500 scenario-based practice questions covering foundation model integration, RAG and agents, AI safety and governance, cost and performance optimization, and testing and troubleshooting. Mapped to the five domains of the official AIP-C01 exam guide.
No credit card required · Upgrade to 500 questions for $11.99
500 Exam-Style Questions
7-Day Money-Back Guarantee
Is this for you?
Your AIP-C01 exam is in the next 2–4 weeks
You've studied the concepts but aren't sure you'll pass
You want questions harder than the real exam
You want to understand why answers are correct, not just memorize
Not for you if:
You're looking for braindumps or exam leaks
You haven't started studying the concepts yet
The AWS Certified Generative AI Developer – Professional (AIP-C01) is a professional-level AWS certification. The exam has 75 questions, runs 180 minutes, costs $300, and requires a passing score of 750/1000. CertSharp provides 500 exam-style AIP-C01 practice questions — 30 free with no credit card — each with a detailed explanation.
Foundation Model Integration, Data Management, and Compliance (31%)
2
Implementation and Integration (26%)
3
AI Safety, Security, and Governance (20%)
4
Operational Efficiency and Optimization for GenAI Applications (12%)
5
Testing, Validation, and Troubleshooting (11%)
Why CertSharp
Built for Exam Day, Not Just Study Time
Study guides teach concepts. We prepare you for the actual AIP-C01 exam.
1
Production GenAI scenarios on Amazon Bedrock: Knowledge Bases, Agents, Guardrails, prompt routing, model evaluation, and quotas
2
Every answer checked against current AWS documentation, with explanations that say why each distractor fails
3
Weighted to match the exam, led by Foundation Model Integration, Data Management, and Compliance at 31%
4
Multiple-response questions included, matching the format of the real exam
Free Sample Questions
AIP-C01 Practice Questions (Free Sample)
Real AIP-C01 exam-style questions from our 500-question AWS Certified Generative AI Developer – Professional bank, each with the correct answer and a detailed explanation.
Question 1 — AIP-C01
A team must plan Bedrock capacity for a new chat feature expecting 40 requests per second at peak. Which method produces the most useful estimate?
A
Count the expected daily active users and multiply by a fixed constant per user session
B
Estimate requests per minute alone and compare that figure to the request quota limit
C
Multiply peak requests per second by average tokens per request, compare with token and request quotas, add headroom
D
Estimate the document storage size and scale the number of model units to that figure
Correct answer: C — Explanation
Foundation model limits are expressed in tokens per minute and requests per minute, so capacity planning must convert peak traffic and average prompt and completion size into token throughput, then add headroom. User counts and storage size do not map to throughput. Request rate alone misses cases where long prompts exhaust the token quota first.
Question 2AI Safety, Security, and GovernanceMedium
A knowledge base retrieval returns low relevance scores for an off-topic question, yet the FM still produces a confident answer. How should the application reduce this hallucination risk?
A
Apply a relevance score threshold to retrieved results and return a cannot-answer message when none pass it
B
Routinely pass the top result to the model so that it has at least some context for the answer
C
Ask the user to rephrase the question three times before responding
D
Switch the embeddings to a higher dimension so each query scores above the threshold
Show answer & explanation
Correct answer: A — Apply a relevance score threshold to retrieved results and return a cannot-answer message when none pass it
Thresholding retrieval scores lets the app recognize missing evidence and decline, rather than letting the model improvise. Passing weak context invites fabrication, forcing rephrasing frustrates users without addressing the cause, and higher dimensions do not make irrelevant documents relevant.
Question 3Implementation and IntegrationEasy
What does the Model Context Protocol (MCP) primarily define?
A
A training format for fine-tuning foundation models with labeled examples
B
A network encryption protocol that replaces TLS between AWS services
C
A billing model for tokens consumed by agents across multiple accounts
D
A standard way for AI applications to discover and call tools and data sources exposed by servers
Show answer & explanation
Correct answer: D — A standard way for AI applications to discover and call tools and data sources exposed by servers
MCP is an open protocol connecting AI applications (clients) to tools, resources and prompts exposed by servers. It is not a training data format, a transport encryption standard, or a billing construct.
