ISTQB CT-GenAI Practice Questions with Explanations
Free ISTQB CT-GenAI practice questions. 14 of them, each with the correct answer, a full explanation, and the reason every other option is wrong.
CT-GenAI practice questions
Question 1
Which statement ACCURATELY describes Classical Machine Learning within the context of AI?
- A. It primarily relies on rule-based systems to model human decision processes.
- B. It utilizes neural networks with multiple hidden layers to automatically discover patterns in raw data.
- C. It processes vast textual datasets to generate new, human-like content by mimicking learned patterns.
- D. It employs a data-driven methodology that involves explicit data preparation and feature engineering.
Show answer and explanation ▾
Correct answer: D
Classical machine learning is data-driven. Teams prepare the data and design features by hand before training a model, which separates it from rule-based and deep learning approaches.
Why the other options are wrong:
- A. Relying on rule-based systems describes symbolic AI.
- B. Neural networks with many hidden layers describe deep learning.
- C. Generating new human-like content from vast text data describes generative models such as LLMs.
Question 2
Which of the following statements BEST describes the purpose of embeddings in Large Language Models (LLMs) when processing and generating content?
- A. They define the maximum length of text an LLM can process at one time to maintain coherence.
- B. They break down input text into smaller units, such as words or sub-words, for individual processing.
- C. They are numerical representations of tokens that capture their semantic, syntactic, and contextual relationships.
- D. They control the randomness or creativity of an LLM's outputs during text generation.
Show answer and explanation ▾
Correct answer: C
Embeddings turn tokens into vectors of numbers that capture meaning, syntax and context, so the model can work with how tokens relate to each other.
Why the other options are wrong:
- A. Maximum text length is set by the context window.
- B. Splitting input into words or sub-words is tokenization.
- D. Randomness or creativity is controlled by settings such as temperature.
Question 3
In the context of Large Language Models (LLMs), which of the following statements (i-v) regarding 'tokenization' and 'context window' are CORRECT? i) Tokenization is the process of breaking down textual input into smaller units called tokens. ii) The context window dictates the maximum number of tokens that an LLM can consider at any given time to maintain coherence. iii) Increasing the size of an LLM's context window generally leads to a reduction in computational complexity and processing time. iv) Tokenization primarily involves converting tokens into highdimensional vectors to capture their semantic relationships. v) A larger context window allows an LLM to maintain coherence over longer passages, such as when analyzing large test logs.
- A. i, ii, and iii
- B. i, ii, and v
- C. ii, iv, and v
- D. iii, iv, and v
Show answer and explanation ▾
Correct answer: B
Statements i, ii and v are true: tokenization splits text into tokens, the context window caps how many tokens the model can consider, and a larger window helps coherence over long inputs. Statement iii is false because larger windows increase computation, and iv describes embeddings rather than tokenization.
Why the other options are wrong:
- A. It includes statement iii, but larger context windows increase computation and processing time.
- C. It includes statement iv, but turning tokens into vectors is embedding, not tokenization.
- D. It includes statements iii and iv, which are both false.
Question 4
In the context of LLMs, which statement BEST characterizes an instruction-tuned Large Language Model?
- A. It is a general-purpose model primarily used for broad tasks without specific adaptation.
- B. It is specifically designed to perform complex multi-step logical inference and problem-solving.
- C. It is a foundation model that has been further refined to better follow human instructions and user intent.
- D. It can create new content like images or code by learning patterns from diverse training data.
Show answer and explanation ▾
Correct answer: C
An instruction-tuned LLM starts as a foundation model and is then refined so it follows human instructions and user intent better.
Why the other options are wrong:
- A. A general-purpose model used without adaptation describes a base foundation model.
- B. Multi-step logical inference describes reasoning models.
- D. Creating images or code from learned patterns describes generative models in general, not instruction tuning.
Question 5
A test team is analyzing screenshots of a mobile application's GUI alongside textual defect reports to identify visual inconsistencies that are not captured by these reports.Which type of Generative AI model is MOST suited to assist with this task?
- A. Symbolic AI, due to its rule-based system for mimicking human decision-making.
- B. Classical Machine Learning, as it excels at data-driven defect categorization.
- C. Multimodal LLMs, specifically those augmented with vision-language models.
- D. Reasoning LLMs, given their emphasis on logical inference for complex problem-solving.
Show answer and explanation ▾
Correct answer: C
Comparing screenshots with text defect reports needs a model that handles both images and text. A multimodal LLM with vision-language capability does that.
Why the other options are wrong:
- A. Symbolic AI follows fixed rules and cannot interpret screenshots.
- B. Classical machine learning categorizes defects from structured data, not from screenshots next to text.
- D. Reasoning LLMs focus on logical inference, not on analyzing images.
Question 6
Which TWO of the following capabilities are primary strengths of Large Language Models (LLMs) when applied to software testing tasks? (Choose two.)
- A. Automating the generation of synthetic test data to cover various combinations and boundary values.
