Understand AI.Use it with judgment.
AI literacy is more than knowing how to write a prompt. It means understanding what different systems can do, where they fail, what data they touch, and when a human should slow down, verify, or say no.
Not every AI product is doing the same kind of work.
These categories overlap, but the distinctions help you choose the right tool, set safer expectations, and explain AI to learners without magical thinking.
Language models
The underlying prediction systems that process and generate language—and increasingly images, audio, code, and files.
- Think of it as
- The engine, not necessarily the whole product.
- Watch for
- Confident errors, stale knowledge, hidden assumptions.
Chatbots & copilots
Conversational products built around one or more models, often with search, file tools, memory, or organization data.
- Best for
- Explaining, brainstorming, drafting, questioning, practice.
- Watch for
- Learners outsourcing the thinking instead of extending it.
AI agents
Systems given a goal, instructions, context, tools, and some ability to choose and execute multiple steps.
- Best for
- Research workflows, repetitive tasks, coordinated actions.
- Watch for
- Permissions, irreversible actions, tool errors, weak oversight.
Media generators
Models and creative tools that generate or edit images, video, audio, music, voices, and design assets.
- Best for
- Visual explanation, prototypes, storyboards, creative iteration.
- Watch for
- Consent, likeness, copyright, misleading realism, provenance.
AI robots
Physical systems combining sensors, models, control software, and hardware to perceive, decide, move, and manipulate.
- Today
- Mostly industrial, research, logistics, and emerging pilots.
- Watch for
- Physical safety, reliability, supervision, and accountability.
Choose for the task and context—not the loudest benchmark.
ChatGPT, Claude, Gemini, and Copilot are products and ecosystems. Each can route work through different models and tools. Model names and availability change quickly, so teach durable selection habits.
General-purpose multimodal chat, study workflows, creation, coding, research, and agent tools. Useful when one workspace must handle several media and task types.
Study Mode ↗Conversational and document-oriented workflows with education offerings that emphasize guided learning and critical thinking rather than answer delivery.
Education ↗Multimodal AI connected to Google’s education and productivity ecosystem, with tools spanning chat, research, images, video, and classroom workflows.
Education tools ↗AI experiences across Microsoft 365 and education, including document-grounded learning and low-code agent creation through Copilot Studio.
Study & Learn ↗Downloadable or self-hostable model families can offer control and customization, but require technical evaluation, infrastructure, security, and maintenance.
Explore models ↗Prompts that preserve thinking instead of replacing it.
Replace the text in brackets. Never paste private student information into a tool unless your organization has explicitly approved that product and data use.
Who should the AI act like?
What useful outcome is needed?
What should it understand first?
What boundaries must it follow?
What shape should the response take?
How should uncertainty be flagged?
Build a thinking-first lesson
You are an experienced [grade/subject] instructional designer. Help me plan a [duration] lesson about [topic] for learners who already know [prior knowledge]. The goal is [learning outcome]. Begin with a short diagnostic question, then provide: a clear explanation, two contrasting examples, one low-cost hands-on activity, three checks for understanding, and a closing retrieval prompt. Do not simply give students answers. Flag likely misconceptions and any factual claims I should verify.
Explain in layers
Teach [concept] in three layers: (1) a plain-language explanation for a learner around age [age], (2) a concrete analogy connected to [interest or experience], and (3) a more precise explanation using the correct academic vocabulary. After each layer, ask one question and wait for the learner’s response before continuing. If the analogy breaks down, explain exactly where.
Run a low-pressure retrieval session
Act as a patient practice coach for [topic]. Ask one question at a time, mixing recall, explanation, and application. Do not reveal the answer immediately. If the learner is unsure, offer one small hint, then a second hint using a different representation. After the learner answers, explain what was strong and what to revisit. Keep a short list of ideas to review again at the end.
Give revision-focused feedback
Review the de-identified student work below using this rubric: [rubric]. Do not rewrite the work. Identify one strength, one high-leverage revision, and two questions that help the learner make the revision independently. Quote only short phrases from the work. Separate observations from inferences, and tell me where the rubric or evidence is too ambiguous for a confident judgment. [PASTE DE-IDENTIFIED WORK]
Differentiate without lowering the goal
The learning goal is [goal]. Suggest three pathways to the same goal for learners who may need: (A) language support, (B) smaller chunks and reduced working-memory load, and (C) additional challenge. Preserve the core thinking in every pathway. For each, recommend instructions, materials, response options, and a quick check for understanding. Avoid diagnosing learners or making assumptions about disability.
Turn a topic into a family conversation
Help me discuss [topic] with a child around age [age] during [dinner/a car ride/a walk]. Give me a 60-second setup, five open questions that move from observation to reasoning, and one simple activity using common household items. Include a respectful way to say “I don’t know—let’s investigate” and two reliable source types we could consult together.
