Your first week with an AI assistant

ChatGPT, Claude and Gemini side by side, and how to pick. About 1 h, beginner level.

You'll be able to

• Set up an AI assistant for university work with the right account and sensible privacy habits.
• Compare ChatGPT, Claude, Gemini and Copilot on the same real task and choose one for a purpose.
• Keep course or project context in one place with projects, custom instructions and files.

What an assistant is, and isn't

ChatGPT, Claude, Gemini and Microsoft Copilot are chat interfaces wrapped around large language models. The model writes each reply by predicting likely text, based on its training and on everything in the current conversation.

Three consequences matter from day one:

• It only knows what's in the chat. Your course, your reader and your deadline are invisible unless you say them.
• Its built-in knowledge stops at a training cutoff. For recent events, use a mode that searches the web, and open the links it shows.
• It sounds equally sure when it's wrong. Fluent writing is not evidence of accuracy.

Around the model, each product adds features: file uploads, web search, voice, image generation, running code on data, and places to save instructions. Those features, more than the chat itself, are what separate the tools in daily use.

Start with the right account

Which account you sign in with decides where your conversations go and what the provider may do with them.

At USC, ChatGPT Edu is provided to students, faculty and staff. Sign in at chatgpt.com with your USC credentials so university work stays in the university workspace. For Claude, Gemini and Copilot, check USC IT Services for current institutional access before relying on a personal account; Copilot and Gemini behave differently depending on whether you're signed in with a school or personal account.

Whatever the tool, some material doesn't belong in a personal account at all:

• student records and grades
• health or other identifiable personal data
• unpublished research, manuscripts under review, or data covered by an agreement
• anything you'd need permission to share outside the university

In a personal account, open the data settings once and decide whether your chats may be used to improve the product.

The big four, side by side

The main assistants overlap far more than their marketing suggests. All four can draft, explain, summarize and answer questions about files. Their differences are mostly in where they live and what they're built around:

• ChatGPT: the broadest set of features in one place, including file and data analysis, image generation, voice and custom GPTs you can share.
• Claude: strong on long documents and careful writing, with Projects for keeping files together and Artifacts for documents and small apps.
• Gemini: built into Google's apps (Docs, Gmail, Drive), with strong image and video understanding and Deep Research reports.
• Microsoft Copilot: works inside Word, Excel, PowerPoint, Outlook and Teams, where your Office files already are.

Models are updated often, so rankings from last semester may not hold. The reliable way to choose is to run your own real task through each one and judge the results.

Keep context in one place

Retyping who you are and what you're working on in every chat gets old fast, and the answers suffer when you forget. Most assistants let you save context:

• Projects (ChatGPT, Claude) group chats with shared files and instructions, such as a syllabus, a reading list or a style guide.
• Custom GPTs (ChatGPT) and Gems (Gemini) are reusable assistants with their own standing instructions.
• Custom instructions apply to every chat: your role, your field, the tone and length you prefer.

Two habits keep answers sharp. Start a new chat when you change topic, because a long conversation carries old details that can pull answers off course. And check whether memory is on: some assistants remember facts across chats, which is convenient but worth reviewing in settings.

Decide what to hand over

The AI Fluency Framework by Rick Dakan and Joseph Feller describes four habits of effective AI use. They're a good checklist for your first week:

• Delegation: decide what the AI should do and what you should do. Hand over tasks where you can judge the result; keep the ones that are the point of your learning or need your expertise.
• Description: tell it clearly what you want, how to approach it and how to behave.
• Discernment: evaluate what comes back. Is it accurate, fit for the audience, and reasoned soundly?
• Diligence: take responsibility for what you use and share, including being open about AI's part in it.

For coursework, delegation also means reading your syllabus. Policies differ between courses, and the policy decides what's allowed, not the tool.

Same task, three assistants

The task: a staff member needs to explain a new room-booking process to 200 students in a short email.

The prompt, identical in each tool: “I coordinate events for a USC student center. Write a 120-word email telling students how to book study rooms from next week: book online, up to two hours a day, cancel 24 hours ahead. Friendly, plain language, one clear call to action.”

Scoring: before reading any answer, decide what good looks like: facts correct and complete, right length, clear action, tone that fits students.

Results: answer A was warm but invented a room-capacity rule nobody mentioned. Answer B was accurate but ran to 200 words. Answer C was accurate, 118 words, and ended with a single link line.

One-line verdict: “C won because it stuck to the facts I gave and hit the length; A would have misinformed students.” That line is the useful part: next time you know what to check first.

Try it

One real task, the identical prompt you gave three assistants, their answers ranked, and one line on why the winner won.