Free Google Cloud Generative AI Leader Practice Questions: Fundamentals of Gen AI

Topic snapshot

Sample questions

These are original IT Mastery practice questions aligned to this topic area. They are not official Google Cloud questions, copied live-exam content, or exam dumps. Use them to preview question style and explanation depth before continuing with topic drills, mixed sets, and timed mocks in IT Mastery.

Question 1

Topic: Fundamentals of Gen AI

A retail company is announcing a generative AI feature that drafts customer-service email replies from policy documents. Leaders want language that encourages adoption, but legal and support managers are concerned about overstating accuracy or implying the system can work without oversight. Which wording is the best balanced recommendation?

Options:

• A. A machine learning model that replaces support representatives

• B. An autonomous agent that resolves customer issues independently

• C. An AI assistant that drafts responses for human review

• D. A knowledge engine that always returns verified answers

Best answer: C

Explanation: For non-technical stakeholders, gen AI terminology should describe the business capability without implying guaranteed correctness or full autonomy. A tool that drafts customer-service replies is best described as an assistant or copilot-style capability because it generates useful content from available context, but its output still needs human judgment. This balances adoption with transparency and governance. Terms like autonomous, always verified, or replacement overstate what the system can reliably do and can create unrealistic expectations, compliance risk, and resistance from users.

• Autonomous agent over-optimizes perceived capability and ignores the stated need for oversight.
• Always verified overstates accuracy because generative AI can produce incorrect or unsupported content.
• Replaces representatives harms adoption and misrepresents the human-in-the-loop role described in the scenario.

Question 2

Topic: Fundamentals of Gen AI

A marketing team wants to generate short product launch videos from text prompts and storyboards for rapid concept testing. Which Google foundation model family best fits this video-generation use case?

Options:

• A. Gemma

• B. Imagen

• C. Veo

• D. Gemini

Best answer: C

Explanation: Veo is the Google foundation model family aligned with video-generation use cases. In this scenario, the core requirement is to create short product launch videos from prompts and storyboards, so the model family should match video creation rather than text, image, or open model customization needs. Imagen is associated with image generation, Gemma with lightweight open models, and Gemini with multimodal reasoning and general assistant-style tasks. The key signal is the requested output format: video.

• Image model mismatch fails because the team needs moving video content, not still images.
• Open model mismatch fails because the requirement is not to use lightweight open models for customization.
• General model mismatch fails because broad multimodal assistance is less specific than a video-generation model family.

Question 3

Topic: Fundamentals of Gen AI

A retail company uses a generative AI tool to draft customer-service replies from approved policy excerpts. Managers need replies that are empathetic, match the brand voice, cite only the provided policy facts, and return a consistent CRM-ready format. The current prompt, “Write a reply to this customer,” produces varied tone, missing fields, and occasional policy claims not found in the excerpt. Which prompt change is the best fit?

Options:

• A. Remove policy excerpts to simplify the prompt

• B. Specify role, tone, source-use rules, output fields, and examples

• C. Use the shortest possible prompt for faster responses

• D. Ask for a more creative and conversational answer

Best answer: B

Explanation: Prompt quality strongly influences generated output because the model uses the prompt as task guidance. In this scenario, the business problem is not just wording; the replies must meet several constraints: empathetic tone, brand voice, factual grounding in approved excerpts, and a repeatable CRM-ready structure. A stronger prompt should define the assistant’s role, the intended audience, required tone, allowed sources, required fields, and any useful examples. That gives the model clearer boundaries for what to include, how to phrase it, and how to format it. Creativity or brevity alone would not solve the trust and consistency issues.

• More creativity may improve style, but it can increase variation and unsupported claims.
• Removing excerpts weakens factual grounding because the model loses the approved source material.
• Shortest prompt may reduce guidance, making tone, format, and consistency less reliable.

Question 4

Topic: Fundamentals of Gen AI

A retail operations team wants a gen AI assistant to summarize weekly store performance for regional managers. A pilot using only point-of-sale data produced confident explanations for sales drops, but the assistant did not have access to promotion calendars, inventory outage records, or local event notes. Leaders want useful summaries quickly, but they also need to avoid misleading recommendations. What is the best balanced recommendation?