- B. Providing real-time, autonomous execution of performance tests without human oversight.
- C. Interpreting complex requirements and suggesting test objectives based on user stories.
- D. Independently designing and implementing test environments with zero human configuration.
- E. Developing new software features based on high-level user feedback.
Show answer and explanation ▾
Correct answer: A, C
Key LLM capabilities for test tasks include generating test data, such as datasets, boundary values and different data combinations, and interpreting requirements or user stories to suggest test objectives and test cases.
Why the other options are wrong:
- B. LLMs support test work under human oversight and do not run performance tests autonomously without it.
- D. Designing and building test environments with no human configuration is not one of the LLM strengths for testing.
- E. Developing new software features is a development activity, not a strength of LLMs in software testing tasks.
Question 7
A test team is looking to leverage Generative AI. They have two immediate needs:Need 1. For rapid, ad-hoc exploration: A new tester needs to quickly understand requirements and generate simple, one-off test ideas interactively.Need 2. For structured, repetitive automation: The team wants to integrate AI into their existing continuous integration (CI) pipeline to automatically generate comprehensive test data for daily regression runs.Which of the following statements BEST describes how AI chatbots and LLM-powered testing applications would apply to these needs?
- A. Both needs are best addressed by an AI chatbot.
- B. Both needs are best addressed by an LLM-powered testing application.
- C. Need 1 is best addressed by an AI chatbot, and Need 2 by an LLM-powered testing application.
- D. Need 1 is best addressed by an LLM-powered testing application, and Need 2 by an AI chatbot.
Show answer and explanation ▾
Correct answer: C
AI chatbots suit fast feedback, dynamic exploration of requirements and onboarding new testers through a conversational interface. LLM-powered testing applications integrate LLMs through APIs into existing frameworks to automate repetitive tasks such as test data synthesis.
Why the other options are wrong:
- A. A chatbot is not the fit for automated, repeatable test data generation inside a CI pipeline.
- B. An integrated testing application is not the fit for quick, interactive, one-off exploration by a new tester.
- D. This reverses the roles of the two approaches.
Question 8
A tester is crafting a prompt for an LLM to generate test cases. One part of the prompt reads:"The generated test cases must be provided in a CSV file format, with columns for Test ID, Description, Preconditions, Test Steps, and Expected Result."In which component of the six-part prompt structure would this line MOST LIKELY appear?
- A. Role
- B. Constraints
- C. Instruction
- D. Output format
Show answer and explanation ▾
Correct answer: D
The output format component states the expected format, structure or characteristics of the response. Requiring a CSV file with named columns describes the shape of the response, so it belongs in the output format.
Why the other options are wrong:
- A. The role sets the persona the model adopts, not the format of the response.
- B. Constraints set restrictions on how the instructions are applied to the input data, not the file layout of the response.
- C. The instruction describes the task to perform, not the structure of the result.
Question 9
A test engineer is learning about the six components of a structured prompt for Generative AI in software testing.Which of the following statements INCORRECTLY describes one of these components?
- A. The Role component specifies the persona the Generative AI model should adopt, such as a tester or test automation engineer.
- B. The Context component provides background information about the test object or specific functionality being tested to help the model determine test conditions.
- C. The Instruction component outlines the specific task the Generative AI model is required to perform, including relevant requirements for the task.
- D. The Constraints component is primarily used to specify the desired output format for the AI's response, such as a markdown table or Gherkin syntax.
Show answer and explanation ▾
Correct answer: D
Constraints set restrictions or special considerations the model must follow and how instructions apply to the input data. Specifying a markdown table or Gherkin syntax is the job of the output format component, so this description is incorrect.
Why the other options are wrong:
- A. This correctly describes the role component as the persona the model adopts.
- B. This correctly describes the context component as background information about the test object.
- C. This correctly describes the instruction component as the task to perform with its relevant requirements.
Question 10
A test lead needs an LLM to generate a comprehensive test plan summary, which requires iterative refinement and review. The task is complex and benefits from breaking it into smaller, verifiable steps, where the output of one step informs the next.Which core prompting technique is BEST suited for this scenario?
- A. Few-shot prompting
- B. Meta prompting
- C. Prompt chaining
- D. Zero-shot prompting.
Show answer and explanation ▾
Correct answer: C
Prompt chaining breaks a task into a series of intermediate prompts, checking and refining each result before the next step, so each response informs the next prompt. It suits complicated test tasks that need decomposition and systematic checking.
Why the other options are wrong:
- A. Few-shot prompting gives the model examples in a prompt and does not split the task into verified steps.
- B. Meta prompting uses the model to generate or refine prompts, not to work through dependent steps.
- D. Zero-shot prompting gives no examples and relies on a single prompt.
Question 11
What is the primary characteristic that distinguishes a system prompt from a user prompt in an interaction with an LLM?
- A. A system prompt changes dynamically with each user input, while a user prompt remains constant throughout the conversation.
