AWS AI Practitioner practice

{{ summaryValue }}

{{ summaryLine }}

Build an AIF-C01 practice round

Unofficial AWS Certified AI Practitioner (AIF-C01) practice · guide 1.1, published April 30, 2026 · verified September 11, 2026. Independently authored; not affiliated with or endorsed by AWS.

40 curated English questions cover selected concepts across all five domains and 14 task statements. This is untimed single-choice practice, sampled uniformly. The real exam has 65 questions in 90 minutes and can include multiple response, ordering, and matching.

Official guide and format · Official preparation resources

Blueprint weights: AI/ML 20%; GenAI 24%; foundation models 28%; responsible AI 14%; security and governance 14%. These are scope references, not this round’s sampling weights. Concept checks and applied scenarios are author-assigned practice levels, not calibrated exam difficulty.

Optional. Blank uses a stable default; a custom seed changes question and option order. An unusable seed falls back to the default.

{{ examSetupValidation }}

{{ answeredCount }} of {{ questionCount }} answered · {{ Math.round(progressPercent) }}%

{{ progressStatusLine }}

Question {{ examCursor + 1 }} of {{ questionCount }}

{{ currentQuestion.stem }}

{{ currentQuestion.accessible_clue }}

Answer recorded. Corrections appear when you view results.

Review this practice round

{{ resultLead }}

AIF-C01 · guide 1.1 · original bank 2026.09.11. This score describes only the items answered. It is not an AWS scaled score, a pass prediction, or evidence of certification. AWS and its service names are trademarks of Amazon.com, Inc. or its affiliates.

{{ readinessReviewTitle }}

{{ readinessReviewLine }}

{{ card.label }}
{{ card.value }}
{{ card.note }}

Missed answer review

  1. {{ item.prompt }}

    Your answer: {{ item.selected_answer_label }}

    Correct answer: {{ item.correct_answer_label }}

    {{ item.explanation }}

    {{ item.source.label.split(' — ')[0] }} · {{ item.source.version }}

Domain performance

Correct and missed answers by official domain for this round. Domains absent from the chart were not sampled. Use the missed-item sources to review that domain, then practise a focused set and check the full official guide.

Loading domain-performance chart…

{{ examChartStatus }}

{{ examChartExportStatus }}

Answer ledger

Question ID Prompt Topic Your answer Correct answer Outcome Explanation Source Copy
{{ row.question_id }} {{ row.prompt }} {{ row.category_label }} {{ row.selected_answer_label }} {{ row.correct_answer_label }} {{ row.outcome_label }} {{ row.explanation }} {{ row.source ? `${row.source.label.split(' — ')[0]} · ${row.source.version}` : 'Official guide' }}{{ sourceUrl(row.question_id) }}

{{ examLedgerStatus }}

AWS Cloud Practitioner Exam Prep Practise AWS Cloud Practitioner CLF-C02 concepts with original questions, immediate explanations and focused reviews of missed answers.
IAM Policy Permissions Analyzer Review AWS IAM policy JSON for wildcard breadth, privilege paths, and policy-shape issues before validating effective access in AWS.
AWS Lambda Cost Calculator Estimate monthly AWS Lambda Functions cost across requests, GB-seconds, provisioned capacity and optional meters with editable regional rates.
S3 Bucket Policy Public Access Checker Review an S3 bucket policy for public or cross-account grants, test trust boundaries, and prepare the live AWS checks needed before approval.

Predicting a delivery time from historical measurements is different from applying a fixed expense limit or answering questions about a policy that changed yesterday. The first task can use learned patterns, the second may need only an exact rule, and the third needs access to current evidence. Choosing an approach that fits the requirement is central to foundational AI literacy.

AWS Certified AI Practitioner (AIF-C01) assesses foundational understanding of artificial intelligence (AI), machine learning (ML), and generative AI (GenAI) on Amazon Web Services (AWS). Its focus includes business use cases, technology selection, and responsible use. The intended learner uses AI technologies but does not necessarily build models. Recognizing suitable services and explaining their limitations matters more at this level than writing training code.

Machine learning uses examples to learn patterns rather than relying only on explicitly written rules. Supervised learning uses known outcomes, such as delivery times paired with package measurements. Unsupervised methods can discover groups without preassigned labels. Generative AI produces content such as text or images, but producing a plausible response does not establish that its facts are correct.

Training and inference
Training adjusts a model's learned parameters. Inference applies a trained model to new input, such as classifying a support message, without itself retraining the model.
Foundation models and large language models
A foundation model is trained on broad data and can support multiple tasks. A large language model (LLM) works with language. Input capabilities still differ, so text support alone does not establish that a model can interpret photographs.
Tokens and context
Tokens are units a model processes, not necessarily whole words. Context is the information available for a request, including instructions and relevant supporting material. Context limits and token usage affect application design and cost.

