artifical intelligence projects

Stop Starting Over: How to Build a Permanent AI Teaching Assistant with ChatGPT

Learn how to use ChatGPT's Custom GPT feature to build a persistent AI teaching assistant that already knows your classroom — no re-explaining yourself every time.

Photo by Levart_Photographer / Unsplash

You've used ChatGPT. You've gotten something useful out of it — a quiz, a lesson outline, a parent email you tweaked and actually sent. And then the next time you opened it, it had no idea who you were.

Back to square one. Re-explain everything. Get a slightly-better-than-generic response. Close the tab.

That loop is the reason most teachers who've "tried AI" don't actually use it consistently. And it's fixable. ChatGPT has a feature called Custom GPTs that lets you build a version of ChatGPT permanently configured for your classroom — your grade level, your students, your formats, your non-negotiables. You set it up once. It's there every time you come back.

This is how to build that.

Receive Email Updates

What a Custom GPT Actually Is

A Custom GPT is a personalized version of ChatGPT you configure for a specific purpose. You give it a name, a set of instructions, and whatever reference files it needs — and it holds all of that permanently. Every conversation starts with it already knowing your context.

The analogy: imagine a capable colleague who volunteers to help you prep materials. First time, you spend twenty minutes explaining everything. After that? They just know. They know your grade level, they know your students struggle with reading fluency, they know you hate jargon, they know your lesson plans don't exceed one page. Every conversation starts from that place — not from zero.

A Custom GPT is that colleague. You build them once. Then you work with them.

(Access note: Custom GPTs are available on paid ChatGPT plans. If you're on the free plan, the memory and custom instructions features get you much of the same result with slightly less control. The concept is identical regardless of which version you're using — give it your context, save it, use it.)

How to Build It (The Part Most People Rush)

The instructions are everything. When you create a Custom GPT, this is where you tell it who it's working with, what it's doing, and how it should behave. Don't skim this step — vague instructions are why AI outputs feel off even when they're technically fine.

Cover these four things, in your own words:

Who you are. Grade level, subject, years of experience. What you care about in your practice. What you explicitly don't want — jargon, generic suggestions, activities that assume a laminator and a prep period.

Who your students are. Not "mixed levels." The real picture: reading range, proportion of ELL learners, students with IEPs, anything relevant about your school or community context. The more honest this is, the more calibrated the output.

How you like to work. Full draft or options to choose from? Questions before attempting, or attempt then ask? Bullet points or prose? What does a "good" piece of output actually look like for you? Spell it out. This is the most skipped step and the one that pays off most.

What this GPT is for. Give it a lane — This GPT helps me build and differentiate materials for my 6th-grade science classes — and it becomes dramatically more useful than a catch-all assistant. Build a second one for parent communication if needed.

Here's a starter template:

You are a teaching assistant for a [X]-year veteran [grade] [subject] teacher. Students include [honest description — reading levels, ELL learners, IEP students, etc.]. Always use plain, direct language. No education jargon. The teacher prefers [format]. Attempt tasks first; ask clarifying questions after. This GPT is focused on [specific purpose].

What to Upload

Custom GPTs let you upload reference files. Use this wisely — not to dump everything you have, but to upload the things that would make any new assistant immediately more useful.

Your pacing guide or curriculum map grounds planning conversations in where you actually are. Rubrics mean feedback language is calibrated to what you actually assess. And if you have materials you're proud of — a strong lesson plan, a unit overview that represents your style — upload those. They teach the GPT your voice and structure faster than any description.

Add files as you realize what's missing. You'll know after the first few conversations what context the GPT is missing. Fill those gaps.

What Actually Changes

Once your Custom GPT has your context, here's what shifts in practice.

Differentiation requests stop requiring a paragraph of background. "Modified version of this for my below-grade readers" already means something specific because it knows who those students are.

Assessment creation becomes faster. Grade level, format, student needs — all baked in. A simple request gets you something calibrated instead of generic.

Parent communication improves because you've uploaded your school's communication style. You're editing a draft, not generating one from nothing — which is a very different experience.

Planning conversations get more grounded. With your pacing guide uploaded, "What would be a good approach to introducing this concept, given where we are in the unit?" is a useful question instead of a shot in the dark.

Explore the STRONG Teache Core Membership

What It Can't Do

Your Custom GPT doesn't know your actual students — it knows the population you described. Your professional judgment about specific kids is irreplaceable, and nothing about this changes that.

Keep confidential information out of it. No student names, no identifying details, nothing you'd be uncomfortable having stored somewhere outside your school. Describe patterns and populations, not individuals.

Read everything before it goes anywhere. Better starting points. Not finished products.

The Deeper Point

The reason most teachers don't get real value from AI isn't the tool. It's that they're using it the hard way — starting from zero every time, getting generic outputs, concluding that AI just isn't really for their situation.

That conclusion is understandable. It's also wrong. (Or at least, it was wrong about the tool — it was right about that particular approach.)

A Custom GPT is a system. Twenty minutes of setup, and every future conversation starts from a better place. Consistency beats intensity — a tool you actually use regularly, because it's already calibrated and ready, is worth more than a powerful tool you approach from scratch every time and eventually stop opening.

Building your Custom GPT and want to think it through with other teachers? That conversation is happening in The STRONG Teacher's Lounge right now. Join us here.

Related Posts (coming soon):