What a Three-Vendor Classroom Looks Like

The whole "which AI should schools use" debate rests on a quiet assumption: that a district picks one. In reality, most won't. A district already pays for Google Workspace, so Gemini shows up for free in the tools teachers use all day. Then OpenAI hands every verified K-12 educator ChatGPT for Teachers free through June 2027, and teachers sign up on their own. Meanwhile a handful of subject departments have quietly standardized on Claude because it writes long-form feedback the way they like. Nobody decided this. It accreted.

That's the three-vendor classroom, and it's the most likely outcome for a mid-to-large district in 2026 — not a clean single-vendor rollout. This piece is the honest look at what that actually feels like day to day: why it happens, what the stack looks like, and the governance and training tax that no procurement slide accounts for.

Key Takeaways

• Single-vendor is the plan; three-vendor is the reality. Free offers, bundled tools, and departmental preference pull districts toward a mixed stack whether or not anyone chose it.
• Each vendor carries its own data-handling rules. Gemini's education terms, ChatGPT for Teachers' training and privacy posture, and Claude's data policy are not identical — and staff can't be expected to track three sets of fine print.
• The real cost is cognitive, not financial. Free tools have a $0 license and a very non-zero training, support, and governance burden.
• Fragmentation is a compliance risk. Three tools mean three places student data can leak, three sets of settings to audit, and three vendors to name in a FERPA review.
• You can govern a multi-vendor stack — but only deliberately. Naming a default, a rule for student data, and one owner per tool turns accidental sprawl into a managed system.

Why districts end up with three vendors

Three forces push in the same direction, and none of them is a considered strategy.

The first is bundling. If a district runs on Google Workspace for Education — and a large share of U.S. districts do — Gemini arrives inside Docs, Gmail, and Classroom at no extra line item. It's the path of least resistance for "help me rewrite this email" because it's already there.

The second is the free offer. OpenAI's ChatGPT for Teachers is free for verified U.S. K-12 educators through June 2027, verified individually through SheerID. A teacher doesn't need IT to provision it; they sign up with a school email and start using it that afternoon. That's a feature for the teacher and a governance headache for the district, because adoption happens below the radar.

The third is departmental preference. Ask an English department and a computer-science department to name their favorite model and you'll often get two different answers. Anthropic's Claude, covered on this site's Claude for Teachers series, tends to win fans for long-form written feedback; others prefer ChatGPT's speed or Gemini's integration. Left alone, each department optimizes locally, and the district inherits the sum.

Put those three together and the outcome isn't a decision — it's a sediment. The question stops being "which one" and becomes "how do we govern the three we already have."

What the three-vendor stack actually looks like

Here's a realistic division of labor, not a recommendation but a description of how these tools tend to settle into roles:

The trap in this table is that it looks tidy. In practice the layers bleed. A teacher drafts a lesson in ChatGPT, pastes it into a Google Doc where Gemini suggests edits, then runs student essays through Claude because the department said to. The same student's work can pass through all three tools in a week — each with its own data terms, its own retention behavior, and its own place in (or out of) the district's compliance paperwork.

The governance tax nobody budgets for

Free is a license cost, not a total cost. The tax shows up in four places.

Data handling. Each vendor has its own answer to "is our data used for training, and for how long is it retained?" ChatGPT for Teachers states that workspace content isn't used to train models by default. Gemini's education terms and Claude's data policy have their own defaults and settings. A teacher juggling three tools cannot be expected to hold three privacy models in their head — so unless the district translates them into one simple rule, someone eventually pastes a class roster into the wrong box.

Auditing. A FERPA review has to name every place personally identifiable student information could land. Three vendors means three audit trails, three admin consoles, three sets of default settings to verify. This ties directly to the concerns in FERPA and ChatGPT for Teachers: the legal duty sits on the district, and a fragmented stack multiplies the surface area of that duty.

