AI Foundations · Module 03 April 2026
Prompt.·Choose.·Verify.

AI tools of the trade — how to talk to them, how to pick between them, and how to get reliable output from each.

What we'll cover 02 / 17
Agenda

From prompting, to choosing the right tool, to using it safely.

01

The art of prompting

Role, task, format, constraints. Zero-shot vs. few-shot. Why specificity wins.

02

The tool ecosystem

What's out there, what's approved, and a framework for choosing.

03

Four tools up close

Gemini · NotebookLM · ChatGPT · Claude. When to use which.

Part One 03 / 17
01
Part One · Prompting

How you ask changes the answer.

Prompting isn't typing — it's briefing. Treat the model like a capable freelancer who just walked into the room and knows nothing about your context.

Prompt patterns 04 / 17
Four levers to pull

Every good prompt sets at least one of these.

01 · ROLE

Who is the AI?

"You are an expert chef" · "Act as a senior Java reviewer" · "Respond as a UX writer."

02 · TASK

What should it do?

"Summarize in 3 bullets" · "Rewrite for tone" · "List the counter-arguments."

03 · FORMAT

How should it answer?

"Explain like I'm 10" · "Return JSON with keys X, Y, Z" · "Use a table."

04 · CONSTRAINTS

What must it avoid?

"Under 200 words" · "No bullet lists" · "Avoid the word quantum."

The most common mistake: being too vague. "Write something about dogs" is not a brief — it's a wish.

Zero-shot vs. few-shot 05 / 17
When to show examples

Show, don't just tell — especially when the task is nuanced.

Zero-shot

No examples. Just the instruction.

Translate this to French: Hello, how are you?

Works well for: simple, common tasks where the model already knows the pattern.

Few-shot

Instruction plus a few examples.

Write a product review: - Wireless headphones → Great sound, comfortable fit. - Coffee maker → Easy to use, makes excellent coffee. - Smartphone case → ?

Works well for: specific tone, unusual format, domain-specific patterns.

Vague vs. specific 06 / 17
The same task, rewritten

Specificity costs you ten seconds — and saves you three rewrites.

✕ Suboptimal

Write about dogs.

✓ Better

Write a 200-word blog post about training puppies for first-time owners. Warm, encouraging tone.

✕ Suboptimal

Fix this code.

✓ Better

Debug this Python function that should calculate Fibonacci numbers but returns incorrect values for n > 10: [code]

Security first: never paste client data, passwords, keys, or proprietary code into public AI tools without explicit IT Security approval.

Part Two 07 / 17
02
Part Two · The ecosystem

Many tools. Few decisions.

You don't need to master everything. You need to know which tool fits which job — and which ones are approved for which data.

The landscape 08 / 17
General-purpose · research · specialized

The tools most people will meet.

GENERAL · LLM

ChatGPT · GPT-5

OpenAI · creative, reasoning, coding, agent mode.

GENERAL · LLM

Claude 4

Anthropic · up to 1M context, strong coding, Computer Use.

GENERAL · LLM

Gemini 2.5

Google · multimodal, real-time web, Workspace integration.

RESEARCH

NotebookLM

Google · source-grounded, multi-doc synthesis, citations.

RESEARCH

Perplexity

Search-native · cited answers, deep-research mode.

OPEN · LLM

DeepSeek · Llama 4

Open-source friendly · competitive on technical tasks.

SPECIALIZED

Midjourney · DALL·E · Flux

Image generation — artistic, prompt-literal, photorealistic.

SPECIALIZED

GitHub Copilot · Cursor · ElevenLabs

Code completion · AI-native IDE · voice synthesis.

Green = general-purpose assistants you'll use daily. Amber = tools you'll reach for when a task calls for them.

What goes where 09 / 17
Data classification for AI tools

One table. Memorize it.

