The Teacher's Guide to Building an AI Assistant: Claude vs. ChatGPT vs. Gemini

You've heard of all three. You've probably tried at least one. And if you're like most teachers who've experimented with AI, the experience went something like this: you asked it something, you got something useful, you asked it something else a week later, and it had no memory of anything.

Start over. Explain yourself again. Get a generic answer. Wonder why people keep saying this is going to change everything.

Here's what nobody tells you upfront: the tool isn't the problem. The missing piece is a setup — a way to build a persistent assistant that already knows who you are, who your students are, and how you like to work. One that doesn't need to be re-briefed every single session.

All three major AI platforms have a version of this. Claude calls it Projects. ChatGPT calls it Custom GPTs. Gemini calls it Gems. Different names, same idea: configure it once, and it holds your context permanently.

This post is the map. What they have in common, where they actually differ, and how to figure out which one to build first — without spending six weeks researching it when you have 87 unread emails.

The Thing That's True of All Three

Before the differences — the sameness. Because this matters more than which tool you pick.

Every AI tool you've found underwhelming was probably used as a one-off. Open it, ask a question, get an answer, close it. That's not how any of these tools are designed to work. That's like hiring a sous chef to pour a glass of water.

What makes AI actually useful for teachers is context. When the tool knows your grade level, your students, your format preferences, your non-negotiables — the output quality jumps. The gap between "AI is kind of useful" and "AI is saving me real time" is almost always a context gap.

Projects, Custom GPTs, and Gems all solve this the same way: write your context once, store it in a dedicated workspace, and every conversation starts from there. No re-explaining. No generic outputs calibrated for some hypothetical teacher in some hypothetical classroom.

The principles for building any of them are identical. The mechanics differ slightly. That's it.

What Goes Into Any Good Setup

Four questions. Get these right and the rest is just details.

Who are you? Grade level, subject, years of experience. What you care about. What you explicitly don't want — jargon, activities that assume a laminator, suggestions that would require three more prep periods to actually implement.

Who are your students? The honest picture — reading levels, ELL learners, students with IEPs, what's actually true about your classroom population. Not the idealized version. The more accurate this is, the more calibrated everything the tool produces.

How do you like to work? Full draft or options? Bullet points or prose? Questions before attempting, or attempt first and ask after? This step gets skipped most often. It's also the one that explains why AI output so frequently feels slightly off even when it's technically fine.

What is this assistant for? One lane. Differentiation. Assessment. Parent communication. Lesson planning. A focused assistant beats a catch-all one — and you can always build a second one.

Write those answers in two or three paragraphs. Paste them into whichever tool you're building in. Done. That's your setup.

Where They Actually Differ

Same principles. Real differences in practice.

Claude — For complex tasks that require actual reasoning

Claude is built to slow down on hard requests rather than confidently producing something plausible-but-wrong. For tasks that require genuine nuance — differentiating an assignment for multiple needs at once, thinking through a difficult parent situation, analyzing student work — it handles complexity carefully. It's also notably honest when it's uncertain rather than filling gaps with confident-sounding nonsense. (Which, for teachers trusting a tool with real materials, matters.)

Claude Projects is a clean workspace. Instructions, uploaded files, persistent context. It holds long, complicated setups without losing track.

Start here if you want an assistant for multi-layered tasks that require judgment — particularly differentiation work or anything that involves nuance over speed.

Full guide: How to Build Your AI Teaching Assistant with Claude (Projects)

ChatGPT — For volume and versatility

ChatGPT generates quickly and handles a wide range of tasks competently. For high-volume work — producing multiple versions of something, generating lots of options to choose from fast, quick drafts you'll revise — it's fast and capable. It's also the tool most teachers have already tried, which means there's a lot of community knowledge, example prompts, and "here's what worked for me" content floating around.

Custom GPTs have a well-developed builder interface. Straightforward to set up, mature feature set, relatively easy to navigate even if you're not particularly technical.

Start here if you need speed and volume across many different task types, or if you're already in the ChatGPT ecosystem.

Full guide: How to Build Your AI Teaching Assistant with ChatGPT (Custom GPTs)

Gemini — For teachers already inside Google

If your school runs on Google — Classroom, Docs, Drive, Meet — Gemini has a practical edge the other tools don't have. Its integration with Google Workspace means you can work with your AI assistant directly inside the tools you already use, rather than copying and pasting between platforms. Less friction. Workflow stays in one place.

Gems work the same way as the others at the core — instructions, files, persistent context. The differentiator is the ecosystem. If you're already deep in Google, the case for staying there is strong.

Start here if your school is all-in on Google Workspace and you want an assistant that works inside those tools rather than alongside them.

Full guide: How to Build Your AI Teaching Assistant with Gemini (Gems)

How to Pick One

Three questions. Seriously, just three.

First: what does your school already use? If you're deep in Google, start with Gemini. If your district has a ChatGPT or Claude license, start there. Remove the access friction first. The best tool is the one you can actually get to.

Second: what's the task costing you the most time right now? Complex multi-layered differentiation and nuanced tasks lean toward Claude. High-volume quick generation leans toward ChatGPT. Anything that lives in Google Workspace leans toward Gemini.

Third — most important — which one will you actually open on a Tuesday afternoon when you're tired? The best assistant is the one that fits how you already work, not the one with the most impressive feature list. A tool you use consistently beats a powerful tool you approach from scratch every time and eventually stop using.

Pick one. Build it. Use it for a month before you decide whether to explore others.

What None of Them Can Do

Worth saying plainly, no matter which platform you choose.

No AI tool knows your actual students. It knows the population you described. Your judgment about specific kids is irreplaceable — nothing about this changes that.

Keep confidential information out of all of them. No student names, no identifying details, nothing you'd be uncomfortable having stored on a third-party server. Describe patterns and populations, not individuals.

Review everything before it goes to students or parents. Better starting points. Not finished products. The judgment call is still yours. That's appropriate.

The Real Shift

Most teachers who've tried AI and moved on weren't wrong that it wasn't useful. They were using it the hard way — starting from zero every time, getting generic outputs, concluding the whole thing was overhyped.

Understandable. Also: premature.

A one-off prompt to a blank AI gets you a generic response. A well-configured assistant that knows your classroom, your students, and how you work gets you something you might actually use. That's not a small difference. That's the difference between a tool that adds work and a tool that removes it.

Twenty minutes of setup. Pick a tool. Build the thing. Come tell us what changed.

Want to work through this with other teachers who are figuring out the same things? The STRONG Teacher's Lounge is where that conversation lives. Share what you're building, ask what's not working, see what other teachers have already figured out. Join the Lounge here.