AIF-C01
AWS Certified AI Practitioner Exam(AI1-C01)
AIF-C01 Exam Guide: What to Study, How to Prepare, and When to Schedule
AWS Certified AI Practitioner (AIF-C01) validates foundational knowledge of AI, machine learning, generative AI, AWS AI tools, business use cases, and responsible implementation. It is aimed at candidates with up to 6 months of exposure to AI/ML technologies on AWS who use, but do not necessarily build, these solutions. This guide helps you decide what the blueprint demands, which AWS concepts deserve hands-on attention, how to sequence study, and whether you are ready to schedule the exam.
What does AIF-C01 validate?
AIF-C01 tests whether you can connect AI concepts and AWS technologies to business problems without requiring you to develop models or build production ML infrastructure. The central decision is not how to code an algorithm; it is which approach fits a use case, what tradeoffs it creates, and how to use it responsibly.
AWS describes the certification as foundational and focused on practical business applications of AI. The exam validates the ability to describe AI, ML, and generative AI concepts, methods, and strategies; identify appropriate technologies for business problems; determine the correct types of AI/ML technologies for specific use cases; and use those technologies responsibly.
That positioning matters when you plan your preparation. A candidate who spends most study time implementing neural networks, tuning hyperparameters, or constructing an MLOps pipeline is working beyond the stated target. A candidate who can explain why a classification model, foundation model, managed API, or retrieval-augmented application fits a particular business requirement is studying closer to the exam’s intended level.
Who is the intended candidate?
The target candidate should have up to 6 months of exposure to AI/ML technologies on AWS. AWS says the candidate uses, but does not necessarily build, AI/ML solutions on AWS. This makes the certification relevant to business analysts, product and project professionals, cloud practitioners, developers, operations staff, and technical managers who need a common working understanding of AI.
AWS recommends familiarity with core AWS services and their use cases, the shared-responsibility model, IAM, and AWS service pricing models. You do not need to treat those topics as separate cloud certification subjects, but you should be able to recognize how they affect an AI solution’s security, access, cost, and operating choices.
What is outside the expected scope?
Developing or coding AI/ML models or algorithms is out of scope for the target candidate. AWS also lists implementing data or feature engineering, hyperparameter tuning, model optimization, building and deploying AI/ML pipelines or infrastructure, mathematical or statistical analysis of models, implementing security or compliance protocols, and developing governance frameworks as out of scope tasks.
Out of scope does not mean that related vocabulary can be ignored. You may still need to identify where fine-tuning belongs in a foundation-model lifecycle, understand what model monitoring is intended to achieve, or recognize why governance is needed. Learn the purpose, selection factors, and business consequence rather than preparing to perform the engineering task.
How is the exam structured?
AWS lists AIF-C01 as a 90-minute exam with 65 questions. The exam includes 50 questions that affect your score and 15 unscored questions that AWS uses to evaluate possible future content. Because you cannot identify which questions are unscored, treat every question as a real opportunity to demonstrate understanding.
Your result is reported as a scaled score from 100–1,000, and the minimum passing score is 700. The official guide warns that section-level feedback should be interpreted cautiously, so use domain feedback as a study signal rather than assuming it maps directly to a precise percentage of correct answers.
Which question formats require the most care?
The exam may include multiple-choice, multiple-response, ordering, and matching questions. Multiple-choice questions have one correct response and three distractors. Multiple-response questions require all correct responses to receive credit. Ordering questions require the correct responses in the correct order, while matching questions require every pair to be matched correctly.
Unanswered questions are scored as incorrect, and AWS states there is no penalty for guessing. Your practical rule should therefore be simple: answer every question, mark uncertain items for review when the interface permits, and return to them only after securing questions you can solve confidently. For multiple-response items, evaluate each option independently instead of stopping at the first plausible answer.
Where can you take AIF-C01?
AWS states that the exam can be taken at a Pearson VUE testing center or as an online-proctored exam. The available choice is a scheduling decision, not a content advantage. Select the environment in which you can reliably meet the provider’s identification, equipment, workspace, and check-in requirements, then verify the current appointment conditions before booking.
AWS lists Arabic, English, French, German, Italian, Japanese, Korean, Portuguese (Brazil), Spanish (Latin America and Spain), Simplified Chinese, and Traditional Chinese as offered exam languages. Confirm the language and current appointment availability through the official certification page when you schedule, because delivery information can change.
