- CT-GenAI (Certified Tester Testing with Generative AI) is the ISTQB certification that proves you can use generative AI and large language models (LLMs) to do better testing.
- It is a Specialist module; the formal prerequisite is the ISTQB Foundation Level (CTFL) certificate.
- The syllabus has 5 chapters and requires at least 13.6 hours of accredited training; almost half of that time is prompt engineering applied to testing.
- The exam has 40 multiple-choice questions, lasts 60 minutes (75 if it is not your native language) and you pass with 30 out of 46 points (65%).
- It covers structured prompts, prompt chaining, few-shot and meta prompting, hallucinations and bias, privacy, RAG, agents, fine-tuning, LLMOps and how to roll out GenAI in a test team.
- Do not confuse it with CT-AI: CT-GenAI is about using AI to test; CT-AI is about testing AI-based systems.
- You can prepare with our live ISTQB Testing with Generative AI course on Microsoft Teams.
Generative AI is already part of every QA team's routine: someone asks a chatbot for test cases, someone else generates test data, another person summarizes a regression report. The problem is that it is usually done by gut feeling, with no method, no way of measuring output quality and little care for company data.
The ISTQB CT-GenAI certification exists to fix exactly that. It gives testers a shared, internationally recognized framework for using generative AI effectively, safely and measurably. In this guide I explain what it is, who it is for, what the syllabus covers chapter by chapter, what the exam looks like and how to prepare.
What CT-GenAI is and where it sits in the ISTQB portfolio
CT-GenAI is the ISTQB Specialist module focused on using generative AI in software testing. ISTQB released version 1.0 of the syllabus in 2025 and, in April 2026, announced version 1.1, a minor update: targeted corrections, terminology updates, clearer descriptions of the evaluation metrics and more context on LLM-powered agents. According to ISTQB, v1.1 does not change the structure, learning objectives or scope, so everything here applies to both.
In the ISTQB portfolio, Specialist modules go deep into one topic (AI, automation, performance, security, DevOps and more) and are taken after Foundation. For the full map and prices in your currency, see our ISTQB certification guide.
CT-GenAI vs. CT-AI
This is the most common question. Both deal with AI, but from opposite sides:
| CT-GenAI | CT-AI | |
|---|---|---|
| Core question | How do I use generative AI to test better? | How do I test a system that uses AI or machine learning? |
| Key content | Prompts, LLM risks, RAG, agents, adoption | Data, ML models, metrics, bias, metamorphic testing |
| Typical profile | Any tester, analyst or lead who wants AI-driven productivity | Testers of AI/ML products, data analysts, technical QA |
| Minimum training | 13.6 h | 19.5 h (v2.0) |
Starting with CT-AI version 2.0, ISTQB removed the "using AI for testing" content from that syllabus and left it to CT-GenAI, so today the two certifications complement each other without overlapping. I cover this in detail in the CT-AI guide.
Who it is for and prerequisites
According to the syllabus, CT-GenAI is aimed at anyone using generative AI in testing: testers, test analysts, test automation engineers, test managers, user acceptance testers and developers. It also suits people who need to understand the topic without doing the hands-on work: project managers, quality and development managers, business analysts, IT directors and consultants.
- Formal prerequisite: the ISTQB Foundation Level certificate (CTFL, any version). Without it you cannot sit the exam. If you do not have it yet, start with our ISTQB Foundation Level 4.0 course.
- Technical background: no programming or machine learning experience is required. It helps to have used a chatbot (ChatGPT, Claude, Gemini, Copilot) and to know the CTFL test process: analysis, design, implementation, execution, monitoring and control.
- Language: the syllabus is available in English and Spanish. The certification has the same value in any language.
What you will be able to do: business outcomes
The syllabus defines five business outcomes, that is, what a certified person can bring to their organization:
| Code | Business outcome | In practice |
|---|---|---|
| GenAI-BO1 | Understand the fundamental concepts, capabilities and limitations of generative AI | Know when an LLM helps and when it is a bad idea |
| GenAI-BO2 | Develop practical prompting skills for testing | Write prompts that produce useful, repeatable results |
| GenAI-BO3 | Gain insight into the risks and their mitigations | Spot hallucinations, protect data and stay compliant |
| GenAI-BO4 | Gain insight into GenAI solutions for testing | Understand RAG, agents, fine-tuning and LLMOps |
| GenAI-BO5 | Contribute to a GenAI strategy and roadmap for testing | Propose how and in what order your QA team adopts AI |
Learning objectives go up to K3 (apply), and the syllabus includes hands-on objectives at three levels: H0 (demo), H1 (guided exercise) and H2 (exercise with hints). Hands-on objectives are not examined, but they are what makes the knowledge usable on Monday morning.
