Most software testers already know how to check a login form or a payment flow. Testing a machine learning model that changes its own behavior over time is a different skill, and that is exactly what the ISTQB Certified Tester AI Testing (CT-AI) exam checks for. This guide walks through who the certification is for, what the exam actually covers, and how to prepare for it in the weeks before test day.
What the ISTQB CT-AI Certification Actually Tests
CT-AI version 2.0 is a specialist certification from ISTQB built around testing AI based systems, including machine learning models and generative AI systems such as large language models. It does not teach you to build models. It teaches you to question them, to design test cases for data and behavior that do not follow a fixed set of rules, and to recognize the quality problems that are unique to AI.
That distinction matters more than it sounds. A traditional functional test case usually has one correct output for a given input, and a tester can write a pass or fail check against that output with confidence. An AI model trained on data can behave differently after a retraining cycle, can produce a slightly different answer to the same prompt twice, and can degrade quietly in ways that a fixed test script never catches. CT-AI exists because testing that kind of system needs its own vocabulary, its own risk categories, and its own test design approach.
The primary audience is testers, test analysts, and test engineers who are being asked to validate AI features inside products that used to be purely rule based. Test managers, consultants, data analysts, data scientists, and developers working on AI components also sit for it, along with user acceptance testers who need to sign off on AI driven features before release. Project managers, quality managers, business analysts, and IT leadership who need a working vocabulary for AI quality make up a smaller secondary group, often so they can ask the right questions in a planning meeting rather than run the tests themselves.
If you work anywhere near a product team that has added a chatbot, a recommendation engine, or an automated document reader in the last year or two, there is a good chance someone on that team already needs this vocabulary. CT-AI gives that person a structured way to learn it instead of picking it up piecemeal from blog posts and vendor documentation.
It also helps to be clear about what CT-AI is not. It is not a data science certification, and it does not ask you to tune hyperparameters or choose between model architectures. It stays firmly on the testing side of the line: given an AI based system, what should a tester check, in what order, and with what evidence that the check actually passed. Keeping that boundary in mind while you study saves you from wandering into data science material that the exam never actually asks about.
Why You Need the CTFL Before You Sit for CT-AI
CT-AI is not an entry level exam. ISTQB requires the Certified Tester Foundation Level (CTFL) certification before you can register for CT-AI. That single requirement changes how you should plan your study time, because it assumes you already know the base testing vocabulary: test levels, test types, defect lifecycle, and the difference between verification and validation.
If your CTFL feels rusty, do not skip straight to AI specific material. Spend a few days refreshing the foundation terms first, because the CT-AI syllabus reuses them constantly without re-explaining them. The syllabus authors assume you can already define terms like test basis, test oracle, and entry and exit criteria without looking them up, and a scenario question can lose you points simply because a foundation term slowed you down, not because the AI concept itself was unfamiliar.
ISTQB Performance Testing exam format is a useful reminder of how ISTQB specialist exams build directly on CTFL concepts rather than restating them, and it is worth a read if you want to see the pattern before you hit it in CT-AI. Both specialist tracks assume fluency in the same base vocabulary and then layer a new domain on top of it, which is exactly the shape CT-AI takes once you get past its introduction chapter.
Treat the CTFL prerequisite as a planning constraint, not a formality. If your CTFL certification needs renewal, or if it has simply been a few years since you last used the terminology day to day, build that refresh into week one of your schedule rather than assuming it will come back to you automatically once you open the CT-AI material.
Inside the Exam: 40 Questions, 44 Points, 60 Minutes
The CT-AI exam has 40 questions worth 44 points in total, because a handful of questions carry more than one point. You need 29 of those 44 points to pass. The standard time allowance is 60 minutes, extended to 75 minutes for candidates who are not sitting the exam in their native language.
That extra time allowance matters more than most candidates expect. AI testing questions often describe a scenario first, such as a model's behavior drifting after a data update, and then ask you to pick the correct testing response. Reading the scenario carefully costs time that a straightforward recall question would not, and a candidate who treats every question like a quick definition check will run short on time well before the end of the exam.
Because some questions are worth more than one point, it also pays to resist the urge to rank questions purely by how long they took you to read. A short scenario can still carry two points if it tests more than one concept at once, so do not assume length and point value always move together. Read the question stem fully before deciding how much time it deserves.
The exam is scheduled through Pearson VUE, delivered under ISQI's exam administration. Budget for an exam fee in the neighborhood of 199 US dollars, though the exact price depends on your region and your chosen exam provider, so always confirm the current fee before you book a date. Booking early also gives you more flexibility in choosing a testing center or an online proctored slot, which matters if your work schedule is tight in the weeks before your target date.