What does each exam domain require?
The blueprint rewards breadth, but not every domain deserves identical study time. Content Domain 3: Applications of Foundation Models represents 28% of scored content, Content Domain 2: Fundamentals of GenAI represents 24%, and Content Domain 1: Fundamentals of AI and ML represents 20%. Content Domain 4: Guidelines for Responsible AI and Content Domain 5: Security, Compliance, and Governance for AI Solutions each represent 14% of scored content.
Use the domain labels whenever you plan your study. A percentage without its domain is not a useful prioritization rule: the larger domains contain different decisions, and the two 14% domains cover risks that can invalidate an otherwise attractive solution. The following breakdown turns the blueprint into study tasks.
Domain 1: Fundamentals of AI and ML
Content Domain 1: Fundamentals of AI and ML represents 20% of scored content and covers basic terminology, practical use cases, and the AI/ML development lifecycle. Build a clean vocabulary for AI, ML, deep learning, neural networks, computer vision, NLP, training, inference, bias, fairness, LLMs, GenAI, and agentic AI.
You should be able to distinguish supervised, unsupervised, and reinforcement learning; labeled and unlabeled data; structured and unstructured data; and batch, real-time, asynchronous, and serverless inference. Study the relationship between a problem and a technique: regression predicts a continuous value, classification assigns categories, and clustering groups similar items without predefined labels.
The domain also asks when AI/ML is not appropriate. Include cost-benefit analysis and situations requiring a specific outcome rather than a prediction in your notes. A deterministic rule can be preferable when the requirement is exact, simple, auditable, or cheaper to implement. For model evaluation, connect accuracy, precision, recall, and F1 score to the type of error a business can tolerate, then connect those technical measures to cost per user, development cost, customer feedback, and ROI.
Review the lifecycle from data and problem definition through training, evaluation, deployment, monitoring, and retraining. Add the AWS services named in the objectives to the stage where they are relevant, including Amazon Bedrock, Amazon SageMaker AI, Amazon Transcribe, Amazon Translate, Amazon Comprehend, Amazon Lex, and Amazon Polly. The objective is service recognition and use-case selection, not implementation.
Domain 2: Fundamentals of GenAI
Content Domain 2: Fundamentals of GenAI represents 24% of scored content. Learn how tokens, chunking, embeddings, vectors, prompts, transformer-based LLMs, foundation models, multimodal models, diffusion models, and agents fit together. Be able to map a business request such as summarization, translation, image generation, code generation, search, or customer service to the appropriate kind of GenAI capability.
The domain emphasizes both capability and limitation. GenAI can be adaptable, responsive, conversational, and effective at generating content, but it can also hallucinate, produce inaccurate or nondeterministic responses, and be difficult to interpret. A strong answer weighs those limitations against the consequence of error, the need for human review, compliance requirements, latency, cost, and the required level of control.
Model selection is a tradeoff among model type, capability, constraints, performance, compliance, cost, latency, and complexity. Revise the effect of token-based pricing on inference cost and performance, along with tradeoffs involving responsiveness, availability, redundancy, regional coverage, provisioned throughput, and custom models.
Know the role of AWS services and features named in the domain objectives, including Amazon Bedrock, Amazon SageMaker AI, SageMaker JumpStart, Amazon Quick, Kiro, Strands Agents, and Amazon Bedrock AgentCore. Study why managed AWS services can lower the barrier to entry, improve efficiency, support speed to market, and help meet business objectives, while still requiring appropriate security, compliance, and cost decisions.
Domain 3: Applications of Foundation Models
Content Domain 3: Applications of Foundation Models represents 28% of scored content, the largest domain. It covers foundation-model design, prompt engineering, training and fine-tuning, and performance evaluation. Make this the anchor of your study plan, especially if your prior experience is mainly with traditional ML or general AWS services.
For model selection, compare cost, modality, latency, multilingual capability, model size, complexity, customization, input and output length, and prompt caching. Understand how inference parameters such as temperature and input/output length affect responses. You should be able to explain RAG, why it can ground responses in relevant data, and how Amazon Bedrock Knowledge Bases and vector stores such as Amazon OpenSearch Service, Amazon Aurora, Amazon Neptune, and Amazon RDS for PostgreSQL can support retrieval.