Evaluate a learning source together
Help us evaluate this source about [topic]: [URL or de-identified excerpt]. Use age-appropriate media-literacy questions: Who made it? What is the main claim? What evidence is shown? What might be missing? What emotions does it try to create? What should we verify elsewhere? Do not assume the source is true or false. End with a two-source verification plan.
Audit an AI answer
Audit the answer below. Create a table with: claim, claim type (fact/inference/opinion), confidence, evidence needed, and a recommended primary source. Identify invented citations, unsupported numbers, outdated statements, and places where wording sounds more certain than the evidence allows. Do not “fix” a claim unless you can explain how it should be verified. [PASTE AI ANSWER]
Design an instructional visual
Create a [16:9 or square] educational illustration of [concept] for learners around age [age]. The visual must make [relationship/process/parts] easy to understand. Use [style and palette], accurate proportions where relevant, strong contrast, and clear negative space for labels. Do not include decorative elements that could be mistaken for data. Avoid copyrighted characters, real children, logos, and text unless specified.
Storyboard a short learning video
Storyboard a [30/60/90]-second learning video explaining [concept]. For each shot provide duration, framing, visible action, narration, on-screen text, and the learning purpose. Use one consistent visual style and avoid rapid cuts or distracting motion. Include a final pause with a question the learner must answer. Flag any scene that could create a factual or safety misconception.
Draft an assignment AI-use policy
Draft a plain-language AI-use policy for this assignment: [assignment]. Divide use into: encouraged, allowed with disclosure, requires permission, and not allowed. Explain the learning reason for each boundary. Include a short disclosure template students can complete, examples of acceptable and unacceptable assistance, and how questions or mistakes will be handled without assuming misconduct.
Create a family AI agreement
Help our family create a one-page AI agreement for a learner age [age]. Cover: what tools may be used, information that must stay private, when an adult should be involved, how to verify important answers, rules for images and other people’s likenesses, disclosure for schoolwork, and what to do after an upsetting or unsafe result. Use collaborative language and include three questions the child should help decide.
No prompts match that search yet. Try a broader word or choose “All.”
Popular examples by capability—not endorsements or permanent rankings.
Product names, models, prices, age requirements, and availability change quickly. Follow the official reference links before adopting a tool.
Chatbots and learning copilots
Conversation-centered products for explanation, drafting, research, tutoring, and working with files.
Agent builders and agent platforms
Systems that combine models with instructions, knowledge, tools, memory, triggers, and actions.
Image generators and editors
Tools for creating, transforming, compositing, and iterating visual assets from prompts and references.
Video generators and creative video AI
Tools that produce or modify short clips from text, images, references, and existing footage.
Robotics and physical AI
Emerging systems that connect language and vision models to simulated or physical movement.
Robotics claims are often demonstrations, research results, pilots, or vendor roadmaps—not evidence of general classroom readiness. Treat physical safety and supervised deployment as separate questions from conversational AI safety.
A short safety checklist before a prompt becomes practice.
Rules should match the age, task, data, setting, and consequences—not rely on a single universal “AI allowed” switch.
Remove names and identifiers. Never enter student records, health information, passwords, payment data, or private communications into an unapproved service.
Use primary sources for consequential claims. Check quotations, dates, numbers, citations, and generated visuals before teaching or publishing.
Decide which thinking belongs to the learner. Use AI for feedback, examples, questions, and scaffolds without quietly completing the core work.
Explain when AI shaped a lesson, image, assessment, or submission. A simple process note supports trust and better evaluation.
Look for stereotypes, missing perspectives, inaccessible formats, uneven performance across languages, and assumptions about families or learners.
AI can recommend or draft. A person should own decisions involving grades, discipline, placement, safety, privacy, or high-stakes communication.
Durable answers in a fast-moving field
What is the difference between a chatbot and an AI agent?
A chatbot mainly responds within a conversation. An agent is configured to pursue a goal using instructions, context, memory, tools, and actions, often across multiple steps. Because agents can act, they require stronger permission boundaries, testing, and review.
Which AI model is best for teachers?
There is no single best model for every teaching task. Compare the learning goal, approved data environment, source and citation needs, file support, accessibility, cost, and the amount of human review required.
What information should never be entered into a public AI tool?
Do not enter student records, private health or disability information, passwords, authentication codes, financial data, or identifiable work unless your organization has explicitly approved the tool and its data practices for that use.
How often should an AI literacy guide be reviewed?
Review vendor-specific guidance at least each term and before a major adoption. The durable principles—privacy, verification, disclosure, bias awareness, learning purpose, and human accountability—should remain stable even when products change.
Turn AI literacy into better learning—not just more output.
Use EduGPT to create reviewable lessons, practice, tutor experiences, and learning paths while educators and parents stay in control.
