Chapter 17 — AB-730 Exam Readiness¶
Part IV — Exam readiness
In 30 seconds¶
- The core idea: consolidate everything for AB-730 (AI Business Professional) into a checklist, a set of high-yield facts, and a full mock exam.
- Why it matters: this is your final rehearsal before exam day.
- The exam angle: pass mark is 700; skills measured as of July 22, 2026.
Exam overview¶
- Domain 1 — Understand generative AI fundamentals (25–30%).
- Domain 2 — Manage prompts and conversations by using AI (35–40%).
- Domain 3 — Draft and analyze business content by using AI (25–30%).
Objective checklist¶
- [ ] Generative AI capabilities across Microsoft 365 (Ch 1, 2, 5)
- [ ] Responsible AI and data protection (Ch 4)
- [ ] Create and manage prompts (Ch 3, 6)
- [ ] Manage conversations (Ch 7)
- [ ] Create and manage agents (Ch 8)
- [ ] Draft business documents and communications (Ch 9)
- [ ] Meetings and collaboration (Ch 10)
High-yield facts¶
Fundamentals & responsible AI (Domain 1)
- Generative AI creates content; predictive AI classifies/forecasts. Match the verb.
- Copilot grounds prompts in your data via Microsoft Graph (RAG); your data is not used to train foundation models.
- Copilot only surfaces content you have at least view permission to; it honors Purview sensitivity labels and stays in the Microsoft 365 service boundary.
- Context that shapes answers: the app you're in, your files/emails/meetings, the web (optional), the conversation so far.
- Chat = open-ended conversation; agent = purpose-built, reusable, knowledge-scoped assistant.
- Four named risks: fabrications, prompt injection, over-reliance, bias. Verify with citation checks and human review; higher stakes → more review.
- Responsible-AI principles: fairness, reliability & safety, privacy & security, inclusiveness, transparency, accountability. Accountability stays with people.
Prompts, conversations, agents (Domain 2)
- Effective prompt = Goal · Context · Source · Expectations.
- Reference a source with
/. Relevance beats length. - Save a prompt (reuse) · Schedule a prompt (runs automatically on a cadence) · Share a prompt (team). The Copilot Prompt Gallery is home base. Sharing shares the prompt, not the data.
- Conversations: find, rename, delete; add to a notebook to keep working on a topic. Notebook = personal; Page = collaborative.
- Agents: check the Agent Store first; build your own (no code) for specific, repeatable needs. Configure instructions, knowledge, capabilities, suggested prompts; then share. Agents honor each user's permissions.
Draft & analyze content (Domain 3)
- Copilot: create a doc from a prompt or from an existing document (e.g., Word → PowerPoint); generate a management summary (audience + brevity); move insights between apps (re-expressed, not copied).
- Meetings: recap / catch-up require transcription or recording.
- Memory remembers context to personalize; instructions are standing preferences. Neither bypasses permissions.
- Always treat output as a draft to verify.
Mock exam — AB-730¶
40 original questions, weighted toward the exam's domains. Answers with explanations are under each item. Target ≥ 70% before sitting the real exam.
Domain 1 — Generative AI fundamentals & responsible AI¶
1. A bank wants to flag likely-fraudulent transactions. Which AI type fits?
- A. Generative AI
- B. Predictive machine learning
- C. A Copilot agent
- D. Image generation
Answer
B. Classifying transactions is predictive ML, not content creation.
2. How does Microsoft 365 Copilot use your organization's data?
- A. It was pretrained on your tenant
- B. It grounds prompts via Microsoft Graph at query time (RAG)