For an employee assistant, relevant information may include an authorized policy passage or an order record. A model does not automatically know the application's current data. Supplying useful evidence is different from teaching a model a preferred writing style, and neither approach removes the need to check the answer. An invented citation is a hallucination even when the surrounding explanation sounds convincing.

Amazon Bedrock provides managed access to foundation models, but model selection still depends on the job. Required languages, image or text inputs, response time, cost, and data restrictions can rule out an otherwise capable model. A support assistant that drafts quickly but requires lengthy corrections may not improve handling time. Representative evaluation examples and measures of the complete workflow are more informative than one impressive answer.

Responsible use also concerns who receives errors and who can access information. High overall accuracy can hide poor performance for a particular group. Clear decision explanations, a route for review, and access limited to the required records address different risks. Content filtering can help, but it does not guarantee that every response is safe or accurate.

Practice questions help expose confused concepts and guide further study. They do not demonstrate hands-on deployment skill, establish that a system meets security or governance requirements, or confer certification. Foundational exam preparation should develop the reasoning behind a service choice rather than just recognition of a familiar answer.

How to Use This Tool:

This is independent, unofficial, untimed practice. A fresh setup starts with All domains, five questions, and Learn — feedback after each answer.

  1. Choose a Domain / practice focus and an offered Question count. The mixed pool contains 40 questions. A focused domain, Concept checks, or Applied scenarios uses a subset of that same bank. An unavailable count resets to five when the focus changes.
  2. Select Feedback timing. Learn — feedback after each answer reveals the correction immediately. Review — feedback at completion withholds corrections until the completed results.
    An optional Replay seed under Advanced changes question and option order. Surrounding spaces are ignored. Blank or unusable text uses a stable default, so leaving it blank does not request a fresh random round.
  3. Select Start exam practice and choose one of four answers. Selection submits and locks the answer immediately. Continue with Next question and open View results after the final answer.
    End round clears the attempt and returns to setup without a partial result. If starting is disabled, select an available focus, feedback mode, and count.
  4. Read the explanations in Missed answer review and the Answer ledger. Retry missed makes a separate round from incorrect items; use Change setup to return to broader study.

Interpreting Results:

The running summary counts submitted answers, not correct answers. At completion, the main fraction reports correct answers out of the round's total. A progress reading of 100% means every question was answered, not that every answer was correct.

Domain performance shows counts of correct and missed answers, not weighted exam scores. A longer bar can simply mean that more questions came from that domain. A domain missing from the chart was not sampled, so it has no result to interpret. Use the explanations to identify the specific distinction you missed rather than treating a small sample as a measurement of the whole domain.

The feedback sources lead to the relevant official domain guide. They provide study context, not AWS approval of an independently authored question or its answer key. Check the linked objectives and, where a service capability matters, its current documentation.

A Perfect round describes only the selected questions. Replaying familiar items or correcting previous misses can reinforce learning, but those results are not independent evidence of exam readiness. Compare understanding with the full official guide and AWS's official preparation resources, not a practice-score cutoff.

When the domain chart is unavailable, use Retry chart and the answer ledger. The chart's loading problem does not change the recorded answers or the answer ledger.

Technical Details:

Single-choice practice measures agreement with an authored best answer for each stated scenario. Every question contributes equally. Changing the topic mix changes what is being tested, even when two rounds have the same number of questions.

Practice points and completion

Let N be the total questions in the round, A the number answered, and C the number answered correctly. Each correct answer earns one point and each incorrect answer earns zero. There is no partial credit or deduction for a wrong answer. The score in points is the correct count, and the possible points equal the round length.

Score points=C Possible points=N Accuracy (%)=100×CN

The total ranges from zero to N points, equivalent to 0% to 100%. The accuracy calculation uses all questions in the round, not only those already answered. An unanswered question therefore stays in the denominator without being classified as an incorrect answer. During a round, completion progress and the number of recorded misses follow these rules.

Progress (%)=100×AN Missed answers=A−C

Question and point counts remain exact integers. The visible progress percentage is rounded to the nearest whole percent, with halves rounded up. A completed review requires every question to be answered and the final View results action. Its review message has only the following two outcomes, not a graded readiness scale.

Review message Condition after all questions are answered Meaning
Start with the missed answers One or more incorrect answers. Review the missed items and their explanations.
Perfect round Zero incorrect answers. Every selected item was answered correctly; no mastery or pass classification is added.

Blueprint and question-bank coverage

The reference target is AIF-C01 exam guide version 1.1, published April 30, 2026. The original English question bank is dated September 11, 2026, and its domain references carry that verification date. The bank is organized against the guide's 14 task statements, but the 40 questions address selected concepts rather than every objective or service. Those labels do not establish exhaustive coverage.

Official domain weights describe the certification exam's scored content. The practice-bank counts below describe available questions, not exam quotas or evidence that a domain has been exhaustively tested.

