode.
Interruptions during checkout: double submit, page refresh, lost network, app sent to the background, and session timeout.
Platform keys used in the cases:

- Web: latest desktop Chrome, Safari, Firefox and Edge.

- mWeb: Safari on iPhone and Chrome on Android.

- iOS: app 4.2 on iOS 17 and the latest iOS.

- Android: app 4.2 on Android 12 and the latest Android.

- All: every platform above.

Assumption: the source of truth for charges is the sandbox dashboard. The source of truth for orders is Account > Orders, or the sandbox dashboard for guests.
Assumption: "one Place order = at most one order and one charge" comes from the 4.1 incident. It is not written in US-5 (see Q1).

Out of scope

Apple Pay and Google Pay.
Payment methods other than cards. Assumption: 4.2 has no other methods.
Whether the tax service's rates are correct. We check only that its result is applied.
Live payment processing, and load or performance testing.
App versions below 4.2, iOS below 17, and Android below 12.
Accessibility and localization. Assumption: no requirements were given for these.

Area	Risk	Why	Case IDs
Duplicate submission and payment interruptions	High	6 customers were double-charged in 4.1 on slow mobile connections. This is a money and trust issue.	TC-01, TC-02, TC-03
Coupon "exclude sale items" rule	High	The rule broke in February, and coupons were applied to sale items, which loses money.	TC-04, TC-05, TC-06
Saved cards	High	The feature is new in 4.2 and stores payment data. A wrong card shown, or a removed card that is still usable, damages trust.	TC-07 to TC-12
Payment declines and 3-D Secure	High	Money path. A lost cart or a silent failure means lost orders.	TC-13 to TC-17
Order total: coupon rules, free shipping, tax	High	Every order uses this calculation. Boundary errors over- or under-charge customers.	TC-18 to TC-26

Claude Opus 5.5

Test Plan and Test Cases – QA test plan prompt for ChatGPT and Claude

^$9.99

Kept in your account
Copy filled in or download
Updates included

Pay once and keep it. Or get every premium prompt and skill with the plan, $19.99 a month. Payments are non-refundable. See our Terms.

votes views

Claude Opus 5.5

This prompt writes a test plan and manual test cases from your requirements. Paste the stories, platforms and past bugs; you get ranked risks, 20 to 30 cases with steps and data, a regression list and questions.

What you get

A scope and risk table that puts past bugs and money paths first
20 to 30 manual test cases with preconditions, exact steps, test data, one expected result and a priority
A regression checklist and the test data to set up before the cycle
Numbered questions wherever the requirements are silent, tied to the cases they block

How to use it

1
Paste the requirements and the history
Add the stories as written, the platforms you support and every past bug in this area. The bugs decide where the most cases go.
2
Read the risk table first
Check that the areas you worry about rank high and have cases, then read the P1 cases.
3
Answer the questions before the cycle
Send each question to its owner, paste the answers back and ask for the Unclear cases rewritten with expected results.

Better results

Paste acceptance criteria word for word. A threshold such as "over $75" only becomes a $75.00 and $75.01 boundary case when the exact words are there.
Follow up with "Give me the test cases as CSV with these columns" and list the import columns of your test management tool.
After the cycle, paste the failures back and ask which checks to add to the regression list for the next release.

Questions

How do I use Claude for QA testing?

Give Claude your requirements or user stories, the platforms you test and the past bugs in the area, and ask for a test plan with manual cases. It writes steps, data and expected results, and raises questions where the requirements are silent. Claude cannot run the tests on your app; your testers do.

What is the best AI for QA testing?

For writing test plans and manual cases, a capable chat assistant such as Claude, ChatGPT or Gemini works well from pasted requirements. For running tests, tools such as testRigor, Momentic or QA.tech automate them. Many teams use both: an assistant to plan, a test tool to run.

Can AI write test cases from user stories?

Yes, AI writes solid test cases from user stories when the stories include acceptance criteria and real values. It covers boundaries, negative paths and interruptions quickly. Where a story is silent, a good prompt makes it ask instead of guessing the expected result.

What should a test plan include?

A test plan should include the scope and what is out of scope, risks ranked by impact, test cases with priorities, the platforms and environments, the test data to prepare, a regression list and open questions. This prompt writes all of them, in that order.

Updated 8 Nov 2026