- C. It emails your data to Bing
- D. Your data trains the foundation model nightly
Answer
B. Grounding through Microsoft Graph; your data is not used to train the model.
3. A user can't see a document by themselves. Will Copilot use it to answer them?
- A. Yes, Copilot ignores permissions
- B. No — Copilot only grounds on content the user can access
- C. Only if it's in SharePoint
- D. Only on weekends
Answer
B. Copilot respects the user's existing permissions.
4. Which is a fabrication?
- A. Copilot cites a real source
- B. Copilot invents a confident but false statistic
- C. Copilot refuses a request
- D. Copilot summarizes an email
Answer
B. A fabrication is confident, plausible, but false content.
5. The best defense against fabrications on high-stakes work is:
- A. Trusting fluent answers
- B. Checking citations and having a human review
- C. Using more tokens
- D. Turning off grounding
Answer
B. Verification (citation checks, human review) scales with stakes.
6. A hidden instruction in a document tries to make Copilot leak data. This is:
- A. Bias
- B. Over-reliance
- C. Prompt injection
- D. A fabrication
Answer
C. Malicious instructions in content = prompt injection.
7. A hiring tool disadvantages one group. Which principle is violated?
- A. Transparency
- B. Fairness
- C. Inclusiveness
- D. Accountability
Answer
B. Systematic disadvantage is a fairness failure.
8. Who is accountable for AI output?
- A. The AI system
- B. Microsoft
- C. The person/organization using it
- D. No one
Answer
C. Accountability stays with people and organizations.
9. Which best distinguishes a chat from an agent?
- A. Agents only work offline
- B. An agent is a purpose-built, reusable assistant with its own instructions and knowledge
- C. Chat can't use work data
- D. There's no difference
Answer
B. Agents are configured and reusable; chat is open-ended.
10. Copilot honors which data-protection control on encrypted content?
- A. Microsoft Purview sensitivity labels
- B. None
- C. Only file names
- D. Screen brightness
Answer
A. Copilot respects Purview sensitivity labels and usage rights.
11. A company fears Copilot will surface an overshared confidential file. The accurate response:
- A. Copilot can expose any file regardless of permissions
- B. Copilot only surfaces what a user can already access; fix oversharing with governance (Purview)
- C. Disable Copilot entirely
- D. Copilot encrypts every file
Answer
B. Copilot enforces existing permissions; oversharing is a governance issue.
12. Which context does not typically shape a Copilot response?
- A. The app you're using
- B. Your recent emails and meetings
- C. The physical color of your monitor
- D. The conversation so far
Answer
C. Monitor color is irrelevant; the others are grounding context.
13. Over-reliance is best mitigated by:
- A. Trusting AI more
- B. Verification habits and training
- C. Faster hardware
- D. Longer prompts
Answer
B. Over-reliance is a human risk addressed by verification and training.
Domain 2 — Prompts, conversations, agents¶
14. The four elements of an effective prompt are:
- A. Goal, Context, Source, Expectations
- B. Who, What, When, Where
- C. Input, Output, Model, Token
- D. Question, Answer, Format, Length
Answer
A. Goal · Context · Source · Expectations.
15. A vague "summarize the project" gives a weak answer. Best fix?
- A. Repeat it
- B. Reference the specific project file/meeting as the Source
- C. Switch language
- D. Use more tokens
Answer
B. Add the missing Source.
16. A user wants a summary delivered automatically every Monday. They should:
- A. Save the prompt
- B. Share the prompt
- C. Schedule the prompt
- D. Rename the chat
Answer
C. Scheduling runs a prompt automatically on a cadence.
17. Where do you discover, save, and share prompts?
- A. The Copilot Prompt Gallery
- B. The Recycle Bin
- C. Purview
- D. The SharePoint admin center
Answer
A. The Prompt Gallery.
18. A teammate runs a prompt you shared. What do they see?
- A. Your data and results
- B. Only content they personally have access to
- C. Nothing
- D. All company data
Answer
B. Sharing shares the prompt, not the data; results are permission-trimmed.
19. To keep developing a topic across sessions, a user should:
- A. Add the conversation to a notebook
- B. Delete the chat
- C. Start fresh each time
- D. Rename and hope
Answer
A. Notebooks make a conversation durable workspace content.
20. Which pairs tool to purpose correctly?
- A. Notebook = collaboration; Page = personal
- B. Page = collaboration; Notebook = personal workspace
- C. Both are meeting-only
- D. Both equal a chat
Answer
B. Pages collaborate; notebooks are personal.
21. A specific, repeatable, knowledge-scoped need with no existing agent calls for:
- A. Re-explaining context in chat each time
- B. Building your own agent (no code), with knowledge and instructions, then sharing it
