October 22, 2026

Module Overview

Google Gemini can support learning, research, writing, programming, and business productivity. However, using AI effectively requires more than knowing how to write prompts. Users must also understand privacy controls, recognize AI-generated errors, protect sensitive information, and verify important outputs.

This module focuses on using Gemini responsibly and evaluating its results before relying on them.

You will learn how to review available settings, understand the limitations of AI-generated answers, follow safe data-handling practices, and complete practical exercises across writing, marketing, research, and coding.

Important: Gemini settings, privacy controls, retention policies, and available features may change over time. Their behavior can depend on your account type, region, age, subscription, and organizational policies. Always review the current settings and explanations shown in your own account.

Learning Objectives

By the end of this module, you will be able to:

• Understand Gemini's available settings for activity, privacy, and personalization.

• Review account controls and make informed privacy decisions.

• Explain what AI hallucinations are and why they occur.

• Verify AI-generated facts, citations, calculations, and code.

• Apply privacy and security practices when using AI tools.

• Recognize when human expertise or additional evidence is necessary.

• Complete practical Gemini exercises across multiple real-world tasks.

Topic 37: Google Gemini Settings Explained: Privacy, Activity and Personalization Controls

1. Introduction to Gemini Settings

Settings determine how certain aspects of a digital service behave. In Gemini, available settings may include account preferences, activity controls, personalization options, connected services, and other feature-specific choices.

Understanding these controls helps you use Gemini according to your needs while making informed decisions about the information you share.

Remember that not every setting affects every Google service. A control in Gemini may have a different scope from a control in another Google product.

2. Understanding Activity Controls

Activity controls help determine how certain interactions are handled or saved, depending on the current product configuration.

Before changing a setting, understand:

• What activity the setting covers.

• Whether it affects future interactions, existing records, or both.

• Whether it influences the availability of certain features.

• Whether separate controls apply to connected services.

• Whether organizational policies override or restrict the setting.

Do not assume that turning off an activity option automatically deletes every existing record or removes information stored elsewhere.

How to review activity settings

Step 1: Open the Gemini app or website.

Step 2: Open the account menu or settings area available in your interface.

Step 3: Locate the relevant activity, privacy, or data controls.

Step 4: Read the explanation provided for each control.

Step 5: Decide whether the current setting matches your privacy requirements.

Step 6: If you change a setting, review its confirmation or explanatory message to understand the effect.

The exact menu names and navigation steps may differ across versions and account types.

3. Understanding Saved Conversations

Conversation history can help you return to previous work and continue an unfinished task.

For example, you may want to revisit:

• A website development discussion.

• A lesson plan.

• A marketing campaign outline.

• A research summary.

• A programming explanation.

However, conversation history and activity records are not necessarily identical. Their storage and management options may differ.

Before discussing sensitive information, review the applicable controls and organizational requirements.

Practical example

Suppose you are using Gemini to draft a business proposal.

You can begin with a fictional company and sample figures while developing the structure. Once the draft is ready, insert approved business information through your authorized workflow.

This approach reduces unnecessary exposure of sensitive details during early brainstorming.

4. Understanding Personalization Controls

Personalization can help tailor responses to recurring preferences when the relevant features are available and enabled.

Examples of useful preferences include:

• Use beginner-friendly explanations.

• Provide concise executive summaries.

• Use British or American English consistently.

• Include practical coding examples.

• Organize complex answers into numbered steps.

• Explain technical terms before using advanced terminology.

Example personalization instruction

"Explain technical topics in clear English. Begin with a concise overview, then provide numbered steps, practical examples, common mistakes, and a short summary. Distinguish verified facts from assumptions."

Personalization can improve consistency, but it does not guarantee that answers will be correct.

Avoid including passwords, API keys, confidential client information, or other unnecessary sensitive details in personalization instructions.

5. Review Connected Apps and Permissions

If Gemini offers connected apps or integrations for your account, review the permissions before enabling them.

Ask yourself:

• What information can the connection access?

• Is the connection necessary for my task?

• Does it support information retrieval, actions, or both?

• Can an administrator restrict the connection?

• How can access be reviewed or revoked?

• What data-handling rules apply?

Only enable connections that are appropriate for the task.

6. Privacy Settings for Students and Businesses

Students

• Avoid sharing student identification numbers unnecessarily.

• Use fictional examples when asking for help with assignments.

• Do not upload confidential school records without authorization.

• Review the applicable school or family rules.

• Check AI-generated explanations before using them in academic work.

Businesses

• Follow internal AI-use and data-classification policies.

• Avoid entering customer records unless the workflow is approved.

• Use anonymized or synthetic data when practical.

• Review permissions for connected business services.

• Require human approval for consequential decisions and external communications.

