The AWS Certified AI Practitioner is the cert everyone is suddenly interested in, and most of the study advice for it is guesswork. The exam is still relatively new, so the internet is full of people describing what they think is on it.
You don’t have to guess. AWS publishes an exam guide with exact domain weights. Study those weights, in that order, and you skip the part where you read forty Reddit threads to reverse-engineer the blueprint.
Here’s the plan.
What the Exam Actually Covers
AIF-C01 is a foundational exam. It tests whether you understand AI, machine learning, and generative AI concepts and how they apply on AWS. It does not require you to write code or train models. It has five domains, and the weights are lopsided in a way that should shape your study time.
- Applications of Foundation Models: 28%. The single biggest domain. Prompt engineering, retrieval-augmented generation (RAG), when to fine-tune versus when to use a base model, evaluating model outputs, and the building blocks in Amazon Bedrock (knowledge bases, agents). If you’re weak here, you’re giving away more than a quarter of the exam.
- Fundamentals of Generative AI: 24%. Foundation models, tokens, embeddings, what generative AI is good at and where it fails (hallucination, non-determinism), and the AWS services that deliver it: Amazon Bedrock and Amazon Q.
- Fundamentals of AI and ML: 20%. The base layer. Supervised, unsupervised, and reinforcement learning; the difference between AI, ML, and deep learning; common use cases; and the AWS AI service catalog (SageMaker, plus the managed services like Comprehend, Rekognition, Textract, Transcribe, and Polly).
- Guidelines for Responsible AI: 14%. Bias, fairness, transparency, and explainability, plus the AWS tooling that addresses them: Guardrails for Amazon Bedrock and SageMaker Clarify.
- Security, Compliance, and Governance for AI Solutions: 14%. IAM for AI services, data governance, encryption, and how compliance obligations apply to AI workloads.
Add it up: the two generative-AI domains together are 52% of the exam. This is an AI Practitioner cert in the generative-AI era, not a classical machine-learning cert. Weight your time accordingly. Someone who spends half their study time on traditional ML algorithms is preparing for a different exam than the one AWS wrote.
The Format
65 questions. 90 minutes. That’s a little under 90 seconds per question, which is plenty for a foundational exam where the questions are conceptual rather than architectural puzzles.
Of those 65 questions, only 50 are scored. The other 15 are unscored questions AWS uses to trial future content. You won’t know which is which, so answer everything as if it counts.
65 questions in 90 minutes, per the AWS exam page. Multiple choice and multiple response are the formats you will see most, but AWS’s exam guide lists further types, so read it rather than assuming only those two. No labs, no code.
Pass score: 700 out of 1000, on a scaled scoring model. AWS uses compensatory scoring, which means you don’t need to pass each domain individually. You only need to clear the overall bar. Don’t try to compute “how many can I miss” from the 700 number; the scaling makes that math unreliable. Aim to be comfortably above the line.
Cost: $100 USD per attempt. Foundational level. The certification is valid for 3 years.
Study Timeline
Plan for about 20 to 30 hours. This is a foundational exam, and over-preparing for it is the most common mistake. How you distribute those hours depends on where you start.
If you already work with AWS (you’ve used the console, you know what IAM and S3 are): 2–3 weeks. Your gap is generative-AI vocabulary and the Bedrock service family, not cloud basics.
If you know AI/ML but not AWS: 2–3 weeks. You understand embeddings and fine-tuning already. What you need is which AWS service does what, and the exam tests service recognition heavily.
If you’re new to both: 4–6 weeks. Build the conceptual base first (what a foundation model is, what RAG solves), then map each concept to its AWS service. Don’t try to memorize the service list cold before the concepts make sense.
If you find yourself past six weeks on a foundational exam, the problem isn’t time. Check your domain scores and find the one that’s dragging.
Common Pitfalls
Studying Classical ML Too Hard
The name says “AI Practitioner,” and people assume that means deep coverage of algorithms, model training, and statistics. It doesn’t. Traditional ML fundamentals are one 20% domain. The generative-AI material is more than half the exam. If you spend three weeks on gradient descent and neural network architecture, you’ve optimized for the wrong blueprint.
Confusing the Bedrock Building Blocks
The exam leans hard on knowing which Amazon Bedrock feature solves which problem, and they’re easy to blur together:
- Prompt engineering: change the input to change the output. No infrastructure.
- RAG (retrieval-augmented generation): give the model your own data at query time, without retraining. This is a Bedrock knowledge base.
- Fine-tuning: further-train a model on your labeled data to change its behavior. More effort, more cost.
- Agents: let the model call tools and take multi-step actions.
When a scenario says “the model needs access to our internal documents but we can’t retrain it,” the answer is RAG, not fine-tuning. Get these four distinctions cold and you’ve locked in a large slice of the 28% domain.
Skipping Responsible AI as “Soft”
Responsible AI and Security are 14% each, 28% together, and candidates treat them as filler. They’re not. Know what bias and explainability mean in practice, what Guardrails for Amazon Bedrock does (blocks harmful content, filters topics), and what SageMaker Clarify measures. These are easy points that people leave on the table because the topics sound less technical than RAG.
Study Approach
Study by domain, in weight order, heaviest first. Don’t read the Bedrock docs front to back.
Start with a diagnostic AIF-C01 practice exam to see where you actually stand. Most people are surprised: they assume they’re weak on responsible AI and score lowest on the foundation-model applications, or the reverse.
Then allocate time proportionally. If Applications of Foundation Models is both your weakest area and 28% of the exam, that’s where your first week goes. Don’t grind a domain you’re already scoring 85% on. For each domain, follow the same loop: read the relevant concept, take questions in that domain, review every explanation (including the ones you got right), repeat. Adaptive practice, where difficulty tracks your level, makes that loop faster than a static question bank.
When to Take the Exam
Book when your domain scores sit above 80% consistently, across all five domains, for three sessions in a row. Not one good day.
If you’re above 80% overall but one domain keeps dipping to 70%, you’re not ready yet. On a compensatory-scored exam a weak domain won’t fail you by itself, but a bad question draw in your weak area plus ordinary variance can put you under the line.
Improving your weakest high-weight domain gives the best return. Going from 75% to 85% in Applications of Foundation Models (28%) is worth more than polishing Security from 85% to 95% (14%).
Book before you feel “100% ready.” That feeling never arrives. If the numbers say you’re ready, trust the numbers.
After the Exam
If you pass: you’ve got three years before renewal, and a foundational cert that’s genuinely current with where the industry is. From here, the Machine Learning Associate or the broader cloud roadmap is the natural next step if you want depth.
If you don’t pass: wait the required period, read your score report (it breaks down performance by domain), fix the weak domain, and retake. Most people who adjust their plan based on the score report pass the second time.
FAQ
Is the AWS AI Practitioner worth it in 2026? For a foundational cert, yes. It’s the one credential that signals you understand generative AI in an AWS context, and demand for that is real right now. It won’t make you an ML engineer, but it was never meant to.
Do I need coding or math? No. AIF-C01 is conceptual. You need to understand what the services do and when to use them, not implement them.
Should I take Cloud Practitioner (CLF-C02) first? Not required. There’s overlap in the cloud-basics material, but AIF-C01 stands on its own. If you’re brand new to AWS, doing CLF-C02 first gives you the cloud vocabulary, but it’s optional.
How do I know when I’m actually ready? Domain scores above 80%, sustained across multiple sessions. Stop waiting for a feeling. Look at the data.