Syllabus content chapter by chapter
These are the five examinable chapters with the minimum training time ISTQB sets for an accredited course:
| Chapter | Minimum time | Key topics |
|---|---|---|
| 1. Introduction to Generative AI for Software Testing | 100 min | Types of AI, LLMs, tokens, context window, reasoning and multimodal models, chatbots vs. LLM-powered apps |
| 2. Prompt Engineering for Effective Software Testing | 365 min | Six-part prompt structure, prompt chaining, few-shot, meta prompting, system prompts, GenAI for analysis, design, regression and monitoring, metrics |
| 3. Managing Risks of Generative AI in Software Testing | 160 min | Hallucinations, reasoning errors, bias, non-determinism, privacy, security, energy use, regulations |
| 4. LLM-Powered Test Infrastructure for Software Testing | 110 min | Architecture, RAG, vector databases, agents, fine-tuning, LLMOps |
| 5. Deploying and Integrating Generative AI in Test Organizations | 80 min | Shadow AI, strategy, LLM/SLM selection, adoption phases, skills and role changes |
Chapter 1. Introduction to Generative AI for Software Testing
The first chapter lays the foundations. It distinguishes the AI spectrum: symbolic AI (rules), classical machine learning, deep learning and generative AI. Then it explains how an LLM works: text is split into tokens, the model works with embeddings and it has a limited context window.
Why should a tester care? Because if you paste an 80-page specification and the model "forgets" half of it, it is not bad luck: you went past the context window. One of the hands-on exercises is precisely counting tokens for a test task.
It also distinguishes three kinds of models: foundation models (trained on huge datasets), instruction-tuned models (the ones behind chatbots) and reasoning models (which work through problems step by step). It then covers multimodal models that accept images as well as text. A practical example: you give the model a screenshot of a sign-up form and ask it to identify UI test conditions.
It closes with the key LLM capabilities for testing: analyzing and improving requirements, supporting test case creation, suggesting expected results (oracles), generating test data, supporting automation, analyzing results and drafting testware such as plans and reports. And it compares two interaction models: the chatbot (conversational, great for quick tasks and exploration) and LLM-powered testing applications integrated via API (more customizable and scalable, and the basis for agents).
Chapter 2. Prompt Engineering for Effective Software Testing
This is the heart of the certification, with more than six hours of minimum training time. It starts with the six components of a structured prompt:
- Role: the perspective the model takes ("act as a test analyst with banking experience").
- Context: the test object and its environment.
- Instruction: the specific task, clear and imperative.
- Input data: user stories, acceptance criteria, screenshots, code, existing test cases.
- Constraints: what the model must respect or avoid.
- Output format: table, Gherkin, JSON, prioritized list and so on.
Then come the three core techniques:
- Prompt chaining: you break the task into steps and check each output before moving on. Ideal when you need precision, such as analyzing a user story, then deriving test conditions and finally writing test cases.
- Few-shot prompting: you give several examples of the output you expect (zero-shot has none; one-shot has one). Perfect for repetitive tasks with a fixed format, such as Gherkin scenarios or keyword-driven scripts.
- Meta prompting: you ask the model to help you build or improve the prompt itself. Useful when the task is new and you do not know where to start.
It also separates the system prompt (fixed rules for the whole conversation, usually set by whoever configures the tool) from the user prompt (each specific request).
The K3 part is applying all this to the test process: test analysis (for example, generating acceptance criteria from a wireframe), test design and implementation (functional test cases, Gherkin scenarios, prioritizing a suite based on dependencies), automated regression testing (keyword-driven scripts, analysis of execution reports) and test monitoring and control (metrics prepared by AI from test data).
Example of a prompt chain for a "reset password" story: step 1, ask for ambiguities and questions for the Product Owner; step 2, with the answers, ask for acceptance criteria; step 3, ask for test cases with boundary values for link expiration. You review each step before moving on.