- C. Emailing a document
- D. Fine-tuning a model
Answer
B. Build and share a purpose-built agent.
22. Which agent setting defines its grounding sources?
- A. Suggested prompts
- B. Capabilities
- C. Knowledge
- D. The name
Answer
C. Knowledge = grounding sources.
23. When should you use the Agent Store instead of building?
- A. Never
- B. When an existing agent already meets the need
- C. Only if you can code
- D. Only for personal use
Answer
B. Reuse an existing agent when it fits.
24. A contractor without access to an agent's SharePoint knowledge uses the shared agent. Result?
- A. It surfaces the restricted content anyway
- B. It won't surface content they lack permission to see
- C. It grants them access
- D. It stops working for all
Answer
B. Agents honor each user's permissions.
25. Building a Microsoft 365 Copilot agent requires:
- A. Writing code
- B. No code — the agent builder is no-code
- C. A data-science team
- D. Fine-tuning
Answer
B. The agent builder is no-code.
26. The best next step when a response is close but too long and formal:
- A. Start a brand-new chat
- B. Iterate: ask to shorten and warm the tone
- C. Accept it
- D. Report it broken
Answer
B. Iterate within the conversation to keep context.
27. Renaming a chat primarily helps you:
- A. Delete it faster
- B. Find it again later
- C. Share your password
- D. Change permissions
Answer
B. A meaningful title aids findability.
Domain 3 — Draft & analyze content¶
28. To build a deck from an existing Word proposal:
- A. Retype it in PowerPoint
- B. Use Copilot in PowerPoint to generate the deck from the Word file
- C. Use Excel
- D. Fine-tune a model
Answer
B. Generate a document from an existing document.
29. What makes a summary a management summary?
- A. Exactly one page
- B. Executive audience + concise, decision-focused format
- C. More tokens
- D. No data
Answer
B. Defined by audience and brevity.
30. "Move data and insights between apps" means:
- A. Copying raw cells unchanged
- B. Copilot re-expresses content for the destination (Excel insight → Word paragraph → email)
- C. Exporting to PDF
- D. Deleting data
Answer
B. Content is re-expressed for the target app.
31. A late joiner wants a meeting catch-up. What's required?
- A. Nothing
- B. The meeting must be transcribed or recorded
- C. Admin rights
- D. Fine-tuning
Answer
B. Meeting grounding needs transcription/recording.
32. To collaborate on a Copilot answer as shared, editable content, use:
- A. A Copilot Page
- B. A personal notebook
- C. A deleted chat
- D. A scheduled prompt
Answer
A. Pages are the collaborative canvas.
33. The difference between memory and instructions:
- A. They're identical
- B. Memory remembers context; instructions are standing preferences you set
- C. Memory bypasses permissions
- D. Both are developer-only
Answer
B. Memory personalizes; instructions guide; neither bypasses security.
34. The responsible final step before sending a Copilot-drafted external proposal:
- A. Send immediately
- B. Review and verify facts, tone, and completeness
- C. Delete it
- D. Add tokens
Answer
B. Copilot drafts; humans verify.
35. Which app's Copilot best summarizes a long email thread and drafts a reply?
- A. Excel
- B. Outlook
- C. PowerPoint
- D. OneNote
Answer
B. Outlook.
36. Which app's Copilot analyzes spreadsheet data and suggests formulas?
- A. Word
- B. Excel
- C. Teams
- D. Outlook
Answer
B. Excel.
37. Generating "from scratch" when a relevant file exists is a pitfall because:
- A. It's faster
- B. It wastes Copilot's grounding advantage and lowers relevance
- C. It uses fewer tokens
- D. It's more secure
Answer
B. Grounding in the source produces a better, faithful draft.
38. Copilot Chat is available on:
- A. Desktop only
- B. Web and mobile
- C. Paper
- D. Fax
Answer
B. Web and mobile experiences exist.
39. Grounding in work data (not just the web) depends on:
- A. The weather
- B. Having the appropriate Microsoft 365 Copilot license
- C. Deleting chats
- D. Using Excel
Answer
B. Work-data grounding requires the license.
40. The single best habit across all Copilot use is:
- A. Trust every answer
- B. Treat output as a draft and verify high-stakes content
- C. Avoid context
- D. Never iterate
Answer
B. Verify — it underpins responsible, effective use.
\u2705 Ready check: if you can explain why each wrong option is wrong, you understand the material, not just the answer. Revisit any chapter where you missed two or more questions in its domain.