Getting started with the AWS Certified AI Practitioner (AIF-C01) is easier than most people expect, as long as you study the right things. It's a foundational exam, but it isn't a vocabulary quiz. AWS wants to know that you can pick between fine-tuning and Retrieval Augmented Generation, explain what temperature does, and spot the guardrail that fits a compliance requirement. This guide walks through what's on the exam, the services you need to know, the traps that catch people out, and a four-week plan to get you ready.
Exam Overview
| Detail | Value |
|---|---|
| Exam code | AIF-C01 |
| Level | Foundational |
| Questions | 65 (50 scored, 15 unscored) |
| Time | 90 minutes |
| Passing score | 700 out of 1000 |
| Question types | Multiple choice, multiple response, ordering, matching, case study |
| Prerequisites | None (AWS suggests up to 6 months of AI/ML exposure on AWS) |
The audience is broad. Business analysts, product managers, sales engineers, and IT support staff all take this one. If you already hold the Cloud Practitioner, you'll recognize the style, but the content goes deeper into AI concepts than anything on CLF-C02.
Exam Domains
| Domain | Weight |
|---|---|
| Applications of Foundation Models | 28% |
| Fundamentals of Generative AI | 24% |
| Fundamentals of AI and ML | 20% |
| Guidelines for Responsible AI | 14% |
| Security, Compliance, and Governance for AI Solutions | 14% |
More than half the exam is about generative AI and foundation models. If you're short on time, that's where your hours should go.
Core Services and Concepts to Master
Amazon Bedrock
Bedrock is the center of gravity for this exam. Know that it gives you API access to foundation models from Amazon and third parties (Anthropic, Meta, Mistral, Cohere, and others) without managing infrastructure. Understand Knowledge Bases for Bedrock (managed RAG), Agents for Bedrock (multi-step task orchestration that calls APIs), Guardrails for Bedrock (content filters, denied topics, PII redaction), and model evaluation. Also know the pricing models: on-demand per token versus Provisioned Throughput for steady, high-volume workloads or custom models.
Prompt Engineering and Inference Parameters
Expect questions on zero-shot, few-shot, and chain-of-thought prompting. You should be able to explain what temperature, top-p, top-k, and maximum tokens do. Lower temperature means more deterministic output. Higher temperature means more creative, varied output. Prompt injection and jailbreaking also show up, usually paired with a question about how to mitigate them.
RAG vs Fine-Tuning vs Pretraining
This is one of the most tested decisions on the exam. RAG pulls fresh or private data into the prompt at query time and is the cheapest way to ground a model in your own documents. Fine-tuning changes model weights using labeled examples and suits a specific style or task. Continued pretraining uses large amounts of unlabeled domain data. Know that vector databases (OpenSearch Serverless, Aurora PostgreSQL with pgvector, and others) store embeddings for RAG.
Amazon SageMaker
You won't need deep ML engineering skills, but you should know the main SageMaker pieces: SageMaker Studio, JumpStart (prebuilt models and solutions), Canvas (no-code ML), Clarify (bias detection and explainability), Model Monitor (drift detection), Data Wrangler, Ground Truth (data labeling), and Model Cards (documentation for governance).
AI/ML Fundamentals
Brush up on supervised, unsupervised, and reinforcement learning. Know classification versus regression, and when you'd use each. Learn evaluation metrics: accuracy, precision, recall, F1, and AUC for classification, and RMSE for regression. For generative AI, know ROUGE, BLEU, and BERTScore. Understand overfitting and underfitting, and the stages of the ML lifecycle from data collection to monitoring.
Prebuilt AI Services
Map each managed service to its use case. Rekognition handles images and video. Comprehend does NLP and sentiment. Textract extracts text and forms from documents. Transcribe turns speech into text, Polly does the reverse, and Translate handles languages. Lex builds chatbots. Personalize powers recommendations. Amazon Q comes in two main flavors: Q Business for enterprise knowledge and Q Developer for coding help.
Responsible AI
Know the core dimensions: fairness, explainability, privacy, safety, transparency, robustness, and governance. Learn how bias creeps in through unrepresentative training data and how SageMaker Clarify detects it. Hallucinations, toxicity, and intellectual property risks all appear in scenario questions.
Security and Governance
The shared responsibility model applies to AI too. Know IAM for access control, KMS for encryption, PrivateLink and VPC endpoints for keeping traffic private, Macie for discovering sensitive data, CloudTrail for auditing API calls, and AWS Config for compliance tracking. AWS Artifact provides compliance reports. Data lineage and data residency show up in governance questions.
Common Exam Traps
Watch for questions where RAG is clearly the right answer but fine-tuning is offered as a tempting distractor. If the requirement mentions frequently changing data or citing sources, think RAG.
Don't confuse Comprehend and Textract. Comprehend analyzes the meaning of text. Textract pulls text out of scanned documents and forms.
Temperature questions are easy points if you remember that lower means more predictable. People overthink these.
Guardrails for Bedrock and SageMaker Clarify both relate to responsible AI, but they solve different problems. Guardrails filter model inputs and outputs at runtime. Clarify detects bias and explains predictions.
Cost questions often hinge on choosing between on-demand and Provisioned Throughput, or between a smaller and larger model. The cheapest option that meets the requirement usually wins.
Finally, read for qualifiers like "least operational overhead." That nearly always points to a fully managed service like Bedrock or a prebuilt AI service rather than training your own model.
Study Plan
| Week | Focus | Activities |
|---|---|---|
| 1 | AI/ML fundamentals | Learn ML types, metrics, the ML lifecycle, and the prebuilt AI services |
| 2 | Generative AI and foundation models | Study tokens, embeddings, prompt engineering, inference parameters, Bedrock features |
| 3 | Applications, RAG, and customization | Compare RAG, fine-tuning, and pretraining; learn Knowledge Bases, Agents, and evaluation |
| 4 | Responsible AI, security, and review | Cover Clarify, Guardrails, IAM, KMS, governance; take full timed practice exams |
If you already work with AWS daily, you can likely compress this into two or three weeks. If you're new to both cloud and AI, give yourself five or six.
Recommended Resources
Start with the official AWS exam guide for AIF-C01. It lists every in-scope service and task statement, so use it as your checklist. AWS Skill Builder offers free foundational courses on generative AI and the AI Practitioner learning plan. The Bedrock and SageMaker documentation pages are worth skimming for feature names, since the exam often tests whether you know which feature does what.
Hands-on time helps more than you'd think. Spend an hour in the Bedrock console trying different models and adjusting temperature. It makes the inference parameter questions feel obvious.
Practice questions are where it all comes together. Working through scenario-based questions shows you exactly where your gaps are and gets you used to how AWS phrases things.
Final Thoughts
The AIF-C01 rewards clear understanding over memorization. If you can explain when to use RAG, what each Bedrock feature does, and how AWS handles responsible AI and security, you're most of the way there. Spend the majority of your time on the generative AI and foundation model domains, then lock in the rest with practice.
When you're ready to test yourself, try our AWS AIF-C01 Practice Exams. You'll get exam-style questions across all five domains with detailed explanations for every answer.