ChatGPT for Teachers reached 55 more US school systems and 100,000 teachers

OpenAI is scaling distribution: more districts, more teachers, more states. Anthropic is running a pilot in one district with a stated research question about wellbeing and practice, alongside a broader free offer.

Both are competing for the same position, and it is a valuable one. Educational software wins through institutional adoption and then persists, because switching costs in a district are enormous and a tool teachers have built their practice around does not get replaced. Free access to schools is a customer acquisition strategy with a long payback and a very long tail.

It is also the second time this decade the sector has been handed free software by companies whose interest is in the students becoming adults who use their products.

The evidence base is thin in both directions

What neither company has published is evidence that the tools improve anything.

The realistic use is administrative: lesson planning, differentiating materials for different reading levels, drafting parent communications, generating practice questions. Teachers do enormous amounts of unpaid preparation, and a tool that reduces it is a genuine benefit even if it never touches instruction.

That is a defensible claim and it has not been tested at scale. Anthropic's Detroit pilot is aimed at exactly this question, which is more than anyone else is doing, and it is one district.

The evidence that does exist from adjacent deployments is not encouraging about unsupervised use. NHS England's watchdog documented AI clinical scribes producing wrong drug names, missing prescription advice and a fabricated diagnosis across 27 systems in use — a simpler task than generating instructional content, with better-trained reviewers.

A Nature Human Behaviour study found that model polishing reduces variance in writing complexity by 21 to 50 percent across more than 880,000 texts. In a classroom, that variance is what teachers assess.

The part nobody is discussing

The students are the unexamined half. These are teacher tools, and the framing keeps the conversation on educator workload rather than on what happens when AI-generated materials are the materials.

The surrounding evidence about AI in schools is meanwhile poor. WIRED documented four teachers describing sexualised deepfake images allegedly created by students. Institutions and platforms were slow to respond in every case.

A district adopting an AI tool for teachers is adopting it into an environment where students already have these capabilities and the institution has not worked out how to handle it.

Neither OpenAI nor Anthropic has published data retention terms for classroom use in detail, what happens to teacher-entered material about identifiable students, or any outcome evaluation. Anthropic has at least started one.