ISTQB CT-GenAI flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Generative AI definition

A class of AI models that produce new text, code, images or data by learning statistical patterns from training data, rather than retrieving a stored correct answer.

Large language model

A neural network trained on large text corpora to predict token sequences, enabling it to draft test cases, code and reports when given a suitable prompt.

Context window

The maximum number of tokens a model can process in one interaction, covering prompt plus response; exceeding it requires chunking or truncation of input.

Temperature parameter

A sampling setting controlling output randomness; values near zero yield repeatable conservative results, while higher values increase diversity of generated test ideas.

Hallucination in GenAI

Fluent but factually incorrect model output, such as invented APIs, selectors or requirement IDs; mitigated by grounding prompts in source material and human verification.

Few-shot prompting

A prompt technique that includes several input-output examples so the model imitates the desired format, useful for standardized test case templates.

Prompt injection risk

A manipulation where untrusted text in the prompt, such as pasted content, overrides instructions, potentially causing unsafe or incorrect model behavior.

Training data cut-off

The date after which training data ends; the model lacks knowledge of newer frameworks or product changes unless that context is supplied in the prompt.

Test condition extraction

Using GenAI to derive candidate test conditions from requirements and user stories, which testers then validate for completeness and traceability.

Boundary value drafting

Prompting a model to propose boundaries and adjacent values for numeric or date inputs; its arithmetic must be independently checked by the tester.

Equivalence partition suggestion

GenAI proposing valid and invalid input classes for a field or parameter, which the tester reviews to confirm classes are genuinely equivalent.

Decision table generation

Asking GenAI to enumerate condition combinations and resulting actions; unreachable or contradictory rules require human pruning before use.

State transition modeling

Using GenAI to list states, events and transitions from a specification, producing candidate transition coverage that testers verify against actual behavior.

Requirement ambiguity detection

Prompting GenAI to flag unclear, untestable or missing requirements; findings are review prompts, not confirmed defects, until humans assess them.

Risk-based prioritization with AI

GenAI proposing likelihood and impact ratings from requirement text; ratings must be calibrated by the team using incident and business knowledge.

Negative test ideation

Prompting GenAI for invalid inputs, misuse cases and error scenarios, a strong use because models generate diverse adversarial ideas quickly.

Structured test case format

Specifying fields such as ID, preconditions, steps, expected result and priority in the prompt so GenAI output is directly usable and consistent.

Coverage gap review

Mapping AI-drafted test cases back to requirements to find duplicates and untested conditions; coverage metrics are computed by the team, not the model.

AI-generated test code

Automation code drafted by a model from a scenario description; it must be reviewed, compiled and executed because APIs and locators may be invented.

Synthetic test data

Artificially generated data covering valid, invalid and boundary values; preferred over production data to avoid privacy exposure, with format constraints validated programmatically.

Locator suggestion risk

GenAI proposing CSS or XPath selectors from guessed DOM structure; selectors are fragile unless the actual DOM is supplied and resilient strategies are preferred.

Self-healing automation

AI-assisted repair of broken locators after UI changes; proposed fixes must be validated against the real application to avoid masking genuine defects.

AI-generated assertions

Expected results and checks drafted by a model; they carry the model's error risk and should be cross-checked against specifications or independent computation.

Automated defect triage

Using GenAI to summarize failures, cluster stack traces and suggest root causes; classifications require engineer verification before routing or closure.

Prompt version control

Tracking prompt changes and generated artifacts alongside code so results are reproducible and regressions in output quality can be traced.

Human approval gate

A pipeline control ensuring AI-generated code or tests are reviewed before merging, preventing unreviewed model output from reaching the main branch.

AI test summary report

A stakeholder-facing summary drafted by GenAI from supplied execution data; all figures must be recomputed and verified before publication.

Defect report enrichment

Using GenAI to draft clear titles, reproduction steps and severity suggestions from logs and screenshots, while severity and priority remain human decisions.

Log clustering with GenAI

Grouping production incidents and log entries by similarity to reveal patterns; correlation found by the model is not causation and needs telemetry confirmation.

Root cause hypothesis

A candidate cause proposed by GenAI from timeline, logs and recent changes; it guides investigation rather than concluding it.

Retrospective summarization

GenAI condensing retrospective notes and recurring defect patterns into process actions; metric definitions must be fixed in advance for comparability.

AI output labeling

Marking content as AI-generated and retaining human sign-off, supporting audit trails and confidentiality control in regulated reporting.