Choosing a Pilot: A Decision Guide for Districts

Every district evaluating an AI pilot starts in the wrong place. The meeting opens with someone asking which model is smartest, and forty-five minutes later everyone is comparing benchmark scores nobody in the room can independently verify. It's the wrong question because, at the frontier, the answer doesn't matter to a lesson plan. ChatGPT, Claude, Gemini, and Copilot are all good enough to draft a differentiated worksheet, summarize a faculty PDF, and rephrase a parent email. The gap between them on the actual work a teacher does is real but small. The gap between them on fit — how they slot into what your district already runs, already pays for, and already worries about — is enormous. That's the decision. Here's how to make it.

Start with the stack you already run

The single most predictive fact about which pilot will succeed is not the model. It's your identity provider. Districts run on either Google Workspace for Education or Microsoft 365, and that choice quietly decides most of the rest.

If you're a Google district, Gemini for Education is the path of least resistance. It starts around $14 per user per month, lives inside Docs, Slides, and Classroom where teachers already work, and inherits your existing Google admin controls and single sign-on. There's no new identity system, no separate roster sync, no third account for a teacher to forget the password to. The friction of adoption — which is where most pilots quietly die — is lowest here.

If you're a Microsoft district, the mirror image holds. Copilot for Education, around $21 per user per month, sits inside Word, Teams, and Outlook, governed by the Microsoft admin center you already staff. A Microsoft shop piloting Gemini, or a Google shop piloting Copilot, is signing up to fight its own identity infrastructure for the length of the pilot. That fight is losable, but you should only pick it for a very good reason.

ChatGPT for Teachers and Claude for Education are the stack-neutral options — they don't assume you live in either productivity suite. That's a liberation if your district is mixed or if the pilot's whole point is to evaluate the tool on its own terms, and a mild tax if you were hoping the AI would show up inside the documents teachers already have open.

Then price the thing honestly

Now overlay cost, because the numbers are not close. Claude for Education and Team run about $20 per user per month; Copilot about $21; Gemini from about $14. And ChatGPT for Teachers is free for verified U.S. K-12 teachers and staff through June 2027, via SheerID verification, with per-district workspace controls.

For a 2,000-staff district, the annual delta between "free" and "$20 a seat" is real money — roughly the cost of several teaching positions. That makes ChatGPT for Teachers the obvious opening bid for a budget-constrained pilot, and it usually should be. But price a free offer the way you'd price any free offer: read the expiry and the scope. The free window closes in June 2027, and the offer covers teachers and staff, not students — a distinction that trips up more district plans than any pricing table, and one the series unpacks in no, students don't get free ChatGPT too. If your pilot's real goal is a student-facing tool, "free for teachers" is answering a question you didn't ask.

Free is the easiest number to compare and the hardest to plan around. A $20 seat is a budget line. A free offer with an expiry date is a budget line you haven't written yet.

Match the tool to your actual priority

With stack and price on the table, the tiebreaker is what you're actually trying to accomplish. Name the priority out loud before you pick, because each tool has a center of gravity.

• Teacher time-savings, fastest. If the pilot's mandate is "give teachers their evenings back" and nothing more, the free ChatGPT for Teachers workspace is hard to beat — no per-seat approval, unlimited access to a flagship model, and a scope deliberately built around the adult in the building.
• Deep writing and reasoning workflows. If your priority skews toward long-form analysis, careful writing feedback, and reasoning you can audit, Claude's education tier is built around that lane, at a real per-seat cost.
• Documents-first integration. If you want the AI to show up inside the files teachers already live in, the answer is simply whichever suite you already own — Gemini for Google, Copilot for Microsoft.

Notice what's not on this list: "the smartest model." At this altitude, priority beats raw capability, because all four clear the bar for the work.

A framework you can actually run

Collapse the above into a decision you can defend to a school board:

1. Default to your suite for anything documents-first. Google district, students in Classroom all day, integration is the goal → pilot Gemini. Microsoft district, everything in Teams → pilot Copilot. Don't overthink the model; the workflow win is bigger than the benchmark win.
2. Default to ChatGPT for Teachers for a fast, teacher-only, budget-zero start. It's free through 2027, verification is per-teacher via SheerID, and it commits you to nothing. It's the ideal low-risk first pilot — provided you've internalized that it doesn't cover students, and that the free window has an end date. The foundational feature-and-cost comparison is worth reading before you sign.
3. Reach for Claude's tier when auditable reasoning is the point and you have budget to spend on it.
4. Whatever you pick, pilot small, measure one thing, and don't marry the vendor. A pilot is a hypothesis, not a marriage. Pick a single cohort — one department, one grade band — define one outcome you'll actually measure (hours saved on prep, or quality of differentiated materials), and run it for a term. The switching cost between these tools is far lower during a 30-teacher pilot than after a district-wide rollout.

The uncomfortable truth is that most of the anxiety in the pilot decision is misplaced. Districts agonize over the model and under-think the identity provider, the expiry date, and the teacher-versus-student scope — the three things that actually determine whether the pilot survives contact with a real building. Get those three right and any of the four tools will do the job. Get them wrong and the smartest model in the world will sit unused behind a login teachers can't remember.

Start with the stack. Price it honestly. Name the priority. Then pick, and keep the pilot small enough that changing your mind is cheap. The goal was never to choose the best AI. It was to choose the one your district can actually run.

Part 97 of 100 in the ChatGPT for Teachers series. Previously: Building the Skill a Free Chatbot Can't Replace. Next: No, Students Don't Get Free ChatGPT Too. Browse more builder insights or explore AI skills for education at aiskill.market.