ReqConsider an electric vehicle manufacturer developing a battery charging feature. The Battery Management System (BMS) must monitor the battery's State of Charge (SoC), terminate charging when the configured maximum level is reached, and notify the driver.
What happens if the SoC sensor provides invalid readings? What if the battery temperature exceeds its permitted operating range? What happens if communication between the BMS and charging controller is interrupted?
An experienced tester can identify these scenarios through requirement analysis and domain knowledge. An AI assistant can also analyze the requirement and suggest relevant test cases within seconds.
But are those test cases complete? Do they reflect the actual system requirements? Can the QA team approve them without further review?
This is where AI-assisted software testing becomes valuable.
AI can help testers generate test scenarios, develop automation scripts, investigate defects, and prioritize regression testing. However, testing knowledge, domain understanding, and professional judgment remain essential.
This article explores practical applications of AI in software testing, its benefits and limitations, and how QA professionals can adapt to an AI-assisted testing environment.
Key Takeaways
- AI can accelerate test case generation, automation, defect analysis, and regression testing.
- AI-assisted requirement analysis can identify potential gaps and ambiguities.
- AI-generated test cases and scripts require human validation.
- Testing AI-powered applications introduces additional challenges involving accuracy, privacy, consistency, and safety.
- The future of QA depends on combining AI capabilities with engineering knowledge and professional judgment.
Table of Contents
1. How AI is Changing Software Testing
Traditional software testing involves understanding requirements, identifying test scenarios, preparing test cases, executing tests, reporting defects, and performing regression testing.
Many of these activities require considerable time and repeated effort.
AI introduces additional capabilities into existing testing processes. Testers can use AI to analyze requirements, suggest scenarios, generate initial test cases, explain automation scripts, and summarize defect information.
Consider an e-commerce application introducing a feature that allows customers to combine promotional offers during checkout.
An AI assistant can suggest scenarios such as:
- Applying multiple eligible promotional codes.
- Applying an expired promotional code.
- Combining discounts with products already on sale.
- Removing an item after applying a discount.
- Modifying an order after the discount has been calculated.
Traditional testing versus AI-assisted testing
| QA Activity | Traditional Approach | AI-Assisted Approach |
|---|---|---|
| Requirement analysis | Manual review | AI identifies potential gaps |
| Test design | Tester develops scenarios | AI proposes initial scenarios |
| Test case review | Peer or lead review | AI-assisted review |
| Automation | Manual script development | AI-assisted script generation |
| Defect reporting | Manual preparation | AI structures observations |
| Regression testing | Existing impact analysis and selection rules | AI-assisted prioritization |
AI does not necessarily replace existing QA processes. It helps reduce repetitive effort and allows testers to focus more on product behavior, risk analysis, and test coverage.
2. AI-Assisted Test Case Generation
Test case preparation is one of the areas where AI can provide immediate practical value.
Testers often spend considerable time translating requirements, user stories, and acceptance criteria into detailed test cases. AI can help generate an initial set of test ideas from these inputs.
However, the quality of AI-generated test cases depends significantly on how the task is presented. A structured prompting approach, such as the ART framework (Act as, Request, Terms), helps testers provide clear instructions and define the expected output.
Example: Battery Charging Control
Consider this hypothetical requirement from an electric vehicle manufacturer.
Requirement:
The Battery Management System (BMS) shall monitor the battery's State of Charge (SoC) during charging and terminate charging when the configured maximum SoC limit is reached.
The system shall also notify the driver when charging is completed.
Using the ART Framework to Generate Test Cases
A tester can provide the requirement to ChatGPT or Claude using the following prompt:
A – Act as
Act as a Senior QA Engineer with expertise in automotive software testing, requirements analysis, and test case design.
R – Request
Analyze the supplied battery charging requirement and generate detailed test cases.
T – Terms
- Include positive, negative, boundary value, and abnormal operating scenarios.
- Identify missing information, ambiguities, and potential requirement gaps.
