AIF-C01 study guide: AWS Certified AI Practitioner domains and a 4-week plan

AIF-C01, the AWS Certified AI Practitioner exam, tests whether you understand AI, machine learning and generative AI well enough to pick the right AWS service for a business problem. It is a foundational exam aimed at people who use AI solutions rather than build them, so there is no coding and no maths. What it does demand is breadth: many concepts and services, each tested at the level of "what is it for, and when would you choose it?"

AIF-C01 at a glance

Details as of September 2026 — confirm on the official exam page before booking.

AWS has announced that the Italian and German versions of the exam retire after 15 October 2026. The certification itself is valid for three years.

Who this exam is for

AWS describes the target candidate as someone with up to six months of exposure to AI and ML technologies on AWS who uses, but does not necessarily build, AI and ML solutions. That covers cloud, development, data, IT and line-of-business roles alike.

The recommended background is familiarity with core AWS services (Amazon EC2, Amazon S3, AWS Lambda, Amazon Bedrock and Amazon SageMaker AI), the shared responsibility model, IAM and AWS pricing models. There is no formal prerequisite. If you are new to IT and AWS, AWS suggests an introductory course first, and the ground covered in our [CLF-C02 study guide](/blog/clf-c02-study-guide) makes the AWS side of AIF-C01 easier.

The exam guide is just as useful for what it leaves out: coding models, feature engineering, hyperparameter tuning, building pipelines, statistical analysis, and implementing security or governance frameworks are all out of scope. If you are memorising formulas or API calls, you have gone too deep.

What the exam covers

There are five domains. Domains 2 and 3 together make up just over half of the scored content, and both are about generative AI.

Domain 1: Fundamentals of AI and ML (20%)

Expect to distinguish AI, ML and deep learning; supervised, unsupervised and reinforcement learning; batch and real-time inference; and labelled versus unlabelled, structured versus unstructured data. The judgement being tested is matching a described problem to a technique (regression, classification or clustering) and recognising when ML is the wrong tool, for example when the business needs a guaranteed, rule-based outcome rather than a prediction. Know the ML pipeline stages in order, the basic ideas of MLOps, and the difference between model metrics (accuracy, F1 score) and business metrics (cost per user, return on investment). Learn the managed AI services by what they take in and what they give back: Amazon Transcribe, Translate, Comprehend, Lex and Polly.

Domain 2: Fundamentals of Generative AI (24%)

This domain covers the building blocks of generative AI: tokens, chunking, embeddings and vectors, transformer-based large language models, foundation models, multimodal and diffusion models, and the foundation model lifecycle. It also asks you to weigh generative AI honestly: its strengths (adaptability, responsiveness) and weaknesses (hallucination, nondeterminism, limited interpretability). On the AWS side, know what Amazon Bedrock, Amazon SageMaker JumpStart and Amazon Q are each for, and understand the cost trade-offs: token-based pricing, provisioned throughput, custom models.

Domain 3: Applications of Foundation Models (28%)

The heaviest domain covers four areas:

• Design choices. Selecting a pre-trained model by cost, modality, latency, language support, size and input/output length; how inference parameters such as temperature change responses; what retrieval augmented generation (RAG) is and where Amazon Bedrock knowledge bases fit; which AWS databases can store embeddings (Amazon OpenSearch Service, Amazon Aurora, Amazon Neptune, Amazon DocumentDB, Amazon RDS for PostgreSQL); and the role of agents in multi-step tasks.
• Prompt engineering. Context, instructions and negative prompts; zero-shot, single-shot, few-shot and chain-of-thought techniques; prompt templates; and the risks, including exposure, poisoning, hijacking and jailbreaking.
• Customisation. The difference between pre-training, continued pre-training, fine-tuning, in-context learning and RAG, what each costs, and how to prepare data for fine-tuning.
• Evaluation. Human evaluation, benchmark datasets, metrics such as ROUGE, BLEU and BERTScore, and whether a model actually meets its business objective.

Domain 4: Guidelines for Responsible AI (14%)

Expect questions on the features of responsible AI (bias, fairness, inclusivity, robustness, safety, veracity), the legal risks of generative AI such as intellectual property claims and loss of customer trust, and how bias and variance show up as overfitting, underfitting or unequal results across demographic groups. You need to know which tools help: Amazon Bedrock Guardrails for filtering content, Amazon SageMaker Clarify and SageMaker Model Monitor for bias and drift, Amazon Augmented AI (Amazon A2I) for human review, and SageMaker Model Cards for documenting a model. Also know how transparent, explainable models differ from opaque ones, and the trade-off between interpretability and performance.