- B. A system prompt contains specific questions or tasks from the user, while a user prompt defines the LLM's overall behavior.
- C. A system prompt is always visible and editable by the chatbot user, unlike the user prompt.
- D. A system prompt is usually set by the developer, while a user prompt is written by the end user.
Show answer and explanation ▾
Correct answer: D
A system prompt is typically defined by the developer or tester to guide the overall behavior of the LLM and stays constant for the session. The user prompt is the actual input from the chatbot user and changes with each interaction.
Why the other options are wrong:
- A. The system prompt stays constant, while the user prompt changes with each interaction.
- B. This reverses them: the system prompt defines overall behavior and the user prompt carries the user's tasks.
- C. In most interfaces the system prompt is not visible or editable by the chatbot user.
Question 12
You are a test analyst using Generative AI to support early test analysis activities for newly written user stories. Your objective is to ensure the stories are clear, testable, and ready for test case design.Which of the following sequences BEST demonstrates how Generative AI can be applied to enhance test analysis through iterative review and refinement? i. Initial Review: Ask the LLM to analyze the user stories and identify any ambiguities, gaps, or testability issues. ii. Refinement Suggestions: Based on the identified issues, prompt the LLM to propose clearer or more testable rewordings. iii. Testability Check: Submit the revised user stories to the LLM for a final assessment of their clarity, completeness, and readiness for test case generation. iv. Risk-Based Prioritization: Use the LLM to assign risk levels to the conditions in the user stories and suggest testing priorities. v. Direct Test Case Generation: Ask the LLM to produce full test cases based on the original user stories
- A. i, ii, and iii
- B. ii, iv, and v
- C. iii, iv, and v
- D. i, ii, and iv
Show answer and explanation ▾
Correct answer: A
Iterative analysis of a user story with GenAI works step by step: first identify ambiguities, then improve testability, and finally assess completeness, with each step using the checked result of the one before. Steps i, ii and iii follow exactly that review and refinement cycle.
Why the other options are wrong:
- B. Skipping the initial review and generating test cases from the original stories breaks the review and refinement cycle.
- C. Without the initial review and refinement steps there is nothing refined to check, and generating from the original stories ignores the improvements.
- D. Risk-based prioritization does not complete the cycle, because it replaces the final check that the revised stories are clear and complete.
Question 13
Which of the following BEST reflects the shift in a tester's responsibility after the adoption of Generative AI in test activities within an organization?
- A. Focusing on the manual creation of test documentation.
- B. Maintaining test-specific prompt libraries and refining AI-generated output.
- C. Delegating all test design activities to the AI.
- D. Managing the overall organizational budget for AI tool procurement.
Show answer and explanation ▾
Correct answer: B
With GenAI, testers become AI-assisted test specialists who guide and verify AI-generated testware. Their tasks expand to include reviewing AI output, refining prompts and maintaining test-specific prompt libraries.
Why the other options are wrong:
- A. The shift moves testers toward guiding and verifying AI output, not toward more manual documentation.
- C. Testers keep their test technique expertise and verify AI output rather than handing all test design to the AI.
- D. Organizational budgets for AI tools are not a tester responsibility described in this shift.
Question 14
You are leading a team that uses Generative AI to enhance automated regression testing. You have identified that adapting test scripts to minor UI or API changes is a frequent challenge, leading to unnecessary failures and maintenance overhead. You want the LLM to proactively adjust test scripts to handle such modifications.Which of the following improvements to a prompt would BEST enable the LLM to support self-healing and adaptive tests?
- A. Add an instruction for the LLM to generate detailed test reports with success metrics and key insights.
- B. Specify a constraint that the LLM should prioritize test execution based on module criticality and historical defect rates.
- C. Expand the instruction to include analyzing code changes to identify high-risk areas for targeted regression testing.
- D. Include a directive for the LLM to automatically analyze UI element changes and modify locators within test scripts to prevent failures.
Show answer and explanation ▾
Correct answer: D
Self-healing and adaptive tests use GenAI to adjust test scripts automatically for minor UI or API changes, such as dynamic locators, so small modifications do not cause unnecessary failures. Directing the LLM to analyze UI element changes and update locators enables exactly that.
Why the other options are wrong:
- A. Detailed test reports support test reporting and insights, not script adaptation.
- B. Prioritizing execution by criticality and defect history does not adapt scripts to UI or API changes.
- C. Analyzing code changes for high-risk areas is impact analysis for targeted regression, not self-healing.
That was 14 of 39.
The full ISTQB CT-GenAI pack has all 39 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. $39.00 USD, paid once, with free monthly updates and a pass-or-your-money-back guarantee.
Get the full packNot ready yet? Get the first 14 free as one PDF by email. Get the Half Pack →
About the CT-GenAI exam
These 14 ISTQB CT-GenAI questions are free here, and as a free PDF by email.
They are drawn from the same bank as the full CT-GenAI pack, which has 39 questions in total.