Question 4Testing, Validation, and TroubleshootingEasy
A team needs a managed way to route low-confidence model outputs to human reviewers through configurable review workflows. Which service provides this?
A
Amazon Lex, which hands the conversation to a human agent through Amazon Connect on request
B
Amazon Kendra, which collects search ratings from users and then retrains its ranking models
C
AWS Step Functions Express workflows, which store the review decisions in the execution history
D
Amazon Augmented AI (Amazon A2I), which defines human review workflows with built-in or custom task UIs
Show answer & explanation
Correct answer: D — Amazon Augmented AI (Amazon A2I), which defines human review workflows with built-in or custom task UIs
Amazon A2I builds human review loops with configurable triggers and task interfaces. Kendra tuning, Lex handoff to a contact center agent, and Express workflow history do not provide a managed reviewer workflow for model outputs.
Question 5Foundation Model Integration, Data Management, and ComplianceMedium
A multi-turn chat application uses the Converse API. How should it preserve conversation context across turns?
A
Send the prior turns in the messages list with alternating user and assistant roles, and put instructions in the system field
B
Merge the history into one large system prompt and send an empty list of messages on each call
C
Send just the newest user message, since Bedrock stores each session history on the server
D
Assign roles randomly to the history entries, because the model ignores role labels in practice
Show answer & explanation
Correct answer: A — Send the prior turns in the messages list with alternating user and assistant roles, and put instructions in the system field
Bedrock inference calls are stateless, so the application must resend the history in order with correct roles, keeping instructions in the system field. A merged system blob loses turn structure, the service does not store history, and role labels matter.
Question 6Operational Efficiency and Optimization for GenAI ApplicationsMedium
A team wants to record actual token consumption for every Bedrock call so finance can attribute cost per feature. The application uses the Converse API. Which source gives the exact token counts for each response?
A
The stopReason field, which reports how many tokens the model generated
B
The metrics.latencyMs field, which converts elapsed time into token estimates
C
The usage object with inputTokens, outputTokens and totalTokens values
D
The additionalModelResponseFields map, which lists billable token totals
Show answer & explanation
Correct answer: C — The usage object with inputTokens, outputTokens and totalTokens values
The Converse response contains a usage object with inputTokens, outputTokens and totalTokens, which can be logged as custom metrics per feature. The stopReason only explains why generation ended, such as end_turn or max_tokens. Latency is a time measure and does not map to billed tokens. additionalModelResponseFields carries model-specific extras, not a standard billing summary.
Question 7AI Safety, Security, and GovernanceHard
A user exercises a right-to-erasure request. Their documents were chunked into a vector store and their chats are stored in S3. Which design makes deletion practical?
A
Delete the S3 source object and assume the vector store updates immediately and automatically
B
Store a user identifier as metadata on chunks and S3 objects, then delete the matching objects, re-sync the knowledge base, and purge the vectors
C
Ask the model to forget the user in the next conversation by sending an instruction
D
Rotate the KMS key so the old data becomes unreadable for the full set of users
Show answer & explanation
Correct answer: B — Store a user identifier as metadata on chunks and S3 objects, then delete the matching objects, re-sync the knowledge base, and purge the vectors
Tagging data with a user identifier lets you locate and delete every copy, including derived embeddings, once sync removes them. Source deletion alone does not remove indexed vectors until a sync, models do not hold per-user memory that an instruction can erase, and rotating a key affects everyone and does not delete data.
Question 8Implementation and IntegrationMedium
A team routes classification tasks to a small model and reasoning tasks to a larger one, with fixed rules. Which approach is the simplest?
A
Application code maps each task type to a model ID kept in configuration
B
Randomly choose a model for each request to spread the load evenly
C
Call the largest model for each request and truncate the output afterward
D
Ask the user to type the model ID into a free text box
Show answer & explanation
Correct answer: A — Application code maps each task type to a model ID kept in configuration
A configuration-driven static mapping gives predictable cost and behavior for known task types. Random choice ignores task needs, using the largest model for each request wastes money, and user-entered model IDs are error-prone and unsafe.