Separate customization choices rather than treating them as interchangeable. In-context learning changes what is supplied in the prompt; RAG retrieves external information; fine-tuning adapts a model using prepared examples; pre-training is a much broader training effort; and model distillation transfers behavior to a smaller model. The exam focus is the rationale and cost tradeoff, not the implementation procedure.
Prompt engineering requires more than writing a longer instruction. Revise context, instructions, negative prompts, zero-shot, single-shot, few-shot, chain-of-thought, templates, specificity, concision, experimentation, guardrails, and prompt management. Also learn the risks of prompt exposure, poisoning, hijacking, and jailbreaking, along with the purpose of Amazon Bedrock Prompt Management.
Evaluation should combine technical and business evidence. Know human-in-the-loop evaluation, benchmark datasets, Amazon Bedrock Model Evaluation, ROUGE, BLEU, BERTScore, and LLM-as-a-judge, while recognizing that no single metric proves that an application meets its objective. For RAG, agents, and workflows, assess retrieval quality, response quality, task completion, user satisfaction, productivity, efficiency, cost per interaction, conversion, or another metric tied to the business outcome.
Domain 4: Guidelines for Responsible AI
Content Domain 4: Guidelines for Responsible AI represents 14% of scored content. Study it as a decision framework: a useful model must also be fair, inclusive, robust, safe, truthful, and appropriate for the people affected by its outputs. Learn how bias, variance, poor data representation, overfitting, and underfitting can create different risks.
Revise dataset characteristics such as diversity, inclusivity, balanced representation, curated sources, and label quality. Human audits, subgroup analysis, and monitoring can help detect bias, trustworthiness problems, and truthfulness concerns. Amazon Bedrock Guardrails is an example of a tool used to support responsible AI controls.
Transparency and explainability are related but not identical. A transparent model or process makes relevant behavior and information easier to inspect; an explainable system helps users understand how or why an outcome was produced. Prepare for tradeoffs between interpretability and performance, and review Amazon SageMaker Model Cards and Amazon Bedrock Model Evaluations as tools named in the objectives.
The domain also includes legal and practical risks of GenAI, including intellectual-property infringement claims, biased outputs, hallucinations, loss of customer trust, and end-user risk. Human-centered design matters: provide understandable decision transparency and mechanisms for user feedback instead of assuming that a technically strong output is sufficient.
Domain 5: Security, Compliance, and Governance for AI Solutions
Content Domain 5: Security, Compliance, and Governance for AI Solutions represents 14% of scored content. Prepare to identify controls for access, data protection, privacy, threats, logging, lineage, governance, and compliance. The key distinction is between selecting or explaining a control and implementing a complete security or governance program, which AWS lists as outside the target candidate’s expected tasks.
Review IAM roles, policies, and permissions; encryption; Amazon Macie; AWS PrivateLink; the AWS shared-responsibility model; Amazon Bedrock Guardrails; and Amazon Bedrock AgentCore Identity and Policy in AgentCore. Connect security threats to mitigations: prompt injection may require input controls and workflow boundaries, while data leakage may require access control, encryption, output filtering, validation, and careful logging.
Source citation and data lineage document where information came from. Data governance also includes lifecycle, residency, retention, monitoring, observation, and audit requirements. AWS names AWS Config, Amazon Inspector, AWS Artifact, AWS CloudTrail, and AWS Trusted Advisor as services or features that can assist governance and compliance.
For hallucinations, know the purpose of grounding through RAG, output validation, and confidence scoring. For governance, revise policies, review cadence, transparency standards, team training, and the Generative AI Security Scoping Matrix. A useful study exercise is to take one application and list its data, identities, logs, model inputs, outputs, review points, and retention decisions.
Which AWS services should you actually revise?
Do not attempt to memorize an isolated catalogue of product names. Learn each service in the context of a task: storing data, building an application, selecting a model, protecting access, monitoring activity, controlling cost, or supporting governance. AWS says its in-scope list is non-exhaustive and subject to change, so use the official list as a boundary and verify current details rather than relying on an old service table.