7. A Practical Privacy Review Checklist

Before using Gemini for an important task, ask:

1. Is the information public, internal, confidential, or highly sensitive?

2. Is this information necessary to complete the task?

3. Can I replace identifying details with fictional examples?

4. Have I reviewed the applicable account and activity controls?

5. Am I authorized to share this information with the service?

6. Does the task require an approved business or school environment?

7. Will the output be reviewed before it is shared or acted upon?

Practical Exercise

Review the settings available in your Gemini account.

Record:

• Which activity controls are available.

• Which personalization options are available.

• Whether connected-app permissions are listed.

• Which controls you would review before handling business or educational information.

• Any questions you need to resolve before using Gemini for sensitive tasks.

Do not include passwords, account identifiers, or confidential information in your exercise notes.

Key Takeaway

Understanding activity controls, personalization, and connected-app permissions helps you make informed decisions about your use of Gemini. Always review the actual scope of each setting rather than assuming that one control covers every service or record.

Topic 38: Google Gemini Limitations and Hallucinations: How to Check AI-Generated Answers

1. What Is an AI Hallucination?

An AI hallucination occurs when an AI system produces information that appears plausible but is false, unsupported, or inconsistent with the available evidence.

For example, Gemini might:

• State an incorrect date.

• Invent a book title or research citation.

• Misinterpret a technical specification.

• Produce a mathematical error.

• Explain code incorrectly.

• Present an assumption as an established fact.

A confident tone does not prove that an answer is accurate.

2. Why Can Hallucinations Occur?

AI systems generate responses based on learned patterns and the information available in the current task. They may lack relevant context, misunderstand a question, or produce an answer that is not adequately supported by evidence.

Common contributing factors include:

• Ambiguous prompts.

• Missing or outdated information.

• Complex reasoning requirements.

• Incorrect assumptions.

• Limited context.

• Misinterpretation of source material.

• Requests for details that are not available.

A longer or more detailed answer is not automatically more reliable.

3. Common Limitations of Gemini

A. Incorrect factual claims

Gemini may provide an incorrect name, date, statistic, or explanation.

What to do: Check important claims against reliable primary sources.

B. Outdated information

A response may not reflect the latest product feature, law, policy, or technical standard.

What to do: Verify time-sensitive information using current authoritative sources.

C. Incomplete answers

An answer may omit a critical limitation, exception, or alternative.

What to do: Ask what assumptions the response makes and which important cases it does not cover.

D. Incorrect citations

A citation may be incomplete, irrelevant, or fail to support the statement it accompanies.

What to do: Open the source and verify the relevant passage.

E. Incorrect calculations

Gemini may make arithmetic or data-interpretation mistakes.

What to do: Recalculate important figures independently.

F. Incorrect code

Generated code may contain syntax errors, logic problems, security weaknesses, or incompatible dependencies.

What to do: Run tests, inspect the implementation, and review security before deployment.

4. The Five-Step Verification Method

Use this method whenever the accuracy of an answer matters.

Step 1: Identify the important claims

Break the response into statements that can be checked.

For example:

• A product supports a particular feature.

• A statistic describes a particular year.

• A code function returns a specific value.

• A source supports a particular conclusion.

Step 2: Find suitable evidence

Prefer sources appropriate to the claim.

Examples:

• Official documentation for product behavior.

• Government publications for official statistics and regulations.

• Original research for scientific claims.

• Source code and tests for software behavior.

• Reputable financial documents for reported financial figures.

Step 3: Check the source directly

Do not rely only on a search-result snippet or AI-generated summary. Read the relevant section of the original source.

Step 4: Compare the claim with the evidence

Check whether the source actually supports the exact statement. Pay attention to dates, definitions, units, scope, and qualifications.

Step 5: Correct or qualify the answer

If evidence is incomplete or conflicting, revise the claim, explain the uncertainty, or omit the unsupported statement.

5. Practical Example: Verify a Research Claim

Suppose Gemini states:

"Technology A is always faster than Technology B."

This claim is too broad without context.

Use the following prompt:

"Evaluate the claim that Technology A is always faster than Technology B. Identify the conditions under which the comparison might be valid, the benchmarks or evidence required, and the factors that could affect the result. Distinguish established evidence from assumptions. Do not invent benchmark results or citations."

You should then review relevant benchmark methodology, hardware, workloads, software versions, and original results.

A more defensible conclusion may depend on the particular workload and measurement method.

6. Verify AI-Generated Citations

When Gemini provides citations, check:

• Does the cited source exist?

• Is the title and author information correct?

• Is the publication date relevant?

• Does the cited passage support the claim?

• Is the source primary, secondary, or opinion-based?

• Does the source discuss the same population, timeframe, or conditions?

If the source cannot be located or does not support the claim, do not cite it as evidence.

Example prompt

"Review the citations in this draft. For each factual claim, identify the cited source and explain whether the source directly supports the claim. Flag missing, inaccessible, outdated, or mismatched citations. Do not create replacement references unless they can be verified."

