AI-901 Cheat Sheet 2026: Quick Reference
Both domains condensed to a table, plus the Content Understanding vs. Document Intelligence call and the Cognitive Services OpenAI role distinction the exam tests directly.
Exam Snapshot
We built this from Microsoft's published AI-901 skills outline, not a generic template — every reference link below points to the current Microsoft Learn page, verified before this went live.
1. AI Concepts & Capabilities (40-45%)
| Concept | Area | Quick note |
|---|---|---|
| Fairness | Responsible AI principle | Equitable treatment across groups |
| Reliability & safety | Responsible AI principle | Consistent, safe operation under real conditions |
| Privacy & security | Responsible AI principle | Protecting data used by and generated from the model |
| Inclusiveness | Responsible AI principle | Accessible to people with diverse needs and backgrounds |
| Transparency | Responsible AI principle | Behavior and limitations are explainable to users |
| Accountability | Responsible AI principle | Clear human ownership and oversight of the system |
| How generative models work | AI model components | Foundational mechanics, not implementation detail |
| Model selection by capability | AI model components | Matching a model to what the task actually needs |
| Deployment options & config params | AI model components | Choosing settings at deployment time |
| Generative & agentic AI workloads | AI workloads | Identifying the right workload category for a scenario |
| Text analysis techniques | AI workloads | Keyword extraction, entity detection, sentiment, summarization |
| Speech recognition & synthesis | AI workloads | Speech-to-text and text-to-speech feature awareness |
| Computer vision & image generation | AI workloads | Vision analysis and generative image features |
| Information extraction | AI workloads | Pulling structured data from text, images, audio, video |
2. Implement AI Solutions With Microsoft Foundry (55-60%)
| Concept | Area | Quick note |
|---|---|---|
| System & user prompts | Generative AI apps | Design effective prompts for a generative model |
| Deploy a model in the Foundry portal | Generative AI apps | Portal-driven deployment workflow |
| Lightweight chat client | Generative AI apps | Built using the Foundry SDK |
| Single-agent solution | Generative AI apps | Created and tested in the Foundry portal |
| Lightweight client for an agent | Generative AI apps | A small app that talks to an existing agent |
| Text analysis app | Text & speech | A lightweight app that includes text analysis |
| Spoken prompts via multimodal model | Text & speech | Responding to speech input directly |
| Azure Speech in Foundry Tools | Text & speech | Building an app with the speech service |
| Visual input via multimodal model | Computer vision | Interpreting images passed into a prompt |
| Visual outputs via generative models | Computer vision | Creating new images, not just analyzing them |
| Vision-capability app | Computer vision | A lightweight app built around vision features |
| Content Understanding (documents/forms) | Information extraction | Schema described in plain English |
| Content Understanding (images) | Information extraction | Same analyzer approach, image source |
| Content Understanding (audio/video) | Information extraction | Same analyzer approach, audio/video source |
| Info-extraction app | Information extraction | A lightweight app using Content Understanding |
Content Understanding vs. Document Intelligence
A single most-tested Domain 2 distinction:
| You need to… | Use this |
|---|---|
| Extract data using a schema described in plain English, from documents, images, audio, or video | Content Understanding |
| Extract data from a deterministic, standard form (invoice, receipt, ID) | Document Intelligence prebuilt models |
| Handle a multimodal source combining several content types at once | Content Understanding |
| Extract data from a well-known, fixed-layout structured form | Document Intelligence |
Cognitive Services OpenAI Role Quick Reference
Least privilege first — the exam tests this distinction even at Fundamentals level:
| Role | Grants |
|---|---|
| Cognitive Services OpenAI User | Inference only — least privilege |
| Cognitive Services OpenAI Contributor | Inference plus deployment management |
| Owner | All of the above, plus role-assignment rights |
Common Mistake: Reaching for Document Intelligence First
A frequent wrong answer on Domain 2 questions is picking Document Intelligence for a scenario that actually calls for Content Understanding. The reflex makes sense — Document Intelligence has been around longer and "document extraction" sounds like its job. But the decision table above draws the real line: Document Intelligence prebuilts are for deterministic, fixed-layout forms like invoices, receipts, and IDs, where the structure never changes. Content Understanding is the answer whenever the source is multimodal (documents, images, audio, or video mixed together) or the extraction schema is described in plain English rather than a rigid template. If a question mentions extracting fields from a mix of scanned contracts and audio call transcripts using a schema you define in natural language, that's Content Understanding, not Document Intelligence — even though both technically "extract data."
Frequently Asked: AI-901
How is this cheat sheet different from the AI-901 study guide?
The study guide explains the exam's most-confused topics in narrative form with a study plan. This page strips the narrative for tables you can scan in the last hour before the exam.
What's the real difference between Content Understanding and Document Intelligence?
Content Understanding is the plain-English, multimodal extractor. Document Intelligence prebuilts are for deterministic, fixed-layout forms like invoices and IDs.
What's the fastest way to review right before the exam?
Scan the two domain tables plus the Content Understanding and OpenAI role tables — those two distinctions account for a large share of missed Domain 2 questions.
Can I print this AI-901 cheat sheet?
Yes — every table here is plain HTML, so a browser print or "print to PDF" renders cleanly.
MSCertQuiz sells practice-exam access for AI-901 and other Microsoft certifications; this cheat sheet is written by the same team that builds those questions.
Where to Go Next
The study plan and narrative behind this quick reference.
15 free questions with detailed explanations.
How this exam differs from its predecessor.
Full exam details and the complete practice bank.
Reviewed the cheat sheet? Now drill it.
Start with free AI-901 questions covering both domains above.
Start Free AI-901 Practice