Skip to content
Teachers With AI

Reference

AI Terms Every Teacher Should Know

22 terms that come up in staff meetings, procurement forms and tool reviews — defined in plain English, without varnish.

Last updated Independently written · affiliate links disclosed

BasicsClassroomPrivacy and lawRisk

Basics

Large language model (LLM)
Software trained on huge amounts of text to predict the next word. It produces fluent text without knowing whether the text is true. ChatGPT, Claude and Gemini are examples.
Prompt
The instruction you type. A prompt that names the grade, the standard, the length and the output format gets a far more usable result than "write a lesson plan".
Token
A chunk of text a model processes — roughly three quarters of a word. Usage limits on free plans are usually counted in tokens or characters, which is why long documents burn a quota fast.
Context window
How much text a model can consider at once. A bigger window lets it read a full unit plan; it does not make the model more accurate.
Hallucination
A confident statement that is wrong — most often a made-up standards code, citation or quote. Treat any specific code or citation from an AI tool as unverified until you check it.
Temperature
A setting controlling how random the output is. Low temperature gives predictable, repetitive text; high gives variety and more errors. Most classroom tools set it for you.
Model
The specific trained system behind a product. Two tools can look different while running the same model, which is why a tool's interface matters as much as its model.

Classroom

AI detector
A tool that guesses whether text was written by a model. It produces a probability, not proof, and false positives fall hardest on second-language writers. Never use a detector score alone as evidence.
Differentiation
Adapting one task to different learners — reading level, scaffolding, extension. AI is good at producing variants and bad at knowing which variant a specific child actually needs.
Rubric
A grid of criteria against named performance levels. A rubric with observable descriptors makes AI feedback far more consistent, because the model has real standards to apply.
Formative assessment
Checks during learning rather than at the end — exit tickets, quick polls, thumbs up. AI speeds up writing them; it cannot judge whether a class actually understood.
Scaffold
Support that lets a student attempt work just beyond what they can do alone. AI scaffolding helps when it is targeted and hurts when it replaces the thinking.
Exit ticket
A two-minute end-of-lesson check. Easy to generate, easy to over-generate — three good questions beat ten rushed ones.
Retrieval practice
Pulling information from memory rather than rereading it. AI-written quizzes are useful here because retrieval is far more effective when questions are varied.

Privacy and law

FERPA
The US federal law protecting education records. It governs what a school may share with a vendor; it does not certify that a product is safe. Check the vendor's data processing agreement.
COPPA
US law covering data from children under 13. Most classroom AI tools either require a school-managed account or exclude under-13 users entirely.
GDPR
EU and UK data protection rules. They give readers and users rights over their data and require a lawful basis for processing — including analytics cookies.
PII
Personally identifiable information: names, student IDs, anything that points at one child. Core planning tools usually work without it; pasting it in is a choice, and a risky one.
Data processing agreement (DPA)
The contract between a school and a vendor describing what data is collected and how it is handled. A vendor claiming FERPA compliance without a DPA is only making a statement.

Risk

Prompt injection
Text hidden in a document or page that tells an AI tool to ignore its instructions. Relevant when a tool reads student work or a website containing untrusted text.
Bias
Systematic skew in output, usually inherited from training data. It shows up in example names, cultural references and how confidently a tool judges writing styles it has seen less often.
Over-reliance
The failure mode that matters most in classrooms: the tool produces something plausible, and nobody checks it. Every page on this site repeats the same instruction — read the output before a student does.

Where to go next

These terms appear throughout our reviews. If you want the practical version, start with Best AI tools for teachers (51 tools, dated pricing) or the testing methodology that explains how we use these concepts when scoring a tool.