- Include preconditions, test steps, test data, and expected results.
- Present the test cases in a structured table.
- Do not invent expected system behavior that is not defined in the requirement.
- List assumptions and clarification questions separately.
Illustrative AI-generated test cases
| Test ID | Test Data / Condition | Scenario | Expected Result |
|---|---|---|---|
| TC-01 | Limit 80%, SoC 79% | Charging below maximum limit | Charging continues |
| TC-02 | Limit 80%, SoC reaches 80% | Maximum limit reached | Charging termination is initiated |
| TC-03 | Limit 80%, SoC reads 81% | SoC exceeds configured limit | Verify behavior against the specified overshoot and protection requirements |
| TC-04 | SoC sensor returns invalid data | Invalid sensor reading during charging | Expected behavior cannot be determined from the supplied requirement |
| TC-05 | Charging reaches configured maximum limit | Driver notification | Verify that the driver is notified when charging is completed |
The generated scenarios also expose potential gaps in the original requirement.
For example, it does not define how frequently the SoC should be monitored, how sensor failures should be handled, or what should happen if the charging controller does not acknowledge the termination command.
These questions need clarification from the system engineer or product owner.
Why engineering review is necessary
AI can generate useful scenarios, but it cannot independently determine the correct expected behavior when the approved specifications do not define it.
Testers must validate generated test cases against requirements, business rules, interface specifications, and relevant safety constraints.
A QA Manager's Perspective: Using AI in Test Case Generation and Review
My experience with AI in software quality assurance has been particularly useful across several activities.
I have extensively used ChatGPT and Claude, and occasionally Google Gemini, to:
- Generate test cases from requirements and user stories.
- Review test cases prepared by QA team members.
- Prepare defect reports and QA documentation.
- Generate automation scripts.
One of the most useful applications has been test case generation. With clear requirements and appropriate instructions, AI can produce structured scenarios covering positive conditions, negative cases, boundary values, and alternative workflows.
I have also used AI to review test cases created by team members. Combining AI-assisted reviews with my own testing experience has helped evaluate their quality, completeness, and relevance. In my experience, the generated test cases and review feedback have generally been very good, often requiring very little correction.
Another useful application has been converting raw defect observations into structured reports that developers can understand and investigate.
One important lesson from these experiences is that the quality of AI-generated output depends significantly on the context provided.
AI-generated output is a starting point, not an approved test asset.
3. How AI Fits into a Real QA Workflow
AI becomes more valuable when integrated into the existing QA workflow rather than treated as a separate activity.
Consider how a QA team could use AI while testing a new software feature.
AI-assisted QA workflow
Requirement Analysis → Test Design → Test Case Review → Test Execution → Defect Analysis → QA Evaluation
| Stage | How AI Assists |
|---|---|
| Requirement analysis | Identifies ambiguities, missing conditions, and potential gaps |
| Test design | Generates initial test scenarios and test cases |
| Test case review | Identifies incomplete steps, weak expected results, and duplicate scenarios |
| Test execution | Assists with analyzing logs and unexpected results |
| Defect analysis | Organizes observations and suggests investigation areas |
| QA evaluation | Helps summarize coverage, defects, and execution results |
The QA team validates the suggestions at each stage and remains responsible for test acceptance and release recommendations.
4. AI in Test Automation
Traditionally, automation engineers write scripts using programming languages and testing frameworks. They identify UI elements, define actions, establish assertions, and maintain scripts when application behavior changes.
AI-assisted development tools can help testers:
- Generate automation scripts from natural-language instructions.
- Explain existing automation code.
- Suggest corrections for failing scripts.
- Generate test data and API test scripts.
- Suggest assertions and validation conditions.
For example, a tester might ask an AI coding assistant to create an automated test that starts an EV charging session, monitors the battery SoC, and verifies that charging terminates when the configured maximum limit is reached.
The assistant may generate an initial script using Playwright, Selenium, or Python-based testing tools.