7. Verify AI-Generated Code

Code must be evaluated by behavior, not appearance alone.

Use this workflow:

1. Define the expected behavior.

2. Review the generated implementation.

3. Run the code in the intended environment.

4. Test normal inputs.

5. Test boundary conditions.

6. Test invalid inputs.

7. Inspect errors and logs.

8. Review security and dependencies.

9. Run regression tests after changes.

Example prompt

"Review this function against the stated requirements. Identify likely syntax errors, logic errors, edge cases, and security concerns. Suggest a minimal correction and provide test cases with expected outputs. Do not claim that the code has been executed or tested unless that has actually occurred."

8. Verify AI-Generated Data Analysis

When Gemini analyzes a spreadsheet or CSV file, check:

• Whether it used the correct columns.

• Whether missing values were handled appropriately.

• Whether duplicate records affected the result.

• Whether units and date formats were interpreted correctly.

• Whether the totals and averages can be reproduced.

• Whether the conclusion goes beyond the available data.

For important decisions, compare results with a spreadsheet, database query, or independently written calculation.

9. Use Uncertainty Effectively

You can ask Gemini to make uncertainty more visible.

Useful prompts include:

• "Separate facts, assumptions, and recommendations."

• "List the claims that require external verification."

• "Identify the strongest counterargument."

• "Explain what information is missing."

• "Give a confidence assessment and justify it with evidence."

• "If the available evidence is insufficient, state that clearly."

A confidence statement is not itself proof. It must still be evaluated against evidence.

10. When Should You Avoid Relying on Gemini Alone?

Do not use an unverified AI answer as the sole basis for high-impact decisions involving:

• Medical diagnosis or treatment.

• Legal rights or obligations.

• Significant financial decisions.

• Safety-critical engineering.

• Cybersecurity incident response.

• Employment or admissions decisions.

• Other consequential decisions affecting people's rights or well-being.

Use appropriate professional expertise, authoritative sources, and established procedures.

Practical Exercise

Ask Gemini to explain a technical concept or analyze a factual claim.

Then:

1. Identify five checkable claims.

2. Find reliable evidence for the important claims.

3. Verify any cited sources.

4. Identify one limitation or uncertainty.

5. Rewrite the answer so unsupported claims are removed or qualified.

Key Takeaway

Gemini can accelerate information work, but accuracy must be established through evidence, testing, and critical review. The correct workflow is not simply to ask for an answer; it is to evaluate whether the answer is supported.

Topic 39: Google Gemini Privacy and Security: Safe AI Practices for Students and Businesses

1. Why AI Privacy and Security Matter

AI tools may be used to work with documents, code, emails, reports, research, and other information. Some of this information may be confidential or sensitive.

Safe AI use requires users to consider what they share, where information goes, who is authorized to access it, and how the resulting output will be used.

Privacy and security are related but different:

• Privacy concerns appropriate handling and use of information about people.

• Security concerns protecting information and systems against unauthorized access, misuse, or compromise.

A workflow can be convenient without being appropriate for every type of information.

2. Classify Information Before Sharing It

A practical approach is to classify information into broad categories.

Public information

Examples:

• Published product descriptions.

• Public documentation.

• Public website content.

This information may generally be suitable for ordinary AI tasks, subject to applicable rights and policies.

Internal information

Examples:

• Internal process notes.

• Draft plans.

• Non-public project details.

Share only through an authorized workflow that meets your organization's rules.

Confidential information

Examples:

• Unreleased financial reports.

• Customer contracts.

• Internal business strategies.

• Confidential source code.

Use only an approved environment with appropriate access and data-handling protections.

Highly sensitive information

Examples:

• Passwords and access tokens.

• Payment credentials.

• Sensitive personal records.

• Private authentication material.

Do not paste these into ordinary prompts. Use secure, purpose-built systems and approved procedures.

The exact classification system varies by organization. Follow the policy that applies to your work.

3. Protect Personal Information

Personal information may include names, addresses, phone numbers, email addresses, identification numbers, account records, and information about a person's private circumstances.

Before sharing information, ask:

• Is the personal information necessary?

• Can the task be completed using fictional data?

• Can identifiers be removed or replaced?

• Do I have permission to share the material?

• Is the chosen AI environment approved for this type of information?

Example

Instead of asking Gemini to summarize a real student record containing identifying details, replace the personal information with fictional values or use an approved anonymized dataset.

Removing a person's name alone may not be sufficient if other details can identify them.

4. Avoid Sharing Secrets in Prompts

Never include passwords, private keys, API keys, session tokens, or other authentication secrets in ordinary prompts.

For example, do not ask Gemini to debug a configuration file while leaving a real secret exposed.

Instead, replace the secret with a placeholder:

API_KEY=YOUR_API_KEY_HERE

This placeholder is illustrative. Do not put the actual secret into the conversation.

