The AWS Certified AI Practitioner (AIF-C01) exam is organized into five domains, each with a specific weight that determines how many questions you’ll see on that topic. Understanding the domain structure is the single most important step in building a study plan — it tells you exactly where to invest your time.
This guide breaks down all five AIF-C01 domains in detail: what each covers, what AWS actually tests, common pitfalls, and a focused study sequence for each.
AIF-C01 Domain Summary
| Domain | Weight | Approx. Questions |
|---|---|---|
| 1. Fundamentals of AI and ML | 20% | 13 |
| 2. Fundamentals of Generative AI | 24% | 16 |
| 3. Applications of Foundation Models | 28% | 18 |
| 4. Guidelines for Responsible AI | 14% | 9 |
| 5. Security, Compliance, and Governance for AI Solutions | 14% | 9 |
A few patterns to internalize:
- More than half the exam (52%) is foundation-model and generative AI material.
- The single largest domain (Domain 3 at 28%) is also the most technical.
- The smaller domains (4 and 5 at 14% each) are easy points if you study, easy losses if you skip.
Domain 1: Fundamentals of AI and ML (20%)
What This Domain Covers
This domain sets the vocabulary and conceptual baseline for the entire exam. AWS expects you to explain AI/ML at a level that lets you participate in technical conversations and map business problems to appropriate AI/ML solutions.
Key sub-topics:
- AI vs. machine learning vs. deep learning vs. generative AI
- Types of machine learning: supervised, unsupervised, reinforcement
- Common ML problem types: classification, regression, clustering, anomaly detection, recommendation, forecasting
- The ML development lifecycle: business problem framing, data collection, data preparation, feature engineering, model training, model evaluation, model deployment, monitoring
- Algorithm families at a conceptual level (decision trees, linear models, neural networks)
- Practical AWS AI service mappings: Comprehend (NLP), Rekognition (vision), Forecast (time series), Personalize (recommendations), Translate, Transcribe, Polly, Lex, Textract, Kendra
What AWS Actually Tests
Expect scenario questions that ask you to:
- Identify which ML problem type fits a business scenario (e.g., predicting churn = classification)
- Choose the correct ML lifecycle stage for an activity (e.g., one-hot encoding = feature engineering)
- Pick the right AWS AI service for a use case (e.g., sentiment analysis = Amazon Comprehend)
Common Pitfalls
- Confusing classification and regression. Classification predicts categories; regression predicts numeric values.
- Confusing unsupervised learning with reinforcement learning. Unsupervised = no labels; reinforcement = reward signals over time.
- Confusing similar AWS services. Comprehend vs. Kendra, Rekognition vs. Textract, Transcribe vs. Polly.
Focused Study Sequence
- Write your own one-line definitions for every key term.
- Create a table mapping AWS AI services to use cases.
- Drill 30 to 50 Domain 1 practice questions before moving on.
Domain 2: Fundamentals of Generative AI (24%)
What This Domain Covers
The second-largest domain. AWS tests your conceptual understanding of generative AI and the AWS infrastructure that supports it.
Key sub-topics:
- Generative AI vs. traditional ML
- Foundation models (FMs) and large language models (LLMs)
- Multimodal models (text + image + audio)
- Tokens, embeddings, vector representations, context windows
- Vector databases and their role in generative AI
- Capabilities and limitations of generative AI (hallucinations, knowledge cutoffs, deterministic vs. stochastic behavior)
- AWS generative AI services:
- Amazon Bedrock — managed foundation model access from multiple providers
- SageMaker JumpStart — pre-trained models and solutions
- Amazon Q Developer — coding assistant
- Amazon Q Business — enterprise knowledge assistant
- Business use cases for generative AI: content creation, summarization, chatbots, code generation, search, translation
What AWS Actually Tests
- Mapping a business scenario to the right AWS generative AI service
- Identifying which features (knowledge bases, agents, guardrails) belong to which service
- Recognizing the capabilities and limitations of LLMs
Common Pitfalls
- Choosing SageMaker JumpStart when Bedrock is the right answer. Bedrock is serverless API; JumpStart requires endpoint deployment.
- Confusing Q Developer and Q Business. Developer = code. Business = enterprise data.
- Underestimating hallucinations. Some scenario questions test whether you recognize a limitation of LLMs.
Focused Study Sequence
- Build a clean one-page diagram of Amazon Bedrock, its features (model access, knowledge bases, agents, guardrails), and how requests flow.
- Memorize the 3-line description of each AWS generative AI service.
- Drill 40 to 50 Domain 2 practice questions and review every wrong answer carefully.
For a deeper service-specific study, see our Amazon Bedrock guide for AIF-C01.
