Exam Blueprints, K-levels and Study Plan — Plus AI Fundamentals for Testers

Why this matters on the job

Your test lead at a Bengaluru IT services company has just been told that a US banking client wants "AI-certified testers" on the team that will test a new loan-scoring model and a GenAI support chatbot. Two people will be sponsored for certification this quarter. You have six weeks, a full-time sprint workload and no machine learning background beyond what you have read on LinkedIn. The people who get picked are the ones who walk in with a realistic study plan, not the ones who buy the most mock-test bundles.

This lesson gives you that plan. You will learn exactly what the two ISTQB AI specialist exams look like, how the cognitive levels (K-levels) change the way you must study, how to split your hours by chapter weight, and the core AI vocabulary that both syllabi assume from the very first page.

Note: this is an independent exam-prep course. It is not affiliated with, sponsored by or endorsed by ISTQB or any ISTQB member board. ISTQB® is a registered trademark of the International Software Testing Qualifications Board. Always treat the official syllabus on istqb.org as the final source.

Concepts

The two certifications at a glance

ISTQB currently offers two AI-related specialist certifications. They answer two different questions. CT-AI (Certified Tester AI Testing) asks: how do you test a system that contains AI, especially machine learning? CT-GenAI (Certified Tester Specialist Level: Testing with Generative AI) asks: how do you use generative AI and LLMs to do testing work better and safely?

Both pages are worth bookmarking: CT-AI v2.0 on istqb.org and CT-GenAI on istqb.org. Because the total points are higher than the number of questions, some questions carry 2 points - in the official CT-AI sample exam, these are usually the application-style (K3) questions. Pacing therefore matters: never leave a 2-point calculation unanswered.

CT-AI v1.0 vs v2.0 - which one should you take?

ISTQB released CT-AI v2.0 in 2026 and it replaces v1.0. According to the official CT-AI v2.0 FAQ, v2.0 shrinks the syllabus from 11 chapters to 7, organises it around the ML lifecycle (input data testing, model testing, ML development testing), adds testing of generative AI and LLMs (including exploratory testing and red teaming) and removes the old chapter on using AI for testing - that topic now lives in CT-GenAI. The exam format is unchanged and there is no bridging exam. v1.0 stays available in English until 21 April 2027 and in other languages until 21 October 2027; existing v1.0 certificates remain valid. If you are starting fresh in late 2026, prepare for v2.0. This course follows v2.0 and CT-GenAI v1.1.

CT-GenAI v1.1 is a minor update of v1.0 (corrections, terminology, refined metrics and risk content, more context on agents). ISTQB states it does not change the structure, learning objectives or scope - see the v1.1 announcement.

K-levels: how deep each objective goes

Every learning objective (LO) in an ISTQB syllabus has a cognitive level. The exam question for that LO is written at that level, so the K-level tells you how to study it.

In CT-AI v2.0 most objectives are K2. The K3 ones you must be able to do include calculating ML functional performance metrics from a confusion matrix (AI-3.3.1), implementing red teaming for GenAI systems (AI-4.2.2) and applying dataset constraint testing (AI-5.1.5). In CT-GenAI the whole of section 2.2 (applying prompting to test analysis, test design, regression testing, monitoring and choosing techniques) is K3, as is identifying hallucinations, reasoning errors and biases in LLM output (GenAI-3.1.2). Keywords are always examinable at K1 even if no LO names them.

The syllabi also contain hands-on objectives (H0 = watch a demo, H1 = guided exercise, H2 = exercise with hints under a time limit). They are not examined directly, but questions often describe exactly those exercises, so doing them makes scenario questions much easier.

Chapter weights and a 6-week plan

Training time per chapter is the best public proxy for how much the exam cares about it. For exact question distribution, check the exam structure documents on istqb.org.

1. Week 1: CT-AI chapters 1-2 plus AI vocabulary (this lesson and lesson 2). Download both syllabi and both official sample exams.
2. Week 2: CT-AI chapter 3 - metrics and neural networks with code (lesson 3). Do 20 confusion-matrix calculations by hand.
3. Week 3: CT-AI chapters 4-7 (lessons 4-5). Take the official CT-AI sample exam under a 60-minute timer.
4. Week 4: CT-GenAI chapters 1-2 (lesson 6). Practise prompts on a local model daily.
5. Week 5: CT-GenAI chapters 3-5 (lessons 7-8). Take the official CT-GenAI sample exam.
6. Week 6: Mock drills, weak-LO revision and booking (lessons 9-10). Book the first exam only when you score 75%+ twice in a row on unseen questions.

AI and ML fundamentals every tester needs on day one

CT-AI chapter 1 (all K2) covers the vocabulary that both exams take for granted:

• AI-based vs conventional systems: conventional systems follow logic written by developers; ML-based systems learn behaviour from data, so the 'specification' is partly the training data.
• Narrow, general and super AI: all deployed systems today are narrow AI - good at one task. General AI (human-level across tasks) and super AI remain theoretical.
• AI technologies: examples include rule-based/expert systems, search and optimisation, ML (including deep learning), and generative models.
• Generative AI: models that produce new content (text, images, code). LLMs are a type of generative model built on the transformer architecture.
• ML hardware: CPUs, GPUs and specialised accelerators differ in parallelism, cost and energy use - training usually needs far more compute than inference.
• Development and hosting: on-premise, cloud, AI-as-a-service APIs and edge devices - each changes what and where you can test.
• ML frameworks: libraries such as scikit-learn, TensorFlow and PyTorch provide data handling, training, evaluation and deployment support.
• Regulations and standards: e.g. the EU AI Act (risk-based obligations), ISO/IEC 25059 (quality model for AI systems), ISO/IEC 42001 (AI management systems), NIST AI RMF. In India, the DPDP Act 2023 governs personal data used to train or test systems.

Quick self-check snippet you can run to see the difference between a rule and a learned model:

# Conventional: the rule is written by a developer
def is_high_value_rule(amount_inr):
    return amount_inr >= 50000

# ML-based: the "rule" is learned from labelled examples
from sklearn.tree import DecisionTreeClassifier
X = [[1000], [5000], [20000], [48000], [52000], [90000], [150000]]
y = [0, 0, 0, 0, 1, 1, 1]          # labels from historical decisions
model = DecisionTreeClassifier(random_state=42).fit(X, y)
print(model.predict([[49000], [51000]]))  # the threshold came from data, not a spec

The tester's takeaway: when behaviour is learned, you test the data, the model and the system - not just the code.

Real-world assignment

Publish a one-page 'ISTQB AI certification study plan' on your GitHub or LinkedIn: the two exam blueprints in a table, your baseline sample-exam score, your 6-week calendar and your LO tracker CSV. Hiring managers in Indian product companies and remote global teams notice testers who plan and measure their own upskilling.