It’s the question that surfaces in almost every school leadership meeting about AI, usually right after someone mentions that a teacher has been using ChatGPT to draft IEP language or a counselor has been summarizing student notes in Gemini.
The instinct is right. But the question itself is too narrow. Price isn’t what determines whether student and staff data ends up in a model’s training set; the account tier and the contract behind it do.
A “free” tool and a “paid” tool from the same company can have opposite data policies, and two free tools from different companies can behave in entirely different ways.
For school leaders, the real question is broader: what agreement governs this tool, and does it treat our students as customers or as products?
Key takeaways
- Price isn’t the signal – the contract is. A free tool and a paid tool from the same vendor can have opposite data policies; what matters is the account tier and the legal agreement behind it, not the price tag.
- Consumer accounts default to training. OpenAI’s Free, Plus, and (in some configurations) Pro tiers, and Anthropic’s consumer Claude Free/Pro/Max tiers, use conversations to improve future models unless a user manually opts out — and that opt-out only protects conversations sent after the setting is changed.
- Education-contracted tiers are built differently. Google Workspace for Education’s Gemini, OpenAI’s ChatGPT for Teachers, and Anthropic’s Claude for Teachers all state that data is not used for model training by default, under FERPA-aligned terms or a Cloud Data Processing Addendum — a contractual protection, not a settings toggle.
- “FERPA-aligned” isn’t “FERPA-compliant.” Vendors write their terms to support FERPA requirements, but whether a specific district’s implementation actually satisfies FERPA is a determination each school still has to confirm with its own counsel.
- Adoption is outpacing governance. Roughly 83% of K-12 teachers already use generative AI and about 30% of students use it daily, while only a small fraction of districts have formal AI policies the exact gap where personal, training-by-default accounts get used for grading, IEPs, or behavioral notes.
The default, on consumer accounts, is yes
Across the major consumer AI platforms, personal accounts are generally opted in to training by default. OpenAI’s consumer tiers Free, Plus, and even the paid Pro plan in some configurations use conversations to improve future models unless a user manually disables “Improve the model for everyone” in settings, and that opt-out only protects conversations sent after the toggle is flipped.
Anthropic’s consumer Claude plans have followed a similar pattern, grouping free and Pro accounts together under default training use.
The common thread: on personal, self-serve accounts, “free” and even “paid individual” don’t mean private. Privacy is something you configure, not something you get automatically.
Education tiers are built differently, and that’s the distinction that matters
This is where the picture changes for schools specifically. The major vendors now offer education- or enterprise-contracted versions of their tools that operate under a different legal basis entirely.
Google is explicit that Gemini inside Google Workspace for Education is a core, contracted service: chats and files aren’t reviewed by humans or used to train models outside the institution’s domain, and the commitment is written into the Cloud Data Processing Addendum rather than left to a settings toggle.
OpenAI’s ChatGPT for Teachers and Anthropic’s newer Claude for Teachers program follow the same logic, pairing free access for verified educators with FERPA-aligned data terms rather than the consumer default.
In other words, the protective feature isn’t the price; it’s whether the tool was issued through a signed institutional agreement or downloaded as a personal app.
Why this gap is the real risk for K-12
The scale of adoption makes this more than a theoretical concern. Recent research puts generative AI use among K-12 teachers at roughly 83%, with about 30% of students using AI tools daily and student use up sharply year over year.
Yet governance hasn’t kept pace: surveys of school AI policy find only a small fraction of districts have formal AI usage guidelines, and a majority of educators say their districts haven’t given clear direction at all. That gap is exactly where consumer-grade, training-by-default tools slip into daily use: a teacher using a personal ChatGPT account for grading, or a staff member pasting behavioral notes into a general-purpose chatbot to save time.
Meanwhile, federal oversight is tightening, not loosening: the Department of Education has moved to require state agencies to certify FERPA compliance, and third-party data-sharing violations have been rising. The exposure isn’t hypothetical; education-sector data incidents have grown year over year, with millions of records affected industry-wide.
What school leaders should actually be asking
The useful question isn’t “is it free”; it’s:
- Is this tool provisioned under our institutional agreement, or is staff using a personal account?
- Does the vendor’s Data Processing Agreement explicitly exclude our data from model training?
- Who is a “school official” under FERPA here, and is that status contractual, not just marketed?
- What’s the retention window, and can we get it in writing?
A free label tells you nothing about the answer. A signed agreement does. For school leaders trying to move fast on AI adoption without creating a data governance problem, that distinction, not the price tag, is the one worth building policy around.