If a real credential has already been exposed, follow the appropriate incident procedure. This may include revoking or rotating the credential and reviewing relevant access logs.

5. Protect Business Information

Organizations should define clear rules for AI usage.

A practical business policy should address:

• Which AI services are approved.

• What information can be shared.

• Which account types may be used.

• How connected apps and permissions are managed.

• Whether generated content requires review.

• How AI-assisted code is tested.

• Who approves external communications.

• How suspected data exposure is reported.

Employees should not assume that a personal AI account is suitable for confidential work simply because it is convenient.

6. Understand Risks in AI-Generated Code

AI-generated code can introduce vulnerabilities if it is copied into a project without review.

Potential problems include:

• Unsafe input handling.

• Exposed credentials.

• Weak authorization checks.

• Insecure file handling.

• Vulnerable dependencies.

• Unvalidated external data.

• Improper error handling.

• Dangerous shell commands.

Safe coding workflow

1. Review the code before running it.

2. Understand any commands it proposes.

3. Use a controlled development environment.

4. Keep secrets in appropriate secret-management systems.

5. Validate input and enforce authorization on the server.

6. Review dependencies and their sources.

7. Run automated and manual security checks.

8. Test the application before deployment.

Never assume that code is safe simply because Gemini generated it.

7. Watch for Prompt Injection and Untrusted Content

Prompt injection is a type of attack in which untrusted content attempts to influence an AI system's behavior in unintended ways.

For example, a webpage or document being summarized might contain text instructing the AI to reveal sensitive information or ignore the user's intended task.

Treat content retrieved from websites, documents, emails, and connected services as potentially untrusted.

Good practices include:

• Use only the permissions needed for the task.

• Do not allow untrusted content to override security rules.

• Review outputs before consequential actions.

• Avoid giving an AI workflow unrestricted access to sensitive resources.

• Require appropriate confirmation before sending messages, modifying records, or executing commands.

8. Check AI-Generated Content Before Publishing

AI-generated text, images, code, and reports may contain:

• False claims.

• Personal information that should not be disclosed.

• Copyright or licensing concerns.

• Misleading statements.

• Inappropriate content.

• Unsupported marketing promises.

• Incorrect technical instructions.

Before publishing or sharing a result, verify its accuracy, ownership, permissions, and suitability for the intended audience.

For business communications, ensure that any required human approval is completed.

9. Safe AI Practices for Students

Students should:

• Use AI as a learning aid rather than blindly copying answers.

• Follow school or university policies.

• Avoid uploading classmates' private information.

• Verify explanations and factual claims.

• Understand code before submitting or deploying it.

• Clearly disclose AI assistance when required.

• Avoid sharing account credentials or private records.

10. Safe AI Practices for Businesses

Businesses should:

• Use approved AI services and accounts.

• Classify information before sharing it.

• Apply least-privilege access.

• Review connected-app permissions.

• Verify outputs before making consequential decisions.

• Keep sensitive credentials out of prompts and code.

• Establish incident-reporting procedures.

• Provide employee training on AI risks.

Practical Exercise: Perform an AI Safety Review

Imagine that a company employee wants to use Gemini to summarize a customer database and generate a marketing campaign.

Evaluate the workflow:

1. Does the database contain personal information?

2. Is the selected AI environment approved?

3. Can a synthetic or anonymized sample be used?

4. Are the relevant permissions appropriate?

5. Could the output reveal confidential details?

6. Are marketing claims accurate and authorized?

7. Who reviews the final campaign?

8. What should happen if sensitive information is exposed accidentally?

Write a one-page safety checklist based on your answers.

Key Takeaway

Safe AI use requires more than adjusting a setting. Users must protect sensitive information, control access, review generated outputs, and follow the security and privacy policies that apply to their work.

Topic 40: 40 Practical Google Gemini AI Exercises for Writing, Marketing, Research and Coding

1. Introduction

The best way to develop practical Gemini skills is through repeated use on clearly defined tasks.

The following 40 exercises cover writing, communication, education, research, multimodal work, productivity, marketing, coding, data analysis, custom workflows, and safety.

The exercises are designed to help you practice prompt writing, evaluate output quality, and learn how to verify AI-generated results.

Exercise rule: Do not assume that Gemini has completed an action, accessed a file, executed code, or verified a source unless the interface or an independent test confirms it.

Category A: Writing and Communication

Exercise 1: Write a Blog Introduction

Task: Draft an introduction for a beginner-friendly article about website development.

Prompt:

"Write a 150-word introduction for a beginner-friendly article titled 'How to Start Learning Website Development'. Explain why the topic matters, identify the intended reader, and preview the article's main sections. Use clear English and avoid unsupported statistics."

Success criteria: The introduction matches the audience, has a clear purpose, and does not contain fabricated statistics.

Exercise 2: Rewrite Text Professionally

Task: Improve an informal business message.