Domain 3: Applications of Foundation Models (28%)
What This Domain Covers
The largest and most technical domain on the exam. AWS tests whether you can actually apply foundation models — choosing the right customization approach, writing effective prompts, and evaluating performance.
Key sub-topics:
- Design considerations for applications using foundation models
- Prompt engineering techniques:
- Zero-shot prompting
- Few-shot prompting
- Chain-of-thought prompting
- Instruction prompts and system prompts
- Prompt chaining
- Retrieval-Augmented Generation (RAG) and when to use it
- Foundation model customization options:
- In-context learning (prompt engineering)
- RAG (knowledge bases)
- Fine-tuning
- Continued pre-training
- Inference parameters and their effects:
- Temperature (randomness)
- Top-p / top-k (diversity sampling)
- Max tokens (response length)
- Stop sequences
- Evaluation metrics for generative models:
- ROUGE (summarization)
- BLEU (translation)
- BERTScore (semantic similarity)
- Perplexity
- Human evaluation
- Choosing between off-the-shelf, customized, and fine-tuned models
What AWS Actually Tests
This is the domain where scenario questions get sharpest:
- Given a use case, choose the right prompt engineering technique
- Given a use case, choose between RAG, fine-tuning, and prompt engineering
- Pick the right evaluation metric for a generative task
- Choose inference parameters to achieve a desired output style
Common Pitfalls
- Choosing fine-tuning when RAG is the right answer. Rule of thumb: changing facts = RAG; stable style/behavior = fine-tuning.
- Confusing temperature direction. Higher temperature = more random/creative. Lower = more deterministic.
- Picking the wrong evaluation metric. ROUGE for summarization, BLEU for translation, BERTScore for semantic similarity.
- Defaulting to “use prompt engineering” for every scenario. Sometimes the correct answer is fine-tuning, continued pre-training, or RAG.
Focused Study Sequence
- Read AWS’s prompt engineering guidelines twice and create your own examples for each technique.
- Build a comparison table: when to use prompt engineering vs. RAG vs. fine-tuning vs. continued pre-training.
- Memorize evaluation metric → use case mappings.
- Drill 60 to 80 Domain 3 practice questions. This domain demands the most practice.
For a deep prompt engineering guide, see our AWS AI Practitioner prompt engineering guide.
Domain 4: Guidelines for Responsible AI (14%)
What This Domain Covers
A smaller domain by weight, but rich in subtle questions about ethics, fairness, and AWS-specific responsible AI tools.
Key sub-topics:
- Core responsible AI principles:
- Fairness
- Inclusivity
- Transparency
- Explainability
- Safety / robustness
- Privacy
- Veracity (truthfulness)
- Bias detection and mitigation strategies
- Diverse and representative training data
- Human oversight and human-in-the-loop patterns
- Legal and ethical considerations in AI deployment
- AWS responsible AI tools:
- Amazon SageMaker Clarify — bias detection and model explainability for classical ML
- Amazon Bedrock Guardrails — content filtering, denied topics, PII redaction for generative AI
- AWS AI Service Cards / Model Cards — documentation of model capabilities, limitations, intended use
What AWS Actually Tests
- Identifying which AWS tool addresses a specific responsible AI concern
- Recognizing principles of responsible AI in scenarios
- Distinguishing between Clarify (bias / explainability) and Guardrails (generative content filtering)
Common Pitfalls
- Confusing SageMaker Clarify and Bedrock Guardrails. Clarify = classical ML bias. Guardrails = generative content filtering. Both are responsible AI tools but for different problems.
- Treating responsible AI as “philosophy.” It has specific AWS-flavored answers, not just opinions.
- Forgetting that responsible AI applies across the lifecycle, not just deployment.
Focused Study Sequence
- Memorize the responsible AI principles list.
- Memorize the three AWS responsible AI tools and their distinct purposes.
- Drill 20 to 30 Domain 4 practice questions and pay close attention to which tool each scenario calls for.
Domain 5: Security, Compliance, and Governance for AI Solutions (14%)
What This Domain Covers
The smallest domain, focused on securing AI workloads using familiar AWS security primitives.