Options:

• A. Connect key missing data before expanding the pilot

• B. Lower the model temperature for more consistent wording

• C. Fine-tune the model on past sales totals

• D. Launch with caveats and add missing sources later

Best answer: A

Explanation: Incomplete data can make gen AI outputs sound confident while missing the facts needed for business judgment. In this case, point-of-sale data shows what happened, but not why it happened. Promotion schedules, inventory outages, and local events are likely causal or contextual inputs for explaining sales changes. A balanced approach is to improve the pilot’s data readiness by connecting the most relevant missing sources before broader use, rather than waiting for a perfect enterprise dataset or launching outputs that managers may overtrust. The key trade-off is speed to value versus decision quality. Faster rollout is not valuable if the assistant produces weak or misleading recommendations from incomplete context.

• Launch-first approach optimizes speed, but it leaves managers exposed to confident explanations that omit known relevant data.
• Fine-tuning on sales totals may improve style or pattern fit, but it still lacks promotion, inventory, and event context.
• Lower temperature can make wording more consistent, but it does not fix missing facts in the input data.

Question 5

Topic: Fundamentals of Gen AI

A benefits provider plans a gen AI assistant that answers employee questions from plan documents. Leadership wants a pilot in 6 weeks, using only approved internal sources. The content team reports that all current plan documents are available in a shared repository, the documents are relevant and mostly complete, several documents list different copay amounts for the same plan and year, and most files are already searchable PDFs.

Which recommendation best balances speed to value and answer quality by addressing the main data-quality issue?

Options:

• A. Convert the PDFs into a new standardized file format.

• B. Purchase third-party benefits datasets to expand coverage.

• C. Launch with the existing documents and rely on user feedback.

• D. Reconcile conflicting values and publish a canonical source for grounding.

Best answer: D

Explanation: The main issue is data consistency: multiple approved internal documents give different values for the same plan and year. For a benefits Q&A assistant, inconsistent facts can lead to conflicting or incorrect answers even when the data is available, relevant, and mostly complete. A balanced pilot should not wait for a large data overhaul, but it should resolve the contradictions that directly affect answer quality. Creating or designating a canonical source gives the model a trusted grounding source while still supporting speed to value.

• Format focus misses that the PDFs are already searchable, so conversion does not solve the conflicting copay values.
• Coverage expansion ignores the constraint to use approved internal sources and does not address contradictions in existing data.
• Feedback-only launch prioritizes speed but risks low-trust answers on high-impact benefits information.

Question 6

Topic: Fundamentals of Gen AI

A retailer wants to predict which subscription customers are likely to cancel next month. It has three years of customer profiles, usage history, support history, and a confirmed cancel-or-renew outcome for each customer. Leaders need reliable performance metrics for governance and cannot yet run experiments that change customer offers. Which approach best balances these priorities?

Options:

• A. Train a supervised learning model on labeled outcomes

• B. Cluster customers with unsupervised learning

• C. Use reinforcement learning to optimize retention offers

• D. Prompt a generative model to estimate churn risk

Best answer: A

Explanation: The core concept is matching the machine-learning approach to the available training signal. This business problem has historical examples where each customer record includes inputs and a known outcome: canceled or renewed. That makes it a supervised learning fit, because the model can learn patterns that predict a labeled target and can be evaluated with governance-friendly metrics such as accuracy, precision, recall, or lift. The constraint against changing offers also makes reinforcement learning a poor first step, because it depends on actions, feedback, and reward signals from interaction. Unsupervised learning can reveal segments, but it does not directly optimize for a known churn label.

• Clustering may help explore customer segments, but it ignores the visible labeled cancel-or-renew signal.
• Reinforcement learning optimizes actions through feedback, but the business cannot yet run offer experiments.
• Prompting a generative model may produce explanations or summaries, but it is not the best primary approach for governed predictive modeling from labeled history.

Question 7

Topic: Fundamentals of Gen AI

A retail company has finished selecting and testing a generative AI model that creates product descriptions. Business users now need the model available inside the merchandising application, with access controlled for approved teams and ongoing monitoring after release. Which machine learning lifecycle stage best fits this work?

Options:

• A. Model training

• B. Data preparation

• C. Model evaluation

• D. Model deployment

Best answer: D

Explanation: Model deployment is the relevant lifecycle stage when the organization is ready to make a model usable by people, applications, or business workflows. In this scenario, the model has already been selected and tested, and the remaining work is to expose it through the merchandising application with access controls and monitoring. Those facts point to operationalizing the model, not improving the dataset or building the model. The key signal is that the model must become available for approved users and ongoing use.

• Data preparation focuses on collecting, cleaning, and organizing data before model development, not releasing a finished model.
• Model training creates or adapts the model using data, but the stem says the model is already selected and tested.
• Model evaluation checks model quality and suitability before release, but the business need is application availability and monitoring after release.