Start with the services explicitly repeated in the domain objectives: Amazon Bedrock, Amazon SageMaker AI, SageMaker JumpStart, Amazon S3, IAM, Amazon CloudWatch, Amazon CloudTrail, Amazon Bedrock Guardrails, Amazon Bedrock Knowledge Bases, Amazon OpenSearch Service, Amazon Macie, AWS Config, AWS Artifact, Amazon Inspector, and AWS Trusted Advisor. Then review the remaining in-scope services by category and use case.
The official in-scope list includes Analytics, Cloud Financial Management, Compute, Containers, Database, Developer Tools, Machine Learning, Management and Governance, Networking and Content Delivery, Security, Identity, and Compliance, and Storage. It names services including Amazon EC2, AWS Lambda, Amazon ECS, Amazon EKS, Amazon Aurora, DynamoDB, Amazon Neptune, Amazon RDS, Amazon VPC, AWS KMS, AWS Secrets Manager, and Amazon S3.
For each service in your notes, answer three questions: what problem does it address, what decision makes it suitable, and what nearby service or approach might be confused with it? This method is more useful than copying descriptions. For example, compare a managed foundation-model access approach with a model-development environment, then ask how data access, customization, latency, and cost alter the choice.
How should you use the AWS exam guide?
Use the official exam guide as your syllabus, not as a reading assignment to finish once. Convert every task statement into a checklist and every objective into a question you can answer without notes. The domain pages provide the most useful detail for this conversion, while the in-scope-services page helps identify the AWS vocabulary that may appear across scenarios.
When a service or concept is unfamiliar, consult current AWS documentation for that item after you understand the blueprint objective. Avoid studying random product announcements or deep implementation tutorials unless they clarify an objective. The exam guide’s target level is foundational and business-oriented; your notes should reflect that level.
How should you study if you are new to AI?
Begin with concepts and decisions before product names. First learn the distinctions among AI, ML, deep learning, GenAI, foundation models, and agents; then connect those distinctions to use cases, data, inference, evaluation, cost, and risk. Only after that should you map the concepts to AWS services. This sequence reduces product memorization and improves scenario reasoning.
Create a one-page comparison sheet for traditional ML and foundation-model applications. Include input data, expected output, training or customization approach, explainability, latency, cost, regulatory sensitivity, and failure modes. Populate it with examples such as classification, forecasting, summarization, retrieval, and an agent that uses tools. The goal is to justify a choice, not to declare one technology universally superior.
Next, study the GenAI lifecycle: data selection, model selection, pre-training, fine-tuning, evaluation, deployment, and feedback. Add the application layer around the model: prompt, context, retrieval, tools, guardrails, monitoring, and human review. This gives you a way to reason through questions that combine model behavior with application architecture.
How should you study if you already know AWS?
Do not assume cloud familiarity covers the exam. An AWS practitioner may know IAM, S3, Lambda, and pricing but still need focused work on tokens, embeddings, vector search, prompt techniques, RAG, fine-tuning, evaluation metrics, hallucinations, and responsible AI. Start with a diagnostic against the domain objectives rather than rereading general AWS material.
Use service scenarios to expose gaps. For example, explain why an application might use Amazon Bedrock, where a knowledge base fits, how access is controlled, how prompts and outputs are evaluated, and which business metric would show value. Then repeat the exercise with a traditional ML use case involving Amazon SageMaker AI or a managed AI service.
How should nontechnical candidates approach the blueprint?
Business and product experience is useful because many objectives ask you to align technology with value, cost, risk, and user outcomes. Build enough technical vocabulary to distinguish training from inference, embeddings from source documents, and model evaluation from business evaluation. Then practice explaining a solution in plain language while preserving the relevant technical tradeoff.
For every use case, write the objective, acceptable error, data sensitivity, required response time, human-review need, expected metric, and cost concern. This turns abstract terminology into a repeatable selection process and exposes when AI is not the right tool.
What is a practical study roadmap?
A workable roadmap has four passes: establish the vocabulary, work through the weighted domains, apply the concepts to AWS scenarios, and verify readiness with retrieval practice. Keep a miss log throughout. Each entry should record the objective, the misleading option, the correct reasoning, and the AWS service or risk involved.
Adjust the number of study sessions to your available time, but keep the order of the passes. Scheduling too early encourages memorization of isolated answers; scheduling too late can lead to endless passive reading. Set a decision point after the diagnostic and again after the final review.