The chapter ends with how to evaluate AI output. The syllabus proposes metrics such as accuracy, precision, recall, relevance and contextual fit, diversity, execution success rate of generated scripts and time efficiency. And refinement techniques: iterative prompt modification, A/B testing of prompts, output analysis, user feedback and adjusting prompt length and specificity. Because LLMs are non-deterministic, it insists that metrics rely on statistically relevant data, not a single run.
Chapter 3. Managing Risks of Generative AI in Software Testing
This is what separates a casual user from a professional. The syllabus defines, and teaches you to detect, three problems:
- Hallucinations: the model confidently makes things up, such as a field that is not in the specification or a business rule nobody asked for.
- Reasoning errors: the logic breaks, such as an effort estimate that does not add up or a contradictory prioritization.
- Bias: test data that always uses the same names, countries or profiles and leaves real users out.
The mitigations are very concrete: provide complete context, split the task into steps (prompt chaining), use clear data formats, pick the right model and compare results across models. For non-deterministic behavior, it mentions lowering the temperature and setting a random seed when the tool allows it.
The second part covers privacy and security: what happens when someone pastes customer data into a public tool, vulnerabilities such as prompt injection, and measures such as data minimization, anonymization or pseudonymization, encryption, team training, systematic review of outputs and choosing a secure environment (an enterprise offering, a private cloud or a model hosted on your own infrastructure). It recommends involving security, legal and the CISO.
It also addresses energy consumption and CO₂ emissions (generating images uses far more energy than generating text) and gives an overview of regulations and frameworks: ISO/IEC 42001:2023, ISO/IEC 23053:2022, the EU AI Act and the US NIST AI Risk Management Framework.
Chapter 4. LLM-Powered Test Infrastructure for Software Testing
This chapter explains what sits behind the AI testing tools on the market. First, the architectural components of an LLM-powered test infrastructure. Then RAG (Retrieval-Augmented Generation): instead of relying on what the model "knows", relevant chunks of your own documents (requirements, existing test cases, manuals) are retrieved from a vector database and added to the prompt. The result is answers grounded in your project and fewer hallucinations.
Next come LLM-powered agents: applications that do not just answer but act by calling tools (opening a browser, querying an API, creating a ticket). The syllabus distinguishes autonomous and semi-autonomous agents and mentions multi-agent architectures with orchestration. It warns that agents inherit the same hallucination risks, so semi-autonomous agents and automated verification are advisable for critical tasks. To see this in practice, read our guide on AI-assisted testing with Playwright MCP.
It closes with fine-tuning (retraining a model on your domain data for a specific task) and LLMOps, the practices for deploying, monitoring and maintaining LLMs in production, much like DevOps does for software.
Chapter 5. Deploying and Integrating Generative AI in Test Organizations
The last chapter moves to strategy. It starts with the risk of shadow AI: people using unapproved tools, with security gaps, regulatory non-compliance and unclear intellectual property.
Then it describes the elements of a GenAI strategy for testing: measurable objectives (productivity, shorter cycles, higher quality), model selection, data quality, training, metrics and guidelines (sensitive data rules, transparency about what was AI-generated, review quality gates). To select an LLM or SLM it proposes four criteria: performance on your tasks, fine-tuning potential, recurring cost, and community and support. One hands-on exercise is estimating the monthly cost of a task based on tokens and frequency.
It defines three adoption phases: discovery, initiation and usage definition, and utilization and iteration. It ends with how roles change: testers become AI-assisted test specialists who guide and verify generated testware and maintain prompt libraries, while test managers take on AI strategy, risk and governance and coordinate hybrid teams of people and agents.
Download the official syllabus
These are the official ISTQB syllabi. Read them in full before the exam: everything except the introduction, hands-on objectives and appendices is examinable. A Spanish version is also available.
For version 1.1 and the official sample exams, check the certification page on istqb.org.
The CT-GenAI exam
Verified structure
| Item | Value |
|---|---|
| Questions | 40, multiple choice |
| Total points | 46 (some questions are worth more than one point) |
| Passing score | 30 points (65%) |
| Duration | 60 minutes; 75 minutes if you take it in a non-native language (+25%) |
| Prerequisite | ISTQB Foundation Level certificate |
| Cognitive levels | K1, K2 and K3 |
Source: official ISTQB exam structure tables and the CT-GenAI page on istqb.org.