Seven Chapters That Make Up the CT-AI Syllabus
The CT-AI v2.0 syllabus is organized into seven chapters, and it helps to group them into five study pillars rather than attack all seven as separate, unrelated topics. Treating seven chapters as seven disconnected reading assignments is the single most common reason candidates feel overwhelmed partway through their preparation.
The first pillar covers the foundation: Introduction to Artificial Intelligence and Quality Characteristics for AI Based Systems. This is where you learn the AI vocabulary and the quality attributes, such as those tied to ISO/IEC 25059, that do not apply to traditional software. Expect terms describing fairness, robustness, and explainability to show up here, each with its own testing implications that later chapters build on.
The second pillar is Machine Learning on its own, covering data preparation, functional performance metrics, and neural network basics. This chapter asks you to understand enough about how a model is trained and scored to know what a tester should check at each stage, even if you are never the one writing the training code yourself.
The third pillar, Testing AI Based Systems, is where generative AI and large language models get direct attention, along with how test levels shift when a system learns instead of following fixed logic. This is often the chapter candidates find most relevant to their day to day work, since it covers the newer systems many teams are shipping right now.
The fourth pillar combines Input Data Testing for Machine Learning Systems and Model Testing for Machine Learning Systems, since both chapters deal with checking what goes into a model and what comes out of it. Input data testing focuses on the quality, completeness, and representativeness of training and test data, while model testing focuses on evaluating the trained model's behavior against defined metrics and edge cases.
The fifth pillar, Machine Learning Development Testing, closes the syllabus by looking at testing activities across the ML development pipeline itself, from experimentation through to deployment and monitoring. It ties the earlier chapters together by showing where each kind of test fits inside a real development workflow.
Treat these five pillars as five separate study blocks rather than reading the syllabus start to finish in one pass. Each pillar builds a different mental model, and mixing them together in one study session makes it harder to recall which concept belongs where on exam day. Candidates who study pillar by pillar, with a short review before moving to the next one, consistently report feeling more confident walking into the exam than those who read the whole document once and hoped it would stick.
Why AI Testing Skills Matter for Your Career Right Now
Generative AI features are showing up inside products that had nothing to do with AI two years ago, from customer support chat to document search to recommendation engines. Someone on the test team has to validate those features, and traditional test case design does not fully cover a system whose output can change based on training data or a model update.
Holding CT-AI signals to a hiring manager that you can design tests for that kind of system without needing a crash course first. For testers who already hold CTFL and want to specialize, it is a natural next step that opens roles on teams building or integrating AI features, rather than leaving that work entirely to data scientists who may not think about quality the way a trained tester does.
It also pairs well with other ISTQB specialist tracks if you are mapping out a longer certification path. ISTQB Finance Testing certification is one example of how ISTQB specialist certifications let you stack a domain specific or technology specific credential on top of CTFL, the same way CT-AI does for AI systems. Testers who hold more than one specialist certification often find they have more options when a team is deciding who leads testing on a new, higher risk feature.
There is also a practical argument for employers here, which is worth knowing if you need to justify the exam fee and study time to a manager. A tester who already understands AI specific quality characteristics can catch problems earlier in a project, before a flawed model reaches production and becomes a much more expensive fix. That argument tends to land well in a budget conversation, since it frames the certification as risk reduction rather than a personal development expense.
Longer term, the vocabulary from CT-AI tends to stay useful even as specific tools and model versions change. The underlying questions, such as how representative the training data is or how a system should behave on an input it has never seen, do not go away when a new model release ships. That durability is part of why the certification is worth the time investment even in a field that otherwise moves quickly.
A 30-Day Study Plan for the CT-AI Syllabus
Thirty days is enough time to work through the CT-AI syllabus properly if you follow a fixed weekly structure instead of studying whatever feels interesting that day. Treat each week as a checkpoint, and resist the temptation to skip ahead before you can summarize the previous week's pillar from memory.
In week one, refresh your CTFL vocabulary and then read the Introduction to Artificial Intelligence and Quality Characteristics chapters twice. Write your own one line definition for every new term instead of copying the syllabus wording, since rephrasing forces you to actually understand it rather than just recognize it on a second read.
In week two, move into the Machine Learning chapter and the Testing AI Based Systems chapter together, since they reference each other constantly. Pay close attention to how testing a generative AI system differs from testing a traditional classifier, because the exam likes to test that distinction directly, often by describing a scenario and asking which approach applies.
