Should Teachers Customize Their AI? The Hidden Trade-Offs
Exploring the benefits of efficiency, the risks to privacy, and how educators can make intentional choices
A couple of weeks ago I wrote a piece about the persistent memory features of the five major Large-Language Models, hoping to give educators a clearer sense of what these chatbots remember about us and how they use that information.
I received multiple requests to focus less on the technical features of persistent memory and write a new blog that provides clear instructions for how teachers can modify the settings in their profiles so that they don’t have to keep repeating themselves. You know, the way you have to tell a bunch of sixth-graders 475 times that the test is on Friday.
Because I work heavily with ChatGPT, I can bypass this hassle by creating my own custom GPTs that complete repetitive work without asking me a bunch of dumb questions. Yes, I’m a sixth-grade teacher in California. Yes, I teach social studies and language arts. Yes, I prefer four-column rubrics with scoring guides. Yes, I want versions of all my assignments in English and Spanish.
There’s even a sustainability angle: The more text you type in a chat window, the more power and water the system uses. Custom instructions help cut down on that digital footprint.
As a public service, here is a teacher’s guide to creating custom instructions for the major LLMs so that they know you well enough to meet your needs with a minimum of fuss and environmental costs.
Prompt Starter
This is a sample prompt that I have used to launch customization:
“My name is David. I teach sixth-grade social studies and English/language arts to sixth-graders at a middle school near Los Angeles. I base all my curriculum and assessments on the California Social Studies/History framework as well as the California Common Core Standards for E/LA Literacy. Nearly 50 percent of my students are English Language Learners with Spanish as a home language. That means all my requests for curriculum, newsletter, emails, and assessments must be generated in both languages. I prefer a conversational tone. My school offers Accelerated Reader. My students have district-issued tablets that they use in class and take home. Google Classroom is our LMS. If my request is ambiguous or missing key details (goal, audience, constraints, context), ask up to 3 specific clarifying questions. End every response by asking me for next steps. Remember this information and use it to shape your outputs.”
Now let’s take a look at customization instructions for each of the major LLMs being used by teachers.
ChatGPT
Setting up custom instructions in ChatGPT is surprisingly straightforward, though you’ll still need to wade into the settings menu — a scary place most users avoid like a middle school cafeteria on burrito day. The guidance you see here includes updates from OpenAI’s new process (as of Sept. 17, 2025).
Here’s how to do it:
Open ChatGPT (web, desktop, or mobile).
Click on your profile picture (bottom left on desktop; top corner on mobile) and open Settings.
Find Custom Instructions (sometimes labeled Personalization).
Toggle it on if it isn’t already.
Two boxes will pop up. Fill them out:
What would you like ChatGPT to know about you to provide better responses?
How would you like ChatGPT to respond?
Save your work, then start a new conversation to see the changes take effect. (Note: Edits don’t retroactively fix old chats.)
Each box gives you up to 1,500 characters, which is plenty of space to set guardrails for how ChatGPT should behave — whether that’s “answer like a patient math tutor” or “channel your inner Anthony Bourdain when discussing recipes.”
And here’s the good news: Custom instructions are now available to all users, not just paid subscribers. If you want to update or disable them, just head back into settings, edit what you’ve written, or flip the toggle off.
Note: Paid subscribers can write longer, more detailed instructions for customization.
Gemini
While Gemini does not have a feature called "Custom Instructions" that remembers context across all conversations, you can achieve a similar result for the duration of a single conversation.
Instead, you front-load your preferences in the first prompt, giving Gemini the role, tone, and context you want it to use throughout the conversation.
For example:
Your Role: "Act as curriculum designer who specializes in social science.”
My Role: "You are my researcher and instructional coach. All your answers should be concise and bulleted."
Preferences: "I prefer short summaries. Always use a professional but friendly tone."
By including this information at the beginning of a chat, Gemini can use that context to tailor responses to you. This works for the entire conversation, so you only need to set it up once for that specific chat session.
Claude
Claude takes a slightly different tack, offering three personalization tools: Profile Preferences, Styles, and Project Instructions:
Profile Preferences (Account-Wide)
This is the closest equivalent to ChatGPT's Custom Instructions:
How to Set Up:
Click your initials in the lower left corner
Select Settings
Under "What preferences should Claude consider in responses?", describe your preferences.
What to Include:
Your preferred approaches or methods
Common terms or concepts you use
Typical scenarios you encounter
General communication preferences
Any preferences you add here will be applied to all of your conversations with Claude.
2. Styles (Communication Format)
Styles customize how Claude communicates with you. Unlike profile preferences and project instructions which provide context and guidance, styles focus specifically on how Claude formats and delivers its responses.
Use Styles When You Want To:
Adjust the tone and format of Claude's responses
Switch between different communication styles (e.g., concise for quick answers, explanatory for learning)
Create custom communication patterns based on your own writing
3. Project Instructions (Pro/Team Plans Only)
Project instructions help Claude understand the specific context and requirements for a particular project. These instructions only apply to chats within that project.
Note: Projects are only available on Claude Pro and Claude for Work (Team and Enterprise) plans.
Grok
Grok also offers customization through a simple “Customize Grok” option available on the web, mobile app, or inside X. All users can access it, though some advanced features require X Premium.
How to set it up:
Log in with your X account and start a new chat.
Look for the Customize Grok option (below the chat box or in settings).
Fill in two fields:
What should Grok know about you? (e.g., role, subject area, goals)
How should Grok respond? (e.g., tone, style, format)
Choose whether these settings apply to all conversations or just one session.
Save, test, and adjust as needed.
Unlike ChatGPT, which applies custom instructions across every chat, Grok lets you turn customization on or off for each session.
Copilot
Copilot has a similar feature, but it’s called “memory” and it’s designed to personalize your experience without needing to re-explain everything every time you chat with this bot.
With memory turned on, Copilot can:
Remember key facts about you — like your name, interests, goals, and preferences.
Adapt tone and style — whether you prefer concise answers, deep dives, or a more casual vibe.
Recall context across chats — so it doesn't start from scratch every time.
Build on past conversations — helping you brainstorm, plan, or reflect with continuity.
Unlike static custom instructions, Copilot’s memory evolves as you chat. You don’t need to fill out a form — just tell the bot what you want me to remember.
Example: “Remember that I’m a teacher of social studies in sixth grade. Forget that I said I taught science in Arkansas — I moved to California and changed jobs.”
How to Manage It
You can view, edit, or delete anything Copilot remembers about you:
Click your profile icon.
Go to Settings > Privacy > Personalization and memory.
From there, you can turn memory on/off or manage what the bot remembers.
Want to personalize Copilot even more?
Try telling the bot:
Your role and what you’re working on
How you like to receive information (e.g., visual, concise, detailed)
Topics you care about (assessment, lesson plans, project-based learning, etc.)
Any quirks or preferences (“I hate jargon” or “Use analogies when explaining theory,” etc.)
As you can tell, the Copilot customization process is more organic and conversation driven.
Final Thoughts
Every company in the ed tech space states that one of the primary value propositions of using generative AI in education is to save teachers time. That process is facilitated by the customization features inherent in the major Large-Language Models that teachers use to complete repetitive or time-consuming tasks.
I can’t help you as you wrestle with the usual cost-benefit analysis. How much information are you willing to share with an AI (and the company that scrapes data from it) in exchange for the time you will save? I operate under the assumption that privacy, at least in the digital age, is more illusion than reality. I am willing to share vast amounts of information in order to save time, which I view as nearly priceless. Folks, you have a cell phone. Give up the illusion of privacy.