Prompt:

"Rewrite the following message in a professional, polite tone while preserving its meaning. Correct grammar and improve clarity. Do not add promises or facts that are not in the original message."

Provide a short, non-sensitive sample message.

Success criteria: The revised text is clear, professional, and faithful to the original meaning.

Exercise 3: Draft a Professional Email

Task: Request a project update.

Prompt:

"Draft a concise professional email requesting a project status update. Include a clear subject, a polite opening, the information required, a reasonable requested response date represented by a placeholder, and a courteous closing. Do not invent recipient names or commitments."

Success criteria: The email is complete, appropriately toned, and easy to customize.

Exercise 4: Summarize a Document

Task: Produce a structured summary of a supplied document.

Prompt:

"Summarize the supplied document in five bullet points. Identify the main argument, key evidence, important limitations, and any action items explicitly stated. Do not add information that is absent from the source."

Success criteria: The summary reflects the original document without invented conclusions.

Exercise 5: Translate and Compare

Task: Translate a short paragraph into another language.

Prompt:

"Translate the supplied paragraph into Hindi. Preserve its meaning and tone. Identify idioms or technical terms that may need adaptation, and provide a brief explanation of any translation choice that could be ambiguous."

Success criteria: The translation preserves meaning and handles context appropriately.

Category B: Learning and Research

Exercise 6: Explain a Difficult Concept

Task: Learn a technical concept at beginner level.

Prompt:

"Explain the concept of an API to a beginner. Use a simple analogy, define the important terms, give one practical example, and explain one limitation of the analogy."

Success criteria: The explanation is understandable without becoming technically misleading.

Exercise 7: Create Revision Notes

Task: Turn supplied lesson material into revision notes.

Prompt:

"Convert the supplied lesson material into organized revision notes. Use headings, definitions, important points, and a short recap. Preserve important qualifications and do not add unsupported facts."

Success criteria: The notes are organized and faithful to the supplied material.

Exercise 8: Generate a Quiz

Task: Test understanding of a lesson.

Prompt:

"Create ten multiple-choice questions from the supplied lesson. Provide four options per question and a separate answer key with explanations. Ensure that each question has one defensible best answer."

Success criteria: The questions cover the material and the answer key is accurate.

Exercise 9: Plan a Research Task

Task: Research an unfamiliar subject.

Prompt:

"Create a research plan for understanding the use of AI in education. List the main research questions, useful source types, potential biases, evidence requirements, and unresolved questions. Do not present the plan as completed research."

Success criteria: The plan is focused and identifies suitable evidence.

Exercise 10: Verify a Factual Claim

Task: Evaluate a claim using appropriate sources.

Prompt:

"Evaluate the following claim. Identify what evidence would confirm or refute it, suggest appropriate authoritative sources, and separate verified facts from assumptions. Do not invent citations or claim verification without examining the evidence."

Success criteria: The response identifies what can and cannot be established.

Category C: Images, Audio and Video

Exercise 11: Create an Image Prompt

Task: Write a prompt for an educational technology illustration.

Prompt:

"Create a detailed image-generation prompt for a clean educational illustration explaining the parts of a computer. Specify the layout, major components, labels, visual style, and readability requirements. Keep the design suitable for a beginner-level lesson."

Success criteria: The prompt is specific, well organized, and appropriate for its audience.

Exercise 12: Analyze a Screenshot

Task: Understand a supplied interface screenshot.

Prompt:

"Analyze the supplied screenshot. Describe the visible interface elements, identify any readable error message, and suggest troubleshooting steps based only on the visible evidence. Clearly state what cannot be determined from the screenshot."

Success criteria: The analysis does not invent hidden interface details.

Exercise 13: Plan a Short Video

Task: Create a 30-second educational video outline.

Prompt:

"Plan a 30-second educational video explaining responsive web design. Divide it into timed scenes with visuals, narration, on-screen text, and transitions. Keep the narration within a realistic speaking pace and avoid unsupported claims."

Success criteria: The scenes fit the duration and communicate one coherent lesson.

Exercise 14: Summarize a Transcript

Task: Extract key information from a supplied transcript.

Prompt:

"Summarize the supplied transcript. Identify the main points, decisions explicitly stated, open questions, and action items with owners only when the transcript names them. Do not invent missing names or deadlines."

Success criteria: The summary distinguishes recorded facts from missing information.

Exercise 15: Generate a Storyboard

Task: Plan a short promotional or educational video.

Prompt:

"Create a six-scene storyboard for a beginner coding course promotion. For each scene, provide the duration, visual description, narration, on-screen text, and transition. Keep the scenes consistent in tone and ensure the total duration is 30 seconds."

Success criteria: The scenes are visually coherent and the durations add up correctly.

Category D: Productivity and Google Workspace

Exercise 16: Draft an Email Reply

Task: Respond to a sample customer inquiry.

