AI Research Agents Cheat Sheet

Microsoft Applied Skills cheat sheet for generating reports with AI research agents, sources, drafting, refinement, and final review.

Use this cheat sheet for Microsoft Applied Skills: Generate reports with AI research agents as a lab-readiness checklist. The assessment evaluates whether you can use the Researcher agent to define report scope, identify sources, create an initial draft, refine the report, and finalize it responsibly.

Research report sequence

    flowchart TD
	  S["Business question"] --> C["Define scope, audience, and decision"]
	  C --> R["Identify source material"]
	  R --> D["Create initial report draft"]
	  D --> F["Refine gaps, weak claims, and structure"]
	  F --> V["Finalize with fact, sensitivity, and policy checks"]

If a lab answer produces a polished report but cannot explain its sources, assumptions, or review path, it is incomplete.

What to know cold

Lab area Decision rule Reject when
Scope Define purpose, audience, business question, output format, and constraints. The report request is vague and the agent guesses the goal.
Sources Identify relevant source material and keep claims traceable. The agent invents facts or uses unknown sources.
Initial draft Ask for structure, sections, tone, length, and evidence expectations. The first draft is accepted as final.
Refinement Ask targeted follow-ups for gaps, weak evidence, unclear logic, and better organization. The user re-prompts vaguely with “make it better.”
Finalization Check facts, sensitive data, policy, audience fit, and actionability. The report is sent without human review.
Responsible use Protect confidential data and document assumptions. Sensitive context is pasted without policy or access review.

Lab-task eliminations

Stem clue Eliminate first Keep in play
report purpose is unclear ask Researcher for “a report” define decision, reader, scope, and output
source quality matters use generic model knowledge only approved sources and source-aware synthesis
initial draft has gaps regenerate randomly ask targeted follow-up questions
claim is important trust fluent text verify source, date, assumption, and relevance
executive audience long raw research dump concise structure, decisions, risks, and recommendations
sensitive business data paste everything minimize data and use approved Microsoft 365 context
final report send immediately review facts, tone, sensitive content, and policy

Prompt patterns

Need Strong prompt shape
define scope “Create a report for [audience] to decide [decision], using [sources], covering [sections], excluding [out of scope].”
draft report “Use these sources, cite or identify where claims come from, and organize by problem, evidence, options, recommendation, risks.”
refine gaps “Find unsupported claims, missing stakeholder concerns, weak assumptions, and unclear recommendations.”
shorten “Condense for [audience] while preserving decisions, evidence, risks, and next steps.”
finalize “Check for factual uncertainty, sensitive data, policy issues, tone, and actionability.”

Final 15-minute review

If the lab asks for… Start with
“define scope” audience, business decision, source boundaries, output structure
“identify sources” approved source material, relevance, freshness, and traceability
“create draft” report sections, tone, source-backed claims, and assumptions
“refine report” gaps, unsupported claims, structure, risks, and recommendations
“finalize” facts, sensitive data, policy, audience fit, and human approval
Revised on Monday, June 15, 2026