Pass 1: Build the conceptual foundation
Start with Content Domain 1: Fundamentals of AI and ML and Content Domain 2: Fundamentals of GenAI. Define each term in your own words, then attach one use case and one limitation. Practice identifying whether a scenario calls for regression, classification, clustering, traditional ML, a foundation model, retrieval, or no AI/ML solution.
Create short explanations for inference types, learning types, data types, model lifecycle stages, tokens, embeddings, vectors, chunking, prompts, context, agents, and token-based pricing. If you cannot explain a term without repeating a product description, mark it for another pass.
Pass 2: Prioritize foundation-model applications
Study Content Domain 3: Applications of Foundation Models next because it represents 28% of scored content. Work through model selection, inference parameters, RAG, vector databases, prompt techniques, customization, fine-tuning data, evaluation methods, and business alignment metrics.
Use a single hypothetical application to compare alternatives. Ask whether the application needs current private information, a fixed response format, multilingual output, low latency, customization, explainability, or strict data controls. Then decide whether prompting, RAG, fine-tuning, a different model, human review, or a non-GenAI approach best addresses the requirement. Record why the other options are weaker.
Pass 3: Add responsibility and security to every design
Study Content Domain 4: Guidelines for Responsible AI and Content Domain 5: Security, Compliance, and Governance for AI Solutions together, while keeping their domain labels distinct. For every application scenario, ask who may be harmed by bias, what data must be protected, how outputs are validated, how users receive transparency, and what activity must be logged.
Build a control map containing identity and access, encryption, privacy, lineage, guardrails, threat detection, output filtering, audit trails, governance review, and retention. Then distinguish a control’s purpose from the AWS service that can support it. This prevents service-name recall from replacing security reasoning.
Pass 4: Test decisions rather than recognition
Use practice questions only after learning the objectives, and treat them as reasoning exercises rather than a source of memorized answers. For every wrong answer, identify whether the problem was terminology, service selection, business alignment, cost, security, responsible AI, or failure to read the constraint in the scenario.
Include multiple-response, ordering, and matching practice because these formats require a different checking habit from standard multiple choice. In an ordering item, identify the lifecycle or dependency before arranging steps. In a matching item, eliminate options by definition and verify every pair. Never rely on exam dumps, leaked questions, or memorized answer sets; they do not establish understanding and do not guarantee a passing result.
What should a final-week review look like?
In the final review, stop expanding the syllabus and repair recurring gaps. Revisit your miss log, domain checklist, service comparisons, metrics, and security controls. Then complete a timed practice session using legitimate preparation material and review the reasoning behind every uncertain response, including answers you happened to guess correctly.
Use the official weighting to decide where to spend the remaining time, but do not abandon the smaller domains. Content Domain 3: Applications of Foundation Models represents 28% of scored content, Content Domain 2: Fundamentals of GenAI represents 24% of scored content, and Content Domain 1: Fundamentals of AI and ML represents 20% of scored content; Content Domain 4: Guidelines for Responsible AI represents 14% of scored content and Content Domain 5: Security, Compliance, and Governance for AI Solutions represents 14% of scored content.
A final review should be able to answer practical questions such as these: When is a specific outcome better served without prediction? Which model or modality fits the constraints? When does RAG make sense? What does a prompt technique change? Which metric reflects the business objective? What could cause a hallucination, leakage, bias, or prompt injection? Which AWS service or control supports the requirement? If your answer is only a product name, add the reason and tradeoff.
When should you schedule the exam?
Schedule when you can explain the blueprint objectives, not merely recognize their terms. A reasonable readiness check is to review each domain without notes, explain the rationale for common AWS service choices, finish practice items without leaving questions unanswered, and show that your weaker areas are improving across repeated attempts.
Confirm the current exam language, delivery option, appointment availability, identification rules, and provider requirements on the official AWS certification page before paying or booking. If your diagnostic still shows broad uncertainty in foundation-model applications or basic GenAI, use that evidence to delay scheduling rather than hoping unfamiliar questions will be favorable.
What should you do during the exam?
Read the business requirement and constraints before examining the answer choices. Identify the requested outcome, data sensitivity, latency, cost, explainability, customization, and operational responsibility. Eliminate options that solve a different problem or introduce an unnecessary implementation burden.