The tester must still validate the test environment, locators, synchronization, assertions, and execution results.
Self-Healing Tests
Some AI-assisted testing platforms support self-healing capabilities.
Suppose an application changes the identifier of a charging-status button. A conventional automation script may fail because its original locator is no longer valid.
A self-healing testing tool may identify an alternative element and attempt to update the locator.
This can reduce maintenance effort, but repairing a locator does not guarantee that the script still validates the intended functionality.
The modified locator, test assertions, and execution results should therefore be reviewed before accepting the repaired test.
5. AI-Powered Defect Analysis and Reporting
Defect analysis often requires testers and developers to examine application logs, error messages, screenshots, reproduction steps, and system behavior.
AI can help organize this information and suggest possible investigation areas.
Consider an EV charging application where charging sessions terminate unexpectedly.
A tester may have access to charging logs, SoC readings, temperature values, communication errors, and charging controller responses.
An AI assistant can help identify patterns and suggest possible relationships between these observations.
Improving defect reporting
Suppose a tester records:
"Charging stopped at 65%. Configured limit was 80%. Temperature normal. Charging controller showed communication timeout."
AI can organize this information into a structured defect report.
| Field | Example |
|---|---|
| Summary | Charging terminates before reaching configured maximum SoC |
| Preconditions | Charging limit configured to 80% |
| Steps | Start charging and monitor SoC |
| Expected result | Charging continues until the configured limit, subject to safety conditions |
| Actual result | Charging terminates at 65% |
| Additional information | Temperature within specified range; communication timeout reported |
This gives developers a clearer starting point for investigation.
However, a possible cause suggested by AI is not a confirmed root cause. Technical investigation and supporting evidence are necessary to establish the actual cause.
6. AI in Regression Testing and Test Prioritization
Regression testing becomes increasingly challenging as software applications grow.
A single application change may affect multiple features, integrations, and dependent components. Executing every available regression test after every change can be time-consuming.
AI-assisted analysis can help prioritize tests based on:
- Modules affected by the change.
- Dependencies between components.
- Historical defect patterns.
- Previous execution results.
- Business criticality.
- Existing requirements-to-test traceability.
Consider an electric vehicle where a software update modifies battery temperature monitoring.
The QA team may need to evaluate its potential impact on charging control, thermal protection, charging recovery, driver notifications, and battery diagnostics.
How requirements traceability supports regression testing
A well-maintained traceability structure helps establish relationships between requirements, system components, test cases, execution results, and defects.
A simplified workflow is:
Changed Requirement → Impact Analysis → Related Test Cases → Regression Test Selection
AI may help identify related requirements and suggest potentially affected test cases.
However, the final regression scope should be based on engineering impact analysis, risk assessment, system architecture, and applicable safety requirements.
A poorly maintained traceability structure can also limit the quality of AI-assisted impact analysis.
For more information, see ReqSphere's guide, Requirements Traceability: A Practical Guide with Real-World Examples.
7. Practical Applications Across Industries
AI-assisted testing can support different business and engineering domains.
| Industry | Practical Testing Applications |
|---|---|
| Automotive | Battery charging, vehicle diagnostics, sensor monitoring, and communication interfaces |
| Logistics | Shipment tracking, delivery status, route changes, and failed delivery workflows |
| HR Management | Employee onboarding, leave calculations, attendance, and payroll integration |
| Healthcare | Patient registration, appointment scheduling, laboratory results, and medical software interfaces |
For example, in an HR application, AI may suggest scenarios involving leave applications submitted across weekends, public holidays, or employee transfer dates.
In healthcare applications, it can help identify scenarios involving missing patient information or incorrect associations between laboratory results and patient records.
Actual expected results must always be based on approved business and technical requirements. Safety-related applications also require appropriate risk management and verification evidence.
8. Testing AI-Powered Features: A Different Testing Challenge
So far, we have discussed using AI to support software testing.
Another important area is testing applications that themselves contain AI-powered features.