Prompt:

"Draft a polite reply to the supplied customer inquiry. Address the customer's main question, distinguish confirmed information from missing details, and use placeholders for anything that requires confirmation."

Success criteria: The reply is useful without inventing company policies or commitments.

Exercise 17: Improve a Document

Task: Review a short report.

Prompt:

"Review the supplied report for clarity, structure, grammar, and consistency. Provide the original issue, the proposed revision, and a short explanation. Preserve the intended meaning and flag factual claims that need verification."

Success criteria: Improvements are traceable and preserve meaning.

Exercise 18: Generate Spreadsheet Formulas

Task: Calculate totals in a spreadsheet.

Prompt:

"Assume unit prices are in column B and quantities are in column C, beginning in row 2. Provide a spreadsheet formula to calculate each row's total in column D. Explain the formula and state how blank or invalid cells should be handled."

Success criteria: The formula matches the stated column structure and the assumptions are clear.

Exercise 19: Plan a Presentation

Task: Outline a short presentation.

Prompt:

"Create an eight-slide presentation outline about cybersecurity awareness for beginners. For each slide, provide a title, three key points, a suggested visual, and speaker notes. Keep technical advice accurate and identify points that need review against organizational policy."

Success criteria: The outline has a logical flow and practical content.

Exercise 20: Create a Weekly Work Plan

Task: Organize a fictional workweek.

Prompt:

"Create a five-day work plan for completing a small website project. Include daily objectives, tasks, dependencies, and a review step. Identify assumptions and leave realistic time for testing and revisions."

Success criteria: The plan is actionable and includes verification time.

Category E: SEO and Digital Marketing

Exercise 21: Generate Blog Topic Ideas

Task: Brainstorm article ideas for a fictional web development training institute.

Prompt:

"Generate 15 blog topic ideas for beginners learning web development. Group them by search intent and learning stage. Avoid claiming that any topic has verified search volume unless actual keyword data is supplied."

Success criteria: Ideas are relevant, organized, and not supported by fabricated metrics.

Exercise 22: Create a Content Calendar

Task: Plan one week of social media content.

Prompt:

"Create a seven-day social media calendar for a fictional coding education brand. Include the topic, platform, content format, key message, and CTA. Vary the content types and avoid repetitive ideas."

Success criteria: The calendar is coherent and suitable for the audience.

Exercise 23: Draft SEO Metadata

Task: Write a page title and meta description.

Prompt:

"Using the supplied webpage content, propose an SEO title and meta description that accurately represent the page. Provide one primary version and two alternatives. Avoid unsupported claims, keyword stuffing, and promises of ranking improvements."

Success criteria: Metadata accurately reflects the page and reads naturally.

Exercise 24: Create Ad Copy

Task: Draft sample advertising text.

Prompt:

"Write three alternative advertisements for a fictional beginner web development course. Each version should include a headline, short body copy, and CTA. Avoid fabricated testimonials, unverified discounts, and guaranteed employment or income claims."

Success criteria: The ads are clear, persuasive, and truthful.

Exercise 25: Develop YouTube Content Ideas

Task: Plan a short educational video.

Prompt:

"Generate ten YouTube Shorts ideas for teaching beginner JavaScript. For each idea, include a title, opening hook, core lesson, and suggested closing CTA. Do not promise viral performance or guaranteed views."

Success criteria: Ideas are practical, varied, and appropriate for short-form video.

Category F: Coding and Data Tasks

Exercise 26: Generate HTML

Task: Create a basic semantic page structure.

Prompt:

"Create a semantic HTML page for a fictional course catalog. Include a header, navigation, main heading, three course cards, and a footer. Use accessible link text and provide only the HTML. Explain how the page structure supports accessibility."

Success criteria: The markup is valid in structure and uses suitable semantic elements.

Exercise 27: Generate Responsive CSS

Task: Style the course catalog.

Prompt:

"Write responsive CSS for the course catalog described in the previous exercise. Use CSS Grid, consistent spacing, readable typography, visible focus states, and a mobile layout. Explain the breakpoints and identify how to test the design."

Success criteria: The CSS meets the layout requirements and includes appropriate responsive behavior.

Exercise 28: Debug JavaScript

Task: Diagnose a logic error.

Prompt:

"Review this function: function square(n) { return n * 2; }. The intended behavior is to return the square of a number. Identify the problem, provide the smallest fix, and give three test cases with expected results."

Success criteria: The fix uses multiplication by the input itself and the tests reflect the intended behavior.

Exercise 29: Analyze CSV Data

Task: Summarize a small dataset.

Prompt:

"Analyze the supplied CSV data. Calculate totals by category, identify missing or duplicate records, and summarize the main observations. Show the calculations and clearly state which conclusions are limited by the supplied data."

Success criteria: Results are reproducible and limitations are explicit.

Exercise 30: Generate Unit Tests

Task: Test a small function.

