Stop Starting Over: How to Build a Permanent AI Teaching Assistant with Claude
Learn how to use Claude's Projects feature to build a persistent AI teaching assistant that already knows your classroom — no re-explaining yourself every time.
You've used Claude. Maybe a few times. You got something useful out of it — maybe even something great — and then the next time you opened it, you had to explain everything from scratch. Your grade level. Your students. The fact that you need things practical and not written like a corporate memo.
So you explained it again. And it worked okay. But here's the thing: a tool that makes you re-introduce yourself every morning isn't actually saving you time. It's just moving the work around.
That friction has a fix. It's called Projects, and it changes how Claude functions for you entirely.
What a Project Actually Is
A Project is a dedicated workspace in Claude with its own permanent instructions, its own file storage, and its own memory of what you've shared. Every conversation you start inside that Project begins with Claude already knowing your context.
Right now, every time you open a new chat, you're meeting a version of Claude with no idea who you are. Competent, willing, completely without context. That's the generic-output problem — and it's not Claude's fault. You just haven't told it anything yet.
Think of a Project as Claude after six months at your school. It knows your grade level, your students' reading range, that three of your sections have a significant number of ELL students, and that you need plain language — not education-speak — and one-page lesson plans. You don't say any of that again because you already put it in the Project.
One setup. Permanent context. Twenty minutes.
The Setup (Do This Once, Do It Right)
When you create a Project, you write instructions that tell Claude how to behave. This is the part that matters most. Don't rush it and don't be vague — vague instructions produce vague outputs, every time.
Your instructions need to cover four things: who you are, who your students are, how you like to work, and what this Project is actually for.
The "who you are" part is shorter than you think. Grade level, subject, years of experience, what you care about, what you don't want. ("No jargon. No activities that require a functioning printer." Whatever is true for you.)
The student description is where most teachers undersell themselves. Don't say "mixed levels." Say: my classes run from about 4th-grade reading level to on-grade, I have several students with IEPs focused on executive function, and roughly a third of my students are ELL at varying stages. That specificity is what calibrates the output. You're not identifying anyone — you're describing your population.
How you like to work matters more than people expect. Full draft or three options to choose from? Should Claude ask clarifying questions before attempting something, or take a shot first and ask after? Bullet points or prose? Spell it out. This is the step most teachers skip, and it's exactly why AI output often feels slightly off even when it's technically correct.
Then name the lane. This Project is for building and refining materials for my 7th grade ELA classes. A focused Project beats a catch-all one. Build a second Project for parent communication if you need it — don't try to make one workspace do everything.
Starter template to adapt:
I'm a [X]-year veteran [grade] [subject] teacher. My classes average [number] students. My students include [honest, specific description — reading levels, ELL learners, students with IEPs, whatever is true]. Plain language over jargon, always. I prefer [format preference]. Give me your best attempt first; ask clarifying questions after if needed. This Project is for [specific focus].
What to Upload
Projects let you store files Claude can reference across every conversation. Use this.
Your curriculum map or pacing guide is the most useful starting point — it grounds every planning conversation in where you actually are in the year. Rubrics are worth including too, so feedback language Claude helps you draft is calibrated to what you actually grade against. And if you have a lesson plan or unit overview you're proud of? Upload it. It teaches Claude your voice and structure better than any description can.
Don't try to upload everything at once. Add things as you notice what's missing. The Project gets more useful over time — which is exactly the point.
What Actually Changes
Differentiation no longer requires a paragraph of setup. You just say "differentiate this for my three reading tiers" and Claude already knows what those tiers look like in your classroom.
Assessment requests stop being vague. "Write a quiz on this standard" now produces something you might actually use because it knows your grade, your students, and your format preferences.
Parent communication gets faster. If you've uploaded your school's communication templates, Claude writes toward those instead of inventing something that sounds like it came from an HR department.
Planning conversations get genuinely useful. With your pacing guide in the Project, you can ask, "Given where we are in the unit, what's worth reviewing before the assessment?" and get a grounded answer. That's different from what most teachers have experienced with AI — and it's the difference between a tool that feels gimmicky and one that actually earns a place in your practice.
The Limits
Claude doesn't know your actual students. It knows the population you described — and your judgment about any specific kid always takes precedence over anything an AI produces. Don't put student names or identifying details into any AI tool. Keep it at the level of role and pattern.
Review everything before it goes to students or parents. A well-built Project gives you better starting points. Not finished products. That call is still yours — and should be.
The Deeper Point
Most teachers use AI like a scratch pad. Occasionally useful, easily forgotten, never really integrated into how they work. A Project is the difference between that and having something that functions like an actual part of your practice — a tool you return to because it already knows where you left off.
Twenty minutes of setup. Everything after that compounds.
Small system. Gets smarter over time. That's the whole thing.
Want to build yours alongside other teachers who are figuring out the same things? That's what The STRONG Teacher's Lounge is for. Come share what you're building. Join the Lounge here.
Related Posts:
- → How to Build Your AI Teaching Assistant with ChatGPT (Custom GPTs)
- → How to Build Your AI Teaching Assistant with Gemini (Gems)
- → The Teacher's Guide to Building an AI Assistant: Claude vs. ChatGPT vs. Gemini
Strong Teacher Pep Talk Playlist