Data type Consumer AI (ChatGPT, Gemini free) Enterprise tier (Business, Team, Workspace)
Public informationAllowedAllowed
Personal non-sensitiveStrip identifying detailsAllowed
Work / client dataNeverWith employer's approval
Health, finance, biometricsNeverWith explicit consent
Passwords · API keys · credentialsNeverNever

Enterprise tiers typically exclude your data from training and add compliance controls. Consumer tiers may not. Check the settings.

Part Three 10 / 17
03
Part Three · Four tools up close

Gemini · NotebookLM · ChatGPT · Claude.

What each does well, where it fits, and when to reach for it instead of the others.

Tool · Gemini 11 / 17
Tool 01 · Google

Gemini

A multimodal research assistant with a live feed of the web.

general-purpose check retention settings
What it does well
Real-time web access · multimodal (text, images, docs) · native Workspace integration · built-in citations.
Where to reach for it
Current events and regulation research · analyzing Google Drive files directly · image analysis for business uses · Deep Research for multi-hour synthesis.
Watch out for
Workspace / Business accounts — chats may be retained as company property. No temporary mode in all tiers. Treat as if everything is logged.
Tool · NotebookLM 12 / 17
Tool 02 · Google

NotebookLM

A research partner that reads your sources and cites them back.

research-grade
What it does well
Source-grounded answers — minimal hallucinations when working from your uploaded documents · multi-document synthesis · automatic citations · web Discover to find adjacent sources.
Where to reach for it
Academic research across many papers · meeting notes and project documentation synthesis · anywhere you need exact citations and traceable claims.
Signature feature
Output formats beyond chat: mind maps, audio podcast summaries, video walkthroughs — useful for handing off research to teammates.
Tool · ChatGPT 13 / 17
Tool 03 · OpenAI

ChatGPT

A versatile creative partner — and a coding collaborator.

public tool · no client data
What it does well
Creative writing with tone and audience control · strong reasoning and explanation · code generation and debugging · custom GPTs that bundle instructions and knowledge.
Where to reach for it
Drafting marketing copy, emails, and presentations · writing and explaining code · brainstorming and strategic thinking · image analysis and creative Q&A.
Watch out for
Default: your data trains the model. Turn this off under Settings → Data Controls if you haven't already. Never paste client, proprietary, or sensitive company data.
Tool · Claude 14 / 17
Tool 04 · Anthropic

Claude

A thoughtful assistant with an exceptional memory — and a careful nature.

general-purpose up to 1M context
What it does well
Very long documents (up to 1M tokens — a small book) · careful reasoning and nuanced writing · strong coding and code review · Computer Use for early agentic tasks · Projects for persistent, scoped context.
Where to reach for it
Analyzing a full contract, thesis, or codebase in one pass · long-form writing and editing · anywhere you want the model to say "I don't know" instead of making something up.
Signature feature
Strong default refusal behavior on sensitive topics — which is a feature for consequential work and a friction for creative exploration. Know which side you're on before you start.
The one-line decision 15 / 17
A shortcut for picking the right tool
Geminiwhen you need current information from the web.
NotebookLMwhen you need cited answers from your own documents.
ChatGPTwhen you need creative writing or coding — with public data only.
Claudewhen the document is long or the reasoning has to be careful.
Where AI already is 16 / 17
You likely interact with 20+ AI systems daily

AI in the applications you already use.

Productivity

Smart compose · priority inbox · meeting scheduling · auto-transcription · smart summaries.

Learning

Duolingo lessons · Khan Academy tutors · Grammarly style and grammar suggestions.

Healthcare

Skin-cancer detection apps · therapy chatbots · personalized workout programs.

Finance

Fraud detection · robo-advisors · expense categorization · automated claims triage.

Entertainment

Netflix · Spotify recommendations · adaptive game difficulty · content curation.

Work

Resume screening · sentiment analysis · ticket routing · copywriting and design assistance.

End of Module 03 17 / 17
// TAKE HOME
Be specific.
Pick the right tool.
Verify what comes back.

AI is a capability multiplier — but only if you brief it like a professional, choose a tool that fits the data, and stay the final authority on what ships.

Next up · Module 04 — agents, synthetic media, your work, your rights. The same tools, at society scale.