Prompt:

"Write unit tests for a function that converts a temperature from Celsius to Fahrenheit. Include normal values, zero, a negative value, and at least one boundary-related test. State the expected results and the testing framework used."

Success criteria: The tests are logically correct and cover more than one ordinary input.

Category G: Custom Assistants and Connected Workflows

Exercise 31: Design a Custom Assistant

Task: Write instructions for a study assistant.

Prompt:

"Create a reusable instruction set for a study assistant. Define its role, intended audience, response structure, exercise format, and rules for uncertainty. Keep the instructions specific enough to test."

Success criteria: The instructions describe a repeatable workflow.

Exercise 32: Improve a Gem

Task: Evaluate and improve a custom assistant.

Prompt:

"Review the supplied custom-assistant instructions. Identify vague requirements, conflicting rules, and missing quality checks. Suggest a revised version and explain how to test whether the revision improves results."

Success criteria: The revision is clearer and its effectiveness can be evaluated.

Exercise 33: Plan a Deep Research Task

Task: Define a multi-source research project.

Prompt:

"Create a structured research plan comparing two website development approaches. Specify the research questions, source types, comparison criteria, evidence requirements, and limitations. Do not invent findings before the research is conducted."

Success criteria: The plan is focused, testable, and evidence-oriented.

Exercise 34: Evaluate a Connected-App Workflow

Task: Plan a safe document-retrieval workflow.

Prompt:

"Design a workflow for locating relevant project notes through an authorized connected service. Specify required permissions, search criteria, source verification, privacy safeguards, and fallback steps if the connection is unavailable."

Success criteria: The workflow respects access limitations and includes verification.

Exercise 35: Build a Reusable Prompt Template

Task: Create a reusable prompt for summarizing reports.

Prompt:

"Create a reusable report-summary prompt with placeholders for audience, report purpose, desired length, required sections, and verification needs. Include rules for missing information and unsupported conclusions."

Success criteria: The template can be reused across different reports without inventing content.

Category H: Accuracy, Privacy and Security

Exercise 36: Fact-Check an AI Answer

Task: Verify an answer that includes factual claims.

Prompt:

"Break the supplied answer into individual factual claims. For each claim, identify the evidence needed, assess the supplied source if available, and label the claim as supported, contradicted, or unresolved. Do not invent sources."

Success criteria: Claims are evaluated individually and uncertainty is visible.

Exercise 37: Review Privacy Risks

Task: Assess a fictional dataset before using AI.

Prompt:

"Review this fictional dataset description for potential privacy risks. Identify direct identifiers, possible indirect identifiers, unnecessary fields, and safer alternatives. Do not request real personal information."

Success criteria: The review identifies data that could reveal identities and suggests proportionate safeguards.

Exercise 38: Review Generated Code for Security

Task: Identify potential code vulnerabilities.

Prompt:

"Review the supplied code for potential security weaknesses, unsafe input handling, exposed secrets, and dependency risks. Prioritize findings, explain the evidence, and suggest safe remediation steps. Distinguish confirmed vulnerabilities from possible concerns."

Success criteria: Findings are specific, evidence-based, and actionable.

Exercise 39: Create an AI Usage Policy

Task: Draft a simple internal AI policy for a fictional training institute.

Prompt:

"Draft a concise AI usage policy for a fictional training institute. Cover approved tools, permitted information, prohibited data sharing, verification requirements, academic integrity, human review, and incident reporting. Mark organization-specific decisions as placeholders requiring approval."

Success criteria: The policy covers the major operational risks without inventing legal requirements.

Exercise 40: Perform a Complete Gemini Workflow Audit

Task: Evaluate an end-to-end AI-assisted task.

Prompt:

"Audit the supplied AI workflow from task definition to final output. Review prompt quality, information sources, privacy, permissions, accuracy checks, human approval, and recordkeeping. Identify gaps and propose a prioritized improvement plan. Do not claim that controls are implemented unless the supplied evidence demonstrates this."

Success criteria: The audit identifies concrete improvements and distinguishes documented controls from assumptions.

Module 10: Revision and Key Takeaways

This final module covered three essential areas of responsible AI use.

Topic 37 — Settings and Personalization: Understand available activity controls, personalization preferences, and connected-app permissions.

Topic 38 — Limitations and Hallucinations: Recognize that AI-generated content can be wrong and verify facts, sources, calculations, and code.

Topic 39 — Privacy and Security: Protect sensitive information, apply appropriate permissions, review generated code, and follow organizational policies.

Topic 40 — Practical Exercises: Apply Gemini across 40 exercises covering writing, research, productivity, marketing, coding, data analysis, custom workflows, and safety.