Support. Every tool has a slightly different login, a different way to share a chat, a different failure mode. Your help desk now fields "which one do I use for this" and "why did it log me out" across three products instead of one.

Shadow adoption. Because ChatGPT for Teachers is provisioned per-teacher, the district may not even know who's using it. You can't govern what you can't see, and "our teachers signed up individually" is not an answer a compliance officer wants to give.

The training problem: three tools, one teacher

A teacher's scarcest resource is attention, and a three-vendor stack spends it. Prompting patterns transfer reasonably well between models — a good rubric prompt is a good rubric prompt anywhere — but the interfaces don't. Where's the file upload? How do I start a fresh chat so last period's context doesn't bleed in? Which tool is cleared for student names and which isn't? Multiply those micro-decisions across a 180-day year and you get real friction.

The instinct to fix this with more training sessions usually backfires. Three separate one-hour trainings teach three tools and no principle. What actually reduces load is teaching one workflow and one data rule, then noting which tool is the default for it. The skill a teacher needs — write a clear prompt, check the output, never paste identifiable student data into an uncleared tool — is the same regardless of the logo in the corner. This is the argument this series eventually lands on in why skill-based AI tutors beat a general workspace: the durable competency is the workflow, not the vendor.

How to make a three-vendor stack survivable

You don't have to collapse to one vendor to regain control. You have to make the sprawl deliberate. Four moves do most of the work:

1. Name a default. Pick one tool as "start here for lesson work" — ChatGPT for Teachers is a strong candidate given it's free, unlimited, and purpose-built for classroom tasks. Defaults kill decision fatigue.
2. Write one student-data rule, not three. The safest rule is the same everywhere: no personally identifiable student information in any general AI tool unless the district has explicitly cleared that specific tool for it. One sentence beats three privacy policies.
3. Assign an owner per tool. Someone is accountable for Gemini's settings, someone for the ChatGPT workspace, someone for Claude. "Everyone's responsible" means no one is.
4. Inventory reality, not the plan. Survey who's actually using what. Shadow adoption you can see is a manageable stack; shadow adoption you can't is a breach waiting to be discovered.

If you're the one holding the pieces together, a lesson-planning workflow that runs the same way on any model — like the patterns in the lesson-plan studio skill or the broader education agent skills — is worth more than betting the district on a single vendor's roadmap. Portability is the hedge.

Frequently Asked Questions

Isn't it simpler to just mandate one AI tool district-wide?

Simpler on paper, harder in practice. You can mandate a default, but you can't easily un-bundle Gemini from Google Workspace or stop individual teachers from signing up for a free ChatGPT for Teachers account. Governing the real stack beats mandating an imaginary one.

Which tool should be the "default" for teachers?

For core classroom work — planning, differentiation, first-pass feedback — ChatGPT for Teachers is a reasonable default because it's free through June 2027, unlimited, and built for the job. See what's actually free and for whom before you commit, and compare against Gemini for Education if Workspace is already your backbone.

Does using three vendors triple our privacy risk?

Not exactly triple, but it does widen the surface area. Three tools mean three places data can go and three sets of settings to audit. The mitigation is one uniform data-minimization rule applied identically across all three, rather than three separate policies teachers have to remember.

How do we train teachers without three separate sessions?

Teach the workflow, not the tool. A good prompt, a habit of checking output, and one firm rule about student data transfer across every model. Cover the principle once and note which tool is the default, rather than running a tour of each interface.

What happens to our ChatGPT layer after June 2027?

That's the open question. The free offer has a fixed sunset, covered in why the sunset date matters. A district relying on it should plan for a "what if it costs money in 2027" scenario now — which is another argument for keeping your workflows portable rather than vendor-locked.

Part 26 of 100 in the ChatGPT for Teachers series. Previously: Why OpenAI Went Free When Rivals Still Charge. Next: A Week of Lesson Plans, Built in ChatGPT. Browse more builder insights or explore AI skills for education at aiskill.market.