Domain 5: Security, Compliance, and Governance for AI Solutions (14%)

This domain maps security and governance needs to AWS services. For protection: IAM roles and policies, encryption at rest and in transit, Amazon Macie, AWS PrivateLink and the shared responsibility model. For compliance: AWS Config, Amazon Inspector, AWS Audit Manager, AWS Artifact, AWS CloudTrail and AWS Trusted Advisor. It also covers data lineage and cataloguing, secure data engineering, prompt injection, data governance (lifecycle, residency, retention, logging) and governance processes such as the Generative AI Security Scoping Matrix. Learn each service by its one-line purpose.

A 4-week study plan

The plan follows the weights: roughly half your time goes on Domains 2 and 3.

Week 1: Domain 1 and the AWS basics. Build a glossary of the terms in the exam guide and write a one-line definition for each in your own words. Make a table of the managed AI services with their input and output. End with a Domain 1 paper and read every explanation.

Week 2: Domain 2. Study how foundation models work, then what each AWS generative AI service is for. Build a second table of pricing options (on-demand tokens, provisioned throughput, custom models) and when each makes sense. Finish with a Domain 2 paper.

Week 3: Domain 3. Give this week the most time. Work through model selection, inference parameters, RAG and vector stores, the customisation options in cost order, prompt techniques and their risks, and evaluation metrics. Take a paper on each area as you finish it, then a mixed Domain 3 paper.

Week 4: Domains 4 and 5, then full mocks. Spend the first half on responsible AI and the security and governance services. Then sit a full-length mock under exam conditions, review it thoroughly, fix the weakest topics, and sit a second mock two or three days later. Our guide on [how to use practice exams effectively](/blog/how-to-use-practice-exams-effectively) explains how to review a mock so it actually moves your score.

Common traps

• Overthinking a foundational exam. Building and tuning models is out of scope. When two answers look plausible, the managed service that meets the need with the least effort is usually what the question is after.
• Mixing up the customisation options. Prompt engineering, RAG, fine-tuning and pre-training differ in cost, effort and what they change. A question about answering from current company documents points to RAG; one about changing the model's tone or domain behaviour points to fine-tuning.
• Confusing similar AI services. Comprehend analyses text, Transcribe turns speech into text, Polly turns text into speech, Translate translates, Lex builds conversational interfaces. Learn them by input and output.
• Mismatching governance services. Config records resource configuration, CloudTrail records API activity, Artifact provides AWS compliance reports, Inspector scans for vulnerabilities, Macie finds sensitive data in S3. Many distractors are real services doing the wrong job.
• Partial answers. On multiple response, ordering and matching questions you must get every selection, position or pair right to receive credit. Check how many answers are required before moving on.
• Chasing product names. AWS renames and adds generative AI services often. If you understand the capability a question describes (bias detection, drift monitoring, human review, content filtering), you can pick the right answer even when the wording is unfamiliar.

How to practise

CertifyCloudx has original AIF-C01 practice questions written against the public exam guide, with an explanation for every option. You can take domain papers of up to 25 questions (60 minutes per 25) to work through the plan above, and full-length timed mock exams with the real exam's 90-minute limit for week 4. The free plan gives you up to 10 questions a day, with no card required.

Start with the [AIF-C01 practice questions](/certifications/aws-ai-practitioner-aif-c01). If AIF-C01 is your first step into AWS and you want to go deeper into architecture afterwards, the [SAA-C03 study guide](/blog/saa-c03-study-guide) covers the associate-level Solutions Architect exam.

Frequently asked questions

How difficult is AIF-C01?

It is a foundational exam that assumes up to six months of exposure to AI on AWS and no coding. The challenge is breadth rather than depth: many terms and services, each tested at the level of what it does and when you would use it.

How long should I prepare for AIF-C01?

With some cloud background, three to six weeks of steady study is a realistic range, and the 4-week plan above suits most candidates. If AWS and AI terminology are both new to you, allow the longer end.

Do I need another AWS certification first?

No. There is no prerequisite. Familiarity with core AWS services, IAM, pricing and the shared responsibility model is recommended, and AWS suggests an introductory course if you are new to IT and AWS.

How is AIF-C01 scored?

The exam has 65 questions, of which 50 are scored and 15 are unscored and not identified. Results are reported on a scaled score of 100–1,000 with 700 needed to pass, and the scoring is compensatory, so you need to pass the exam overall rather than each domain.

Are CertifyCloudx questions real AIF-C01 exam questions?

No. Every CertifyCloudx question is original and written against the public exam guide. Using leaked exam content breaks the AWS candidate agreement and can cost you the certification, as our article on [why exam dumps put your certification at risk](/blog/why-exam-dumps-put-your-certification-at-risk) explains.

CertifyCloudx is independent and not affiliated with Amazon Web Services. AWS Certified AI Practitioner is a trademark of its owner. All CertifyCloudx practice questions are original.