Research · AI models
You can't choose if you can't see.
We reviewed around 50 tools used in schools. Most did not name the model behind the student feature in a current source.
Why it matters
The model inside a tool shapes what students learn.
Students meet AI in different ways
A chat box, a tutor, feedback on an essay, or a feature they don't know is AI. Each one shapes how much thinking students do.
Some tools label their AI. Others build it in quietly, so students and teachers may not know it's there.
Models change often
AI companies release new models every few months and retire old ones. When the model changes, the tool changes, often without notice.
A tool you approved in the fall may run on a different model by spring. If the tool doesn't name its model, you can't tell.
The models under the hood differ
Most tools run on a model from ChatGPT (OpenAI), Claude (Anthropic) or Gemini (Google). These differ in accuracy, tone and safety, so two tools with the same features can teach very differently.
In Common Sense Media's 2025 risk ratings for young users, ChatGPT was rated moderate risk, Claude minimal risk, and Gemini's kids and teen products high risk.
Schools shouldn't choose blind
You can't compare, approve or explain a tool you can't see inside.
Districts are left to vet AI tools themselves. New York City already requires vendors to explain what their AI tools do.
What we found
8 in 10 tools don't name the model behind the student feature.
The label is missing because buyers rarely asked for it. Schools are starting to ask now.
How we research
Set the rules, then freeze the sample.
Apply each rule to every tool.
Keep dated evidence. Log every judgment call.
State our ties. Say what we don't claim.
Put it to work.
For schools
Four questions for every edtech company
Where do students meet AI in your tool, and does it give answers or ask questions?
Which model runs each feature, and from which company?
How will you tell us before the model changes?
Can we turn AI features off, or choose the model?
For edtech companies
Four questions to ask yourself
Do students know when they're using AI, and does it push them to think?
Can a school find which model runs each feature on your site?
Do schools hear from you before the model changes?
Can a school turn AI features off, or choose the model?