Question 8

Topic: Fundamentals of Gen AI

A CIO must choose where to focus a first gen AI investment. The business wants quick productivity gains for nontechnical employees, strong user adoption, and minimal custom development. The team is willing to accept less customization if the solution works inside existing collaboration tools and keeps enterprise governance in place. Which gen AI landscape layer is most directly affected by this decision?

Options:

• A. Infrastructure layer

• B. Model layer

• C. Application layer

• D. Platform layer

Best answer: C

Explanation: The decision most directly maps to the application layer because the CIO is prioritizing employee-facing productivity, adoption, and speed to value over custom model or infrastructure choices. In the gen AI landscape, applications are the finished or packaged experiences users interact with, such as assistants embedded in collaboration or productivity tools. The visible trade-off is accepting less customization to gain faster rollout and easier adoption within governed enterprise workflows. A platform-layer choice would fit better if the business needed to build, customize, evaluate, or deploy gen AI solutions. A model-layer or infrastructure-layer decision would focus more on selecting foundation models or compute capabilities.

• Infrastructure focus over-optimizes compute and scalability, but the stem does not require choosing GPUs, TPUs, or hosting capacity.
• Model focus emphasizes foundation model selection, but the key decision is not about Gemini, Imagen, Veo, or model behavior trade-offs.
• Platform focus fits custom building and governance tooling, but the stated priority is a ready employee-facing experience with minimal development.

Question 9

Topic: Fundamentals of Gen AI

A regional insurance company is piloting a generative AI assistant to summarize claim histories for adjusters. The pilot goal is to reduce review time, but adjusters report that summaries omit recent claim notes and sometimes include outdated customer details. Compliance leaders require explainable outputs that staff can trust before expanding adoption. What is the best next step?

Options:

• A. Increase the model temperature to make summaries more flexible

• B. Replace adjuster training with a shorter AI usage guide

• C. Launch to all adjusters and collect feedback later

• D. Assess and improve data completeness, freshness, and source traceability

Best answer: D

Explanation: Data quality is a core AI-readiness factor because model outputs depend on the accuracy, relevance, completeness, and timeliness of the data they use. In this scenario, the assistant is not failing because adjusters need more creativity or because rollout is too small. It is producing untrusted summaries because the underlying claim data is incomplete, stale, or not clearly traceable to reliable sources. Improving freshness, completeness, and source traceability supports business value by making summaries useful, reduces compliance risk by limiting outdated or misleading information, and builds stakeholder trust through explainable outputs. A rollout decision should wait until the data issues behind the output problem are addressed.

• More creativity fails because higher temperature can increase variation, not fix missing or outdated source data.
• Broader rollout fails because scaling an untrusted pilot increases operational and compliance risk.
• Shorter training fails because user education cannot compensate for incomplete, stale, or untraceable data.

Question 10

Topic: Fundamentals of Gen AI

A manufacturer is piloting a gen AI assistant that summarizes customer health for sales, support, and finance leaders. The assistant pulls from CRM records, support tickets, and finance spreadsheets, but the same customer often has different renewal dates and revenue figures in each source. Leaders need board-ready summaries they can trust across departments. What is the best professional decision before expanding the pilot?

Options:

• A. Ask each department to validate outputs manually after generation

• B. Define authoritative sources and reconcile inconsistent customer data

• C. Fine-tune the model on the current combined dataset

• D. Increase the model temperature to improve answer variety

Best answer: B

Explanation: Gen AI output quality depends heavily on the quality and consistency of the data it uses. In this scenario, the assistant is not failing because it lacks creativity or a larger model; it is using conflicting facts from different departmental systems. Before scaling the pilot, the organization should identify authoritative sources, reconcile mismatched records, and establish data governance for shared customer fields such as renewal date and revenue. This improves trust because all departments see outputs grounded in consistent business data. Fine-tuning or broader rollout would likely amplify the inconsistency rather than fix it.

• Temperature tuning affects randomness and style, not whether conflicting business records become accurate.
• Fine-tuning current data can teach the model patterns from inconsistent records, which may make unreliable outputs more persistent.
• Manual validation may reduce immediate risk, but it does not solve the underlying cross-department data inconsistency.

Continue in the web app

Use IT Mastery for interactive Google Cloud Generative AI Leader practice with mixed sets, timed mocks, topic drills, explanations, and progress tracking.

Try Google Cloud Generative AI Leader on Web

Related focused pages

• Free Google Cloud Generative AI Leader Full-Length Practice Exam
• Google Cloud’s Gen AI Offerings
• GenAI Output Quality
• GenAI Business Strategy