Consider an electric vehicle manufacturer introducing an AI-powered voice assistant.
The driver can ask:
- What is my current battery charge?
- Where is the nearest charging station?
- How much driving range is remaining?
- Find a charging station along my route.
Unlike conventional software functionality, generative AI may produce different responses depending on context, input phrasing, model behavior, and available information.
This introduces additional testing challenges.
| Test Area | Example Scenario |
|---|---|
| Accuracy | Does the assistant report the correct battery charge? |
| Relevance | Does it understand the driver's question? |
| Consistency | Does it provide consistent information for repeated queries? |
| Context handling | Can it understand follow-up questions? |
| Safety | Does it avoid misleading vehicle operation advice? |
| Privacy | Does it prevent access to another driver's information? |
| Hallucination | Does it invent information unavailable in vehicle data? |
| Prompt injection | Can malicious instructions bypass intended restrictions? |
| Resilience | What happens when the AI service becomes unavailable? |
| Model drift | Does behavior change following a model update? |
For example, if a driver asks for the remaining driving range, the assistant should use appropriate vehicle data rather than inventing a value.
Similarly, when the AI service becomes unavailable, the application should provide an appropriate fallback according to its requirements.
Testing AI-powered applications therefore requires more than checking whether the application responds. Testers must evaluate accuracy, reliability, privacy, robustness, and the appropriateness of generated responses.
For safety-related automotive functionality, conversational correctness must also be distinguished from safety-critical vehicle control.
9. Limitations and Risks of Using AI
Although AI offers several benefits, relying on it without appropriate controls can introduce new risks.
Incorrect or incomplete test scenarios: AI may overlook important conditions or generate scenarios that do not match approved requirements. Testers must validate suggestions against specifications and business rules.
Hallucinated information: AI can produce incorrect technical explanations, fabricated API details, or unsupported conclusions. Technical claims must be verified against authoritative documentation and actual execution evidence.
Data privacy and confidentiality: Uploading customer information, business documents, or production logs to an unapproved AI service can introduce security and compliance risks. Organizations should establish clear AI usage policies and mask sensitive information.
Unreliable automation: Generated scripts may contain incorrect locators, weak assertions, or incomplete validation logic. They should be reviewed and executed in controlled environments.
Overdependence on AI: Accepting generated results without questioning them can cause testers to overlook defects that require deeper investigation or domain understanding.
Incomplete context: AI may not have access to the complete application architecture, interface specifications, or historical engineering decisions. Testers must recognize when additional information is necessary.
The practical approach is to treat AI-generated results as proposals requiring appropriate validation.
10. Three Skills Software Testers Need in the AI Era
The growing use of AI does not make testing knowledge less important. In many situations, it makes strong testing fundamentals even more valuable.
1. Analytical Thinking and Domain Knowledge
Testers must understand business processes and technical behavior.
For example, testing an EV battery charging feature requires understanding the relationship between SoC, charging limits, temperature monitoring, and charging control.
Without this knowledge, even numerous AI-generated test cases may overlook important system-level risks.
2. AI-Assisted Testing and Prompting
Testers should learn to provide clear requirements, context, constraints, and expected output formats when interacting with AI tools.
A useful approach is the ART framework (Act as, Request, Terms):
- Act as: Define the role or expertise AI should assume. For example, act as a Senior QA Engineer specializing in automotive software testing.
- Request: Clearly describe the task AI needs to perform, such as analyzing a requirement and generating test cases.
- Terms: Specify the instructions, constraints, expected output format, and quality criteria. For example, include positive and negative scenarios, boundary conditions, and expected results in a structured table.
Using ART, testers can provide more structured instructions and obtain outputs that are easier to evaluate and use.
However, a well-structured prompt does not guarantee technically correct results. Reviewing AI-generated output against approved requirements and engineering specifications remains essential.
3. Technical Testing and Automation Skills
Understanding APIs, automation frameworks, test data, and application architecture helps testers evaluate AI-generated scripts and technical recommendations.