In week three, work through Input Data Testing and Model Testing for Machine Learning Systems, then close with Machine Learning Development Testing. These three chapters are the most technical in the syllabus, so slow down here rather than rushing to finish the reading. If a concept does not make sense on the first pass, re-read the relevant section of the official syllabus rather than guessing at the meaning and moving on.
In week four, stop reading new material entirely and sit two full timed practice exams under real exam conditions, meaning a quiet room, a visible clock, and no pausing partway through. Review every missed question by chapter, not just by counting your score, so you know exactly which pillar needs one more pass before test day. If one pillar accounts for most of your missed questions, give it a dedicated evening rather than a quick skim.
Exam-Day Strategy and Common Pitfalls
With 40 questions and 60 minutes, you have roughly a minute and a half per question, though some scenario based questions will take longer than others. Skim the full exam first and answer the shorter recall questions immediately, then come back to the longer scenario questions with whatever time remains. This approach protects you from running out of time on the easy points because you got stuck early on one difficult scenario.
A common mistake is treating CT-AI like a pure knowledge test and skipping scenario practice entirely. The exam frequently describes a situation, such as a model's accuracy dropping after retraining, and asks which testing activity catches that problem. Candidates who only memorized definitions struggle with these questions because they never practiced applying the vocabulary to a situation, and recognizing a term in isolation is a different skill from recognizing which term applies inside a described scenario.
Another frequent pitfall is confusing data testing with model testing, since both chapters sit next to each other in the syllabus and cover similar sounding ground. Keep a short note comparing the two as you study, listing what each one checks and when it happens in the pipeline, so the distinction stays clear under exam pressure. A simple rule of thumb many candidates use is that data testing happens before training and model testing happens after it, though the syllabus goes into more nuance than that shorthand captures.
A final pitfall worth naming: non native English speakers sometimes forget to request the extended 75 minute sitting when they register. If English is not your first language, confirm that extension is applied to your booking before exam day, not after you are already seated, since fixing it on the day itself is rarely possible.
It is also worth pacing your review at the end of the exam if time remains. Rather than re-reading every question from the start, flag the two or three you were least confident about as you go, and spend your remaining minutes only on those, since blanket re-reading tends to waste time without changing many answers.
How Practice Tests Close the Preparation Gap
Reading the syllabus twice is not the same as answering exam style questions under time pressure. Practice questions expose the gaps between knowing a definition and applying it to a scenario, which is exactly the skill CT-AI is built to test, and that gap only becomes visible once you try to apply a concept under a time limit.
CT-AI sample practice questions from ProcessExam give you a factual sense of question style and difficulty before you commit to a real exam date. Work through a batch after each study pillar rather than saving all your practice for the final week, so you catch weak spots while there is still time to revisit the chapter instead of discovering them the night before your exam.
If you are also weighing how to talk about AI testing skills in an interview or a portfolio, a decision many testers get wrong is worth reading, since it breaks down a choice point in AI test design that practice questions alone will not teach you, particularly around how to frame your experience to someone evaluating your skills rather than grading an exam.
Before you register, read through the official CT-AI certification page one more time to confirm the current exam structure has not changed, and keep the official CT-AI syllabus open as your primary reference while you study rather than relying only on secondary summaries. Secondary sources, including this guide, are useful for structure and planning, but the syllabus itself remains the authoritative description of what the exam can ask.
Frequently Asked Questions
Do I need the CTFL certification before I can take CT-AI?
Yes. ISTQB requires the Certified Tester Foundation Level certification as a prerequisite for CT-AI. Make sure that certification is active before you register for the AI testing exam.
How many questions are on the CT-AI exam and how long do I get?
The exam has 40 questions worth 44 points, and you need 29 points to pass. You get 60 minutes, or 75 minutes if you are not testing in your native language.
Does CT-AI require a programming or data science background?
No. CT-AI is written for testers and other quality roles, not for people building the models. You need to understand testing concepts and AI vocabulary, not write machine learning code.
Which chapters of the CT-AI syllabus carry the most weight on the exam?
The official CT-AI page lists the seven chapters but no percentage weight for each one. Study all seven chapters evenly, and give extra review time to the ones that feel least familiar.
How long should I study before attempting the CT-AI exam?
A focused 30 day plan that covers each syllabus pillar once and ends with two timed practice exams is realistic for most candidates who already hold CTFL. Extend that if AI concepts are new to you.
Final Thoughts
CT-AI rewards candidates who treat AI testing as its own discipline rather than an extension of rules they already know. Work the syllabus in pillars, practice on real scenario questions, and confirm your exam logistics early, and the 60 minutes on test day will feel like a formality rather than a surprise.