Key sub-topics:
- Securing AI/ML systems:
- IAM and least privilege for AI workloads
- VPC isolation and PrivateLink for AI services
- Encryption at rest (KMS) and in transit (TLS)
- Data privacy and protection in AI workloads:
- Data classification
- PII handling
- Data residency
- Governance and compliance considerations:
- Industry regulations relevant to AI (GDPR, HIPAA, financial regulations)
- Internal AI policies and acceptable use
- Monitoring and auditing AI systems:
- AWS CloudTrail — API activity logging
- Amazon CloudWatch — metrics and logs
- Bedrock model invocation logging
- SageMaker Model Monitor — drift detection
- Data lineage and model lineage
What AWS Actually Tests
- Selecting the right AWS security service for a specific AI workload concern
- Recognizing how IAM, VPC, KMS, and PrivateLink apply to AI services
- Identifying which AWS service handles a specific compliance or audit requirement
Common Pitfalls
- Forgetting that AI workloads use the same AWS security primitives as any other workload.
- Choosing CloudWatch when CloudTrail is the right answer (or vice versa). CloudTrail = who did what. CloudWatch = metrics and logs.
- Missing the role of PrivateLink for keeping traffic on the AWS backbone.
Focused Study Sequence
- Refresh AWS security basics — IAM, VPC, KMS, PrivateLink — if you don’t already have a Cloud Practitioner background.
- Read AWS’s Bedrock security documentation.
- Drill 20 to 30 Domain 5 practice questions.
How to Allocate Study Time Across Domains
Based on weighting and difficulty, here’s a recommended time allocation for a 30-hour study plan:
| Domain | Recommended Hours | Reasoning |
|---|---|---|
| Domain 1: Fundamentals of AI and ML | 5 hours | Vocabulary-heavy but conceptually accessible |
| Domain 2: Fundamentals of Generative AI | 7 hours | Larger weight + new content for most candidates |
| Domain 3: Applications of Foundation Models | 10 hours | Largest weight, most technical, hardest to master |
| Domain 4: Guidelines for Responsible AI | 4 hours | Small but nuanced |
| Domain 5: Security, Compliance, and Governance | 4 hours | Small; reuses general AWS security knowledge |
Add 4 to 6 hours for full-length mock exams and review.
Domain-by-Domain Practice Strategy
Practice questions should mirror the domain distribution. If you’re working through 200 practice questions, aim for roughly:
- 40 questions on Domain 1
- 48 questions on Domain 2
- 56 questions on Domain 3
- 28 questions on Domain 4
- 28 questions on Domain 5
Our AWS Certified AI Practitioner mock exam bundle is calibrated to exactly these official AIF-C01 domain weightings, so every full-length mock you take has the right distribution. Each mock includes a domain-level performance breakdown so you know exactly which domain to focus on next.
Tracking Domain Performance
Maintain a simple tracker:
| Mock | D1 % | D2 % | D3 % | D4 % | D5 % | Overall |
|---|---|---|---|---|---|---|
| Mock 1 | 80 | 70 | 60 | 75 | 85 | 71 |
| Mock 2 | 85 | 75 | 70 | 80 | 85 | 77 |
| Mock 3 | 90 | 80 | 78 | 85 | 90 | 83 |
When your overall score is consistently above 80 percent and no single domain is below 70 percent, you’re ready.
If a single domain is dragging you down, do focused study on that domain for 3 to 5 days before the next mock.
FAQ: AIF-C01 Exam Domains
Q: Which domain has the most questions? A: Domain 3 (Applications of Foundation Models) at 28%, which translates to roughly 18 of the 65 questions.
Q: Which domain is the hardest? A: Domain 3, for the same reason — it covers the most technical material (prompt engineering, RAG, fine-tuning, evaluation metrics, inference parameters).
Q: Can I skip the smaller domains? A: No. Domains 4 and 5 are 28 percent of the exam combined. Skipping them is a common reason candidates fail.
Q: How are unscored questions distributed across domains? A: AWS doesn’t publish this. Assume they’re roughly proportional. Treat every question as if it counts.
Q: Is the domain weighting the same in every exam version? A: Yes, current AIF-C01 weightings are 20 / 24 / 28 / 14 / 14. Always verify the latest version against the official AWS exam guide.
Q: How do I know which domain I’m weakest in? A: Take full-length mocks with per-domain reporting. Our AIF-C01 mock exam bundle provides exactly this.
Conclusion
The AIF-C01 domain structure is the map for your study plan. The biggest mistake candidates make is treating all five domains equally — they aren’t. Domain 3 alone is 28 percent of the exam and the most technically demanding. Domains 1 and 2 set the foundations. Domains 4 and 5 are smaller but full of easy points if you’ve studied.
Allocate your time according to weight and difficulty, track domain performance through mock exams, and book the real exam only when no single domain is dragging you below 75 percent.
Ready to test your domain-by-domain readiness? Our AWS Certified AI Practitioner mock exam bundle gives you 8 full-length exams aligned to the official AIF-C01 domain weightings, with detailed per-domain reporting so you can target your weakest areas with confidence.