Testers do not necessarily need to become full-time developers, but they should be comfortable understanding generated scripts and identifying incorrect assertions.
For more guidance, see ReqSphere's guide, 10 Rules for Writing Good Software Requirements.
11. How QA Teams Can Start Using AI
QA teams do not need to introduce AI into every testing activity immediately. A gradual approach is more practical.
Step 1: Start with Requirement Analysis
Use an approved AI assistant to identify ambiguities, missing acceptance criteria, and potential negative scenarios.
Step 2: Generate Draft Test Scenarios
Ask AI to suggest positive, negative, boundary, and exception scenarios. Review them before adding them to the test repository.
Step 3: Improve Test Documentation
Use AI to improve test case descriptions, organize defect reports, and summarize execution observations.
Step 4: Explore Automation Assistance
Experiment with AI-assisted script generation and code explanation using a non-production application or controlled test environment.
Step 5: Evaluate the Results
Compare AI-assisted work with the team's existing approach.
Useful measurements include:
| Metric | What It Measures |
|---|---|
| Test preparation time | Time saved during test design |
| Scenario relevance | Valid AI-generated scenarios accepted after review |
| Requirement gap identification | Valid gaps discovered through AI assistance |
| Defect reporting time | Time required to prepare structured reports |
| Automation reliability | Whether generated scripts execute consistently and validate the intended behavior |
A higher number of generated test cases does not necessarily indicate better test coverage. Generating 100 repetitive test cases may provide less value than identifying five previously overlooked high-risk scenarios.
The objective should be measurable improvement in software quality and testing efficiency.
For a broader perspective on AI-assisted development, see ReqSphere's article, The Future of Software Development: Will Artificial Intelligence (AI) Prompts Replace Coding?
12. Will AI Replace Software Testers?
AI can automate or accelerate several testing activities, particularly repetitive tasks involving test generation, data preparation, script assistance, and result analysis.
However, software testing involves much more than executing predefined instructions.
Testers need to understand user expectations, interpret business requirements, evaluate risks, investigate unexpected behavior, and decide whether a product is ready for release.
Consider the EV battery charging example again. AI may generate dozens of test scenarios, but the QA team must still determine whether they adequately cover approved requirements, whether expected results are correct, and whether sufficient evidence exists for release decisions.
The more likely direction is that AI will change the responsibilities of testers rather than eliminate the need for testing professionals.
Testers who combine AI capabilities with domain knowledge, analytical thinking, and strong testing fundamentals will be better positioned to adapt.
Conclusion
AI is becoming a valuable assistant in software testing. It can help testers analyze requirements, generate test scenarios, accelerate automation development, investigate defects, and prioritize regression testing.
My experience with AI-assisted QA activities has demonstrated its practical value in test case generation, test case review, defect reporting, documentation, and automation.
However, AI-generated outputs are not automatically correct, complete, or aligned with business expectations. The responsibility for validating software quality remains with the people and teams developing and testing the product.
The future of software testing will increasingly involve collaboration between testing professionals, automation technologies, and AI-powered assistants.
AI can help generate tests. Experienced testers know which tests matter, why they matter, and whether the results can be trusted.
Further Reading
Explore these related ReqSphere guides:
- What Is Requirements Engineering? A Practical Guide for Beginners
- Requirements Traceability: A Practical Guide with Real-World Examples
- 10 Rules for Writing Good Software Requirements
- The Future of Software Development: Will Artificial Intelligence (AI) Prompts Replace Coding?
References and Further Resources
- ISTQB – Certified Tester AI Testing (CT-AI).
- ISTQB – Certified Tester Testing with Generative AI (CT-GenAI).
- ISO 26262 – Road Vehicles: Functional Safety.
- ISO/IEC 42001 – Artificial Intelligence Management System.
- ISO/IEC 23894 – Artificial Intelligence Risk Management.
- NIST – AI Risk Management Framework.
Discussion
Post a comment