AIF-C01 surprises people in both directions. Some assume that because it is foundational, they can walk in cold and pass on general AI awareness. They usually fail on the service mapping questions and the responsible AI vocabulary. Others assume it will demand SageMaker configuration and Python, then relax once they realize the exam never asks them to train a model or tune a hyperparameter. The challenge sits in the middle: AWS wants you fluent in generative AI concepts and able to match the right AWS service to a business scenario, using precise terminology, without ever touching code.
AIF-C01 is built from five scored domains, and the weighting is lopsided enough that it should shape your entire study plan.
Fundamentals of AI and ML, 20% — distinguishing AI, ML, deep learning, and generative AI from each other; supervised versus unsupervised versus reinforcement learning; basic terms like training, inferencing, bias, and fit that show up as trick questions when a scenario is worded to sound like one type of learning but is actually another.
Fundamentals of Generative AI, 24% — how large language models work at a conceptual level, foundation model characteristics, prompt engineering styles (zero-shot, few-shot, chain-of-thought), and where generative AI fits against traditional ML.
Applications of Foundation Models, 28% — the single largest domain, covering model selection trade-offs, fine-tuning versus prompt engineering versus Retrieval-Augmented Generation, Amazon Bedrock, Bedrock Agents, Bedrock Guardrails, and knowing when each approach is appropriate for a stated business problem.
Guidelines for Responsible AI, 14% — fairness, bias detection, explainability, and AWS tools that support responsible AI practices such as SageMaker Clarify.
Security, Compliance, and Governance for AI Solutions, 14% — data privacy for training and inference data, IAM controls around AI services, and governance frameworks for AI workloads.
Notice that Applications of Foundation Models and Fundamentals of Generative AI together make up 52% of the scored exam. If you spend equal time across all five domains, you are underinvesting in the half of the exam that decides whether you pass.
AIF-C01 draws a wider and less technical crowd than most AWS certifications. Solutions architects and developers take it, but so do product managers, business analysts, and project leads who need to speak credibly about AI initiatives without building the models themselves. A large share of candidates have never touched SageMaker or Bedrock in a console before studying for this exam, and AWS built it with exactly that audience in mind. There is no formal prerequisite, though AWS suggests some prior exposure to AI and ML concepts helps. People coming from a Cloud Practitioner background tend to find the AWS service knowledge familiar and the generative AI material genuinely new.
The official AIF-C01 exam runs 90 minutes for 65 questions, of which 50 are scored and 15 are unscored trial questions that AWS uses to evaluate future content. You cannot tell the scored questions apart from the unscored ones, so every question deserves full attention. Beyond standard multiple choice and multiple response, AIF-C01 also uses ordering questions, where you sequence three to five steps correctly, and matching questions, where you pair items from two lists. Both formats give you no partial credit unless every pairing or every position is correct, which is a different kind of pressure than picking one answer from four options. Scoring is scaled from 100 to 1,000, and you need 700 to pass. The exam costs 100 USD.
The ordering and matching formats are where candidates who only memorized flashcard-style facts tend to lose points, since these formats require you to understand a full process or a full set of relationships rather than recognize one correct fact in isolation.
Do not just check whether you got a question right. For the Applications of Foundation Models questions especially, work out why RAG was the correct choice over fine-tuning, or why a smaller foundation model was preferred over a larger one given the stated latency or cost constraint. That is the actual skill AWS is testing in this domain, not memorized service names.
Pay close attention to your accuracy on ordering and matching style questions specifically, since they behave differently from single-answer multiple choice and are easy to underestimate in casual review. If you are consistently strong on straightforward multiple choice but shaky on ordering questions, that is a format problem, not a knowledge gap, and it needs targeted practice with that specific question type.
The most common mistake is confusing prompt engineering with fine-tuning as strategies for improving model output, when the exam expects you to know which one fits a stated constraint, such as needing a fast, low-cost adjustment versus needing the model to reliably use a company's own proprietary data.
A second common mistake is underestimating the responsible AI and governance domains because they feel less technical than the generative AI material. Together these two domains make up 28% of the exam, equal to the single largest individual domain, and candidates who skip them to focus purely on Bedrock and foundation models often lose enough points there to fail despite strong technical answers elsewhere Try a CLF-C02 practice exam practice exam for free.
A third mistake is treating this exam like Cloud Practitioner. AIF-C01 assumes you already understand basic AWS structure and instead tests genuinely new generative AI vocabulary and reasoning. Candidates who study it the same way they studied Cloud Practitioner, by memorizing service names and one-line definitions, tend to struggle on the scenario-based Applications of Foundation Models questions.
Learn to classify a scenario correctly before you look at the answer choices. A large portion of the Fundamentals of AI and ML domain is testing whether you can identify supervised classification, unsupervised clustering, or reinforcement learning from a business description before you ever see the options.
Get comfortable with the actual mechanics of Retrieval-Augmented Generation, since it shows up repeatedly across the Applications of Foundation Models domain in different guises. Understand what a knowledge base is doing in that pattern and why it reduces hallucination compared to relying on the model's training data alone.
Read AWS's own responsible AI documentation rather than relying only on definitions from third-party study guides. AWS tests its own specific framing of fairness, explainability, and bias, and that framing is what the exam questions are built from.
Work through full 65-question sets under a 90-minute clock so the ordering and matching formats stop feeling unfamiliar on exam day. Track your score by domain, and if Applications of Foundation Models or Fundamentals of Generative AI is not clearing 80%, go back to those two domains before touching anything else, since together they decide more than half your result.
Last updated on Oct, 9 2026