Exam options
- Online from home with remote proctoring (iSQI FLEX): available 24/7. You need a webcam, a microphone, your phone as a second camera and a government-issued ID. Book with enough notice.
- Pearson VUE: at a test center or online with OnVUE.
- On-site at your company: for groups of 6 or more, Cool Testers brings a proctor to you.
Paper-based exams are no longer offered.
Where to buy it
The exam is not included in the course. You have two options:
- Directly from iSQI, our partner and official exam provider, with a 20% coupon off their listed price for your first attempt, which we send when you enroll. This is the best option if you are paying yourself.
- Through Cool Testers: our certification exam with 1 or 2 attempts. Best if your employer pays or you need an invoice. You have 1 year to take it.
How to prepare
- Get your CTFL first. You cannot sit the exam without Foundation. See our CTFL 4.0 guide.
- Read the whole syllabus and build your own glossary of GenAI terms: each chapter's keywords are examinable even if they do not appear in a learning objective.
- Practice with real prompts. Take user stories from your project (with no sensitive data) and apply the six-part structure, chaining, few-shot and meta prompting.
- Master chapter 2. It has the most time and the K3 objectives: expect scenario questions where you must choose the right technique for a task.
- Learn to recognize risks. Be clear on the difference between hallucination, reasoning error and bias, and the mitigation for each.
- Work through the official sample exam and analyze why each wrong option is wrong.
- Take an accredited course. In our ISTQB Testing with Generative AI course you do the hands-on exercises live on Teams, with recordings available for 3 years and the guarantee to retake it free in any open group within 6 months.
If you are also taking Foundation, ask about the Foundation + GenAI bundle with 50% off the second course.
Benefits and career paths
- Instant credibility: everyone says they "use AI"; few can prove with an international certification that they do it methodically and with risk control.
- Measurable productivity: you learn to measure the quality of AI output and the time you save, a strong argument with your manager or client.
- An emerging role: the syllabus describes the shift toward AI-assisted testers who maintain prompt libraries and verify generated testware, a profile that shows up more and more in QA job postings.
- Leadership: if you are a test lead or manager, it gives you the vocabulary to propose a GenAI strategy, prevent shadow AI and talk to security and legal.
- For companies: a certified team adopts AI with clear rules and less risk. We can deliver the course to your team; request a quote.
Related certifications and next steps
- Before: CTFL 4.0, mandatory.
- Alongside: CT-AI, if your product includes AI or machine learning models you need to test.
- Automation: CTAL-TAE for the technical side and CT-TAS for organization-wide automation strategy.
- Leadership: CTAL-TM (Test Management) if you will define the test strategy, AI included.
- DevOps: CT-QDO if your team works with CI/CD pipelines.
ISTQB suggests the Core Advanced modules (Test Analyst, Technical Test Analyst, Test Management, Test Automation Engineering) as a next step, followed by the Expert Level.
Frequently asked questions
Do I need to know how to code for CT-GenAI?
No. The syllabus does not require programming. Being able to read a script helps with the automation topics, but it is not a requirement.
What is the prerequisite for the CT-GenAI exam?
You must hold the ISTQB Foundation Level (CTFL) certificate before taking the exam. It is the only formal requirement.
How many questions are on the exam and what is the passing score?
40 multiple-choice questions worth 46 points in total. You pass with 30 points (65%). You have 60 minutes, or 75 if you take it in a non-native language.
What is the difference between CT-GenAI and CT-AI?
CT-GenAI teaches you to use generative AI to test. CT-AI teaches you to test systems that use artificial intelligence or machine learning. They complement each other.
Does the syllabus cover specific tools like ChatGPT?
No. It is vendor-neutral and talks about LLMs, chatbots and LLM-powered applications in general. What you learn works with any tool.
Is the exam included in the course?
No. You can buy it directly from iSQI with the 20% coupon we send when you enroll, or through Cool Testers if your employer pays or you need an invoice.
Can I take the exam from home?
Yes, online with remote proctoring (iSQI FLEX) or with Pearson VUE OnVUE. You can also go to a Pearson VUE test center.
What changed in syllabus version 1.1?
It is a minor update: corrections, terminology, clearer evaluation metrics and more context on agents. The structure, learning objectives and scope stay the same.