The Responsible Gemini Workflow

Use the following process for important tasks:

1. Define: Specify the goal and expected result.

2. Prepare: Provide appropriate context without unnecessary sensitive information.

3. Prompt: Give clear instructions and constraints.

4. Review: Inspect the generated output for errors and omissions.

5. Verify: Check factual claims, calculations, citations, and code.

6. Protect: Confirm that privacy, permissions, and security requirements are satisfied.

7. Approve: Obtain appropriate human review before consequential use.

8. Improve: Refine the prompt or workflow based on the results.

Final Module Assessment

Choose the best answer for each question.

1. What should you do before sharing sensitive information with Gemini?

A. Paste the entire dataset immediately
B. Review the information, permissions, applicable policies, and available controls
C. Assume that every account has identical privacy settings
D. Include all personal details to improve the response

2. What is an AI hallucination?

A. A guaranteed accurate response
B. A computer hardware failure
C. An AI-generated statement that may be false or unsupported
D. A type of spreadsheet formula

3. Why should you verify citations?

A. Every AI-generated citation is guaranteed to be correct
B. A source may be missing, irrelevant, or fail to support the claim
C. Citations never require checking
D. All sources have equal authority

4. What is the safest approach to AI-generated code?

A. Deploy it without testing
B. Assume that readable code is secure
C. Review, test, and assess it before use
D. Remove its validation logic

5. What is a good way to reduce unnecessary privacy exposure?

A. Share more personal information than needed
B. Use synthetic or appropriately anonymized examples where practical
C. Put passwords in reusable prompts
D. Ignore connected-app permissions

6. What should you do if Gemini cannot verify a claim?

A. Present it as confirmed anyway
B. Invent a supporting source
C. Mark it as unresolved or seek suitable evidence
D. Remove all uncertainty from the report

7. What is the purpose of regression testing?

A. To check whether existing behavior remains correct after changes
B. To automatically guarantee privacy
C. To create a marketing slogan
D. To replace all code review

8. What should a business do before using AI with confidential information?

A. Use any personal account without review
B. Follow approved tools, access controls, and data-handling policies
C. Share credentials to improve context
D. Assume that AI tools are exempt from company rules

9. What should you do before enabling a connected app?

A. Grant every permission
B. Review its actual access and determine whether it is necessary
C. Assume that it can perform every possible action
D. Ignore the connected account

10. Which statement best describes responsible Gemini use?

A. AI output should always be accepted
B. Verification is needed only for code
C. Clear prompts, appropriate data handling, verification, and human judgment work together
D. Personalization eliminates errors

Answer Key

1. B — Review the information, permissions, policies, and available controls.

2. C — An AI-generated statement may be false or unsupported.

3. B — Sources must be checked against the claims they support.

4. C — Review and test code before using it.

5. B — Synthetic or appropriately anonymized examples can reduce unnecessary exposure.

6. C — Mark unresolved claims or seek suitable evidence.

7. A — Regression tests check that existing behavior remains correct.

8. B — Use approved tools and follow data-handling policies.

9. B — Review access and necessity before enabling a connection.

10. C — Responsible AI use combines good prompts, safeguards, verification, and judgment.

Course Completion Project: Build a Responsible AI-Assisted Workflow

Project Objective

Create a small, complete workflow that uses Gemini to solve a practical problem while demonstrating accuracy, privacy, and quality control.

Choose one project:

• Draft and fact-check an educational article.

• Build and test a small JavaScript application.

• Create a social media content calendar.

• Analyze a fictional sales dataset.

• Research and compare two technical approaches.

• Design a reusable custom-assistant instruction set.

Required Deliverables

1. Task Definition

Document the goal, intended audience, expected output, and success criteria.

2. Prompt Design

Save the prompt you used and explain how its context, constraints, and output requirements support the task.

3. AI-Generated Output

Include the resulting draft, code, analysis, or plan. Label any unverified claims or assumptions.

4. Verification Evidence

Show how you checked important claims, calculations, citations, or program behavior.

5. Privacy Review

Explain what information was supplied, why it was necessary, and how sensitive information was protected.

6. Improvement Log

Document at least three improvements made after reviewing the first output.

7. Final Reflection

Explain what Gemini did well, what required human judgment, and what you would change in a future workflow.

Completion Checklist

• The task has clear success criteria.

• The prompt includes relevant context and constraints.

• No unnecessary sensitive information was shared.

• Important factual claims were verified.

• Code or calculations were tested where applicable.

• The final output was reviewed by a human.

• Limitations and unresolved questions were documented.

• At least three improvements were recorded.

• The final deliverable meets the original requirements.

Final Conclusion

You have now completed the 10-module Google Gemini AI tutorial course, covering fundamentals, prompting, writing, learning, multimodal work, Google Workspace, digital marketing, coding, custom workflows, and responsible AI use.

The most valuable skill is not simply knowing which buttons to click or which prompt to copy. It is learning how to define a task, provide suitable context, evaluate the response, verify important claims, and use AI within appropriate privacy and security boundaries.