For multiple-response questions, select only options you can support from the scenario and verify that you have not omitted a required response. For ordering and matching questions, check the complete sequence or every pair before moving on. Answer all questions because unanswered questions are scored as incorrect and AWS states there is no penalty for guessing.
Do not let a familiar AWS service decide the answer by itself. A question may be testing model choice, prompt design, evaluation, responsible AI, or governance rather than the service most prominently mentioned. Keep the explanation tied to the stated requirement and use the remaining time to revisit marked questions.
What common preparation mistakes reduce readiness?
The most damaging mistake is studying AIF-C01 as either a pure AI theory exam or a miniature solutions-architect exam. The blueprint combines concepts, AWS services, business selection, responsible AI, and governance. Preparation must therefore move between definitions and scenarios without drifting into unsupported implementation depth.
Four habits are especially costly. First, memorizing service names without use cases leaves you unable to distinguish similar choices. Second, treating GenAI output as automatically correct hides hallucination, evaluation, and grounding requirements. Third, ignoring cost and business metrics misses the decision context. Fourth, avoiding the 14% domains creates preventable risk in responsible AI and security.
Mistake: overstudying coding and model mathematics
Coding, hyperparameter tuning, model optimization, mathematical analysis, and pipeline implementation are not the target candidate’s expected tasks. Review their purpose and vocabulary when they appear in lifecycle discussions, but redirect study time toward identifying suitable technologies, interpreting limitations, explaining tradeoffs, and selecting responsible controls.
A useful test is whether your notes help you choose or explain an approach. If they instead describe how to implement an algorithm line by line, they are probably too deep for the stated scope.
Mistake: treating the blueprint percentages as a pass formula
The domain percentages describe the distribution of scored content, not a promise that mastering one domain compensates for neglecting another. Content Domain 4: Guidelines for Responsible AI and Content Domain 5: Security, Compliance, and Governance for AI Solutions each represent 14% of scored content, and both can shape the correct answer in a question that also mentions a model or AWS service.
Use weights to allocate review time, not to calculate a personal passing threshold. AWS cautions that section-level feedback should be interpreted carefully, so diagnose the underlying objective whenever you miss a question.
Mistake: confusing technical evaluation with business success
A model can score well on a technical metric and still fail the business objective through excessive latency, cost, poor user experience, or unsafe outputs. Conversely, a lower technical score may be acceptable when the application has human review or a low-risk use case. Practice pairing metrics such as accuracy, precision, recall, F1 score, ROUGE, BLEU, or BERTScore with an outcome such as task completion, user satisfaction, productivity, efficiency, ROI, or cost per interaction.
What should you do next?
Download or open the current AWS AIF-C01 exam guide and turn every task statement into a checklist. Mark each objective as explain, distinguish, select, evaluate, or recognize; that verb tells you how to study it. Then compare the checklist with your experience and schedule a diagnostic session before choosing a preparation timetable.
Next, create five working pages: an AI/ML vocabulary map, a GenAI and foundation-model comparison, an AWS service-to-use-case table, a metrics and evaluation sheet, and a responsible-AI/security control map. Fill them from the official domain pages and update them when you verify a current service detail.
Finally, choose a review date and a scheduling decision point. If you can reason through the objectives and explain why an option fits the constraints, move toward booking. If you are still relying on recognition, isolated flashcards, or remembered answer patterns, return to scenario practice and the miss log. The certification is designed to validate practical foundational judgment; prepare for that judgment directly.
Official material to keep open
Use the main AWS exam guide for the target candidate, scope, question types, scoring information, and domain weighting. Use the domain pages for objectives and the in-scope-services page for the current service boundary. Check the AWS certification page for delivery, language, and scheduling information immediately before making an appointment.
These sources are the appropriate reference point for changes. The in-scope list is explicitly non-exhaustive and subject to change, so do not treat a copied service list or an older course outline as permanent.
Conclusion
AIF-C01 preparation is strongest when it mirrors the decisions the exam asks you to make: identify the problem, choose an appropriate AI or GenAI approach, understand the model and application tradeoffs, measure business value, and account for responsible use, security, compliance, and governance. Start with the official blueprint, prioritize Content Domain 3: Applications of Foundation Models and the other larger domains, keep the two 14% domains in your plan, and schedule only after you can explain the reasoning behind your choices.