Question 9Testing, Validation, and TroubleshootingMedium
Before cutting over to a new model version, a team must confirm hallucination rates have not worsened. What should the validation do?
A
Run a single sample prompt in the console and approve the cutover when the answer looks reasonable
B
Run the golden dataset with source passages through the new version, score faithfulness with a judge, and block on a worse rate
C
Check that the new version has a larger context window and approve it on that basis alone
D
Compare the average token counts of the two versions and block the cutover when the counts differ
Show answer & explanation
Correct answer: B — Run the golden dataset with source passages through the new version, score faithfulness with a judge, and block on a worse rate
Hallucination validation scores faithfulness on a fixed golden set and compares the rate with the baseline, gating the release on the result. Token counts, a single console prompt, or a larger context window do not measure fabricated content.
Want more AIP-C01 practice questions?
Take the free AIP-C01 practice test: 30 questions per session with instant explanations. Unlock all 500 questions plus timed exam mode for $11.99 one-time.
Everything you need to know about AWS Certified Generative AI Developer – Professional
What is the AIP-C01 exam?
AIP-C01 is the AWS Certified Generative AI Developer – Professional exam. It validates integrating foundation models into applications and running GenAI solutions in production on AWS. The exam costs $300, has 75 questions in 180 minutes, and requires a scaled score of 750 out of 1000 to pass.
How many questions are on AIP-C01?
The AIP-C01 exam has 75 questions: 65 scored and 10 unscored. Questions are multiple choice or multiple response, and multiple response questions need every correct option selected. The time limit is 180 minutes.
What is the AIP-C01 passing score?
The AIP-C01 passing score is 750 on a scale of 100–1000. Scoring is compensatory, so you need to pass the exam overall rather than every domain.
What are the AIP-C01 exam domains?
AIP-C01 has five domains: Foundation Model Integration, Data Management, and Compliance (31%); Implementation and Integration (26%); AI Safety, Security, and Governance (20%); Operational Efficiency and Optimization for GenAI Applications (12%); and Testing, Validation, and Troubleshooting (11%).
How is AIP-C01 different from AIF-C01?
AIF-C01 (AI Practitioner) is a Foundational exam about AI concepts and which AWS AI service fits a use case. AIP-C01 is a Professional exam for developers who build GenAI applications: RAG pipelines, agents, guardrails, evaluation, and cost and performance tuning on Amazon Bedrock.
Which AWS services are on AIP-C01?
Amazon Bedrock is central (models, Knowledge Bases, Agents, Guardrails, prompt management, model evaluation), together with vector stores such as OpenSearch Service, and Lambda, Step Functions, API Gateway, IAM, KMS, and CloudWatch for integration, security, and monitoring.
What experience does AWS recommend for AIP-C01?
AWS recommends 2 or more years of experience building production-grade applications on AWS or with open-source technologies, general AI/ML or data engineering experience, and 1 year of hands-on experience implementing GenAI solutions.
How long does it take to prepare for AIP-C01?
Plan for 8–12 weeks at 1–2 hours per day if you already build software. Build a RAG application and an agent on Bedrock, add guardrails and evaluation, then use timed practice exams to cover the governance and optimization domains.
How much does the AIP-C01 exam cost?
The AWS Certified Generative AI Developer – Professional (AIP-C01) exam costs $300 USD plus applicable taxes, at a Pearson VUE test center or online with a remote proctor.
Is there a free AIP-C01 practice test?
Yes. CertSharp offers a free AIP-C01 practice test for the AWS Generative AI Developer – Professional exam: 30 questions per session with detailed explanations, no credit card required. The full 500-question AIP-C01 bank, including timed exam mode, is $11.99 one-time with lifetime access, or included in Pro.
Also on CertSharp
Related AWS Certification Practice Tests
Same 500-question depth, with 30 free questions per session.