Official domain Exam weight Bank questions Selected practice concepts
Fundamentals of AI and ML 20% 8 Learning approaches, inference, held-out evaluation, recall, and speech-to-text service selection.
Fundamentals of GenAI 24% 10 Tokens, embeddings, hallucinations, input capabilities, context, agent concepts, and business value.
Applications of Foundation Models 28% 12 Retrieval, prompting, customization, evaluation, model selection, and controlled tool use.
Guidelines for Responsible AI 14% 5 Subgroup performance, content safeguards, model documentation, transparency, and rights review.
Security, Compliance, and Governance for AI Solutions 14% 5 Least privilege, private connectivity, data origins, activity history, and processing locations.

Selection is uniform without replacement within the chosen pool: a question cannot appear twice in one round. No official domain quota is enforced. All domains names the eligible pool; a short mixed round can omit domains entirely. Selecting all 40 questions includes the full bank, with the domain counts shown above.

The 18 concept checks and 22 applied scenarios are author-assigned practice groupings, not calibrated levels of official exam difficulty. Their scores cannot be compared as though the two pools were equally difficult or sampled identical subject matter.

Repeatability and question scope

Each item has one authored best answer and three authored alternatives. The seed determines question selection, question order, and option order. A seed reproduces the round when the focus, count, feedback mode, included question pool, and bank are unchanged. Remembering an answer's position is therefore less useful than understanding its wording.

Replay same seed clears responses and recreates the round. Retry missed creates a new round from the incorrect items only, which can contain just one question even though the normal setup menus start at two. Try a new seed uses a different seed; it does not add questions or guarantee unseen ones. When the full pool is already selected, a new seed can change order but not which questions belong to that pool. Use Change setup to return from a missed-item subset to broader practice.

Distinguishing model adaptation methods

Several foundation-model questions turn on whether a method supplies information for a request or changes learned model parameters. These differences explain why a plausible-sounding alternative may not meet the stated requirement.

Method What changes Important boundary
Few-shot prompting A few examples of the desired response are included in the request. The examples guide inference without updating model parameters.
Retrieval Augmented Generation (RAG) Relevant source material is retrieved and supplied to support generation. Source freshness, access permissions, and citation accuracy still require attention.
Supervised fine-tuning Training examples adjust a supported model's parameters for a task or response style. A completed training run does not automatically incorporate documents written later.
Model distillation A student model is trained using a teacher model's outputs. A smaller student still needs evaluation; equal performance is not guaranteed.

Embeddings are numerical representations used to compare meaning for similarity-based retrieval. Similarity helps find potentially relevant material; it does not establish that a passage is true or authorized for a user. Lower temperature generally reduces variation in generated text, but it does not verify facts or ensure identical results in every deployment.

These practice scenarios are conceptual, not checks of a live AWS configuration. Model support, service availability, and processing locations can vary. For a residency requirement, the actual inference route matters, including cross-Region behavior. A selected source Region or private connection alone does not establish where every request is processed.

Why the official exam score is different

The official exam has 65 questions in 90 minutes, including 50 scored and 15 unidentified unscored questions. Its formats can include multiple choice, multiple response, ordering, and matching. Official unanswered questions count as incorrect, with no additional penalty for guessing. This practice is untimed and single-choice, with every selected item contributing to its result.

AWS reports a scaled score from 100 to 1,000, with 700 as the minimum passing score. That is not a raw 70% rule. No published conversion maps this practice fraction to an AWS scaled score, and no pass prediction or certificate is produced. The questions are independently authored, not official examination questions, and this practice is not affiliated with or endorsed by AWS.

Privacy and Saved Attempts:

Answer checking happens in your browser, but an attempt is not necessarily private to that device. The page address carries encoded round settings, responses, and the current position so that a compatible attempt can be restored.

  • The encoded address is not encrypted. Sharing it can disclose your responses, and it can remain in browser history or bookmarks. On reload, the address can also reach the website. Do not put personal information or credentials in the replay seed.
  • Page resources, optional chart and document resources, and opened references can use the network. Browser scoring is not a promise of fully offline operation or zero transmission.
  • Saved addresses depend on compatible question content and attempt rules. They are not permanent archival records. Exported answer ledgers also contain your responses and should be shared deliberately.

A malformed or incompatible saved address returns to setup with Saved attempt unavailable. Start a new round with the available settings rather than interpreting the rejected attempt as a zero score.

Worked Examples:

Nine correct answers in one domain

Select Applications of Foundation Models, 12 questions, and Review — feedback at completion. Finishing with nine correct answers produces a main result of 9/12, equivalent to 75% under the practice formula, with three missed answers. The domain chart contains nine correct and three missed answers for that domain only. Review those three explanations; the result says nothing about the four domains that were not selected.

A perfect retry has a smaller denominator

In a 40-question mixed round, suppose the only mistake is choosing clustering for the delivery-time question with known historical outcomes. The result is 39/40, equivalent to 97.5%, with one missed item. Retry missed then creates a one-question round. Choosing supervised learning earns 1/1 because the examples pair inputs with known target values. That perfect retry records a correction on one familiar question; it does not replace the earlier 39/40 with a new 40/40 result.

References: