AI is becoming a key QA tool, aiding in faster scenario generation, risk detection, and test planning. At the Online TestConf, Arbaz Surti showed how effective prompting using roles, context, and output format helps to get clear, relevant, and actionable test scenarios. AI can boost testers, but human judgment is needed to ensure relevance and quality.
AI is going to be part of the QA toolkit, and testers who learn how to shape prompts effectively will be the ones using it to speed up scenario generation, uncover risks faster, and focus their time on higher-value work, Surti explained:
In my Baskin-Robbins project, we had to test how menu item availability synced to the ordering system. I gave the AI a detailed prompt explaining the application and asked it to generate edge cases. It came up with several scenarios I hadn’t considered — including one where an item marked as “available” in the app could actually be out of stock in the store.
Catching that prevented a serious bug that would have led to customers placing unfulfillable orders and stores dealing with messy refunds, Surti said.
Prompt engineering is an extension of what we already do best: asking good questions. If you give AI a vague prompt, you’ll usually get a vague answer, Surti said. If you apply the same discipline testers bring to writing test cases, you get results that are much closer to what testers actually need: clear, relevant, and actionable test scenarios they can trust.
Writing effective prompts for test cases is like writing clear requirements; the more context you give, the better the output. He suggested using a simple structure:
- Define the role: you are a senior QA engineer, a performance tester focused on load and stress scenarios, a security tester skilled in identifying vulnerabilities in authentication flows, or a QA lead responsible for risk-based prioritization.
- Add context: what feature or system are you testing, any rules or constraints.
- Set the format you want the output in.
When you do this, instead of a random list of test cases, you’ll often get structured, prioritized scenarios that you can plug directly into your QA workflow, Surti said. It’s about guiding the AI the same way you would guide a junior tester.
AI-powered testing workflows empower testers instead of replacing them. AI plugs into different parts of the QA cycle, Surti explained:
During test planning, it can help you brainstorm coverage faster. For test case creation, it can structure scenarios and even suggest edge cases you might miss. When you hit execution and automation, AI can assist with generating scripts or prioritizing high-value tests. And in reporting, it can summarize results, spot patterns, or translate technical findings into business-friendly language.
The workflow is still the same, but AI acts like a smart assistant at each step, helping testers move quickly, and focus on higher-value thinking, Surti said.
For prompting in QA, Surti suggested focusing on guiding principles rather than specific steps:
- Clarity over completeness: prioritize asking for exactly what I need rather than trying to get everything in one prompt.
- Format for reuse: structure outputs so they can plug directly into test plans, tables, or checklists.
- Think like a tester: treat AI responses as something to validate and iterate on, just like test results.
- Risk-aware prompting: steer the AI to focus on high-impact or edge-case scenarios rather than generic answers.
- Reflective iteration: each prompt is an opportunity to refine and learn, improving results over time.
AI is a powerful accelerator, not a replacement for testers, Surti said. It shines at brainstorming, generating scenarios, or surfacing edge cases you might miss under time pressure. But its real value comes when you guide it with good prompts and then apply human judgment to filter what matters. AI helps you cover more ground, faster, but it’s the tester who ensures the depth and relevance of that coverage, he concluded.
InfoQ interviewed Arbaz Surti about prompting for software testing.
InfoQ: How can we write effective prompts to generate test cases?
Arbaz Surti: Here’s an example prompt with the output that is generated:
Role: You are a senior QA engineer experienced in e-commerce checkout testing.
Context: The feature under test is the e-commerce checkout flow for a web app. It must support guest and authenticated checkout, multiple payment methods (credit card, gift card, PayPal), promo codes, multi-device continuation, shipping options, and tax calculation. The system enforces payment validation and must resist common security attacks (SQLi, XSS). There is a 15-minute session/checkout timeout.
Constraints: Produce a Markdown table with these columns:
| Test Scenario | Category (Functional / Edge Case / Security / Usability) | Input / Action | Expected Outcome | Priority (High / Medium / Low) |
Requirements:
– Provide 15–20 distinct scenarios covering Functional, Edge Case, Security, and Usability.
– Include core happy paths, negative/payment failure cases, edge cases (empty cart, timeouts, multi-device), and security tests (SQL injection, XSS).
…
Output only the Markdown table (no extra commentary).
Result:
Test Scenario
Category
Input / Action
Expected Outcome
Priority
Add item to cart
Functional
Select an item and click “Add to Cart”
Item appears in cart with correct price
High
Apply discount code
Functional
Enter valid/invalid coupon
Valid: discount applied; Invalid: error message
High
Login during checkout
Functional
Login mid-checkout
Cart retained, user redirected to checkout
Medium
…
InfoQ: Which prompting techniques exist, and how do you use these techniques?
Surti: Techniques taken from A Systematic Survey of Prompt Engineering Techniques that give testers the most impact are:
- Role prompting helps shape the AI’s “voice” and expertise – for example, making it act like a senior QA engineer.
- Decomposition is about breaking big tasks into smaller, more testable steps.
- Structured outputs are key too, because when you ask for results in a table, JSON, or a test charter, you get something you can plug into your workflow.
- And you can ask AI to self-evaluate its coverage, which helps catch gaps.
There are more advanced techniques like Chain-of-Thought prompting (CoT), where you guide the AI to reason step by step. But for most testers, starting with these core techniques can already transform the quality of the output.
