Custom GPTs vs Claude Projects vs Gemini Gems: Which Should You Use?

If you use AI for the same jobs every week (writing proposals, answering customer emails, summarising meetings), you have probably noticed that pasting the same long prompt again and again gets old fast. Every major assistant now lets you save that setup: ChatGPT has Custom GPTs and Projects, Claude has Projects and Gemini has Gems. They sound similar, and they overlap, but they are built for slightly different jobs. This guide explains the difference in plain English, helps you decide which one to use for which task, and shows how to set up the same assistant in each so you can try them side by side.

The short answer

• Custom GPT (ChatGPT): a reusable tool. Best when you want a single-purpose assistant you can share with others, by link or publicly, and that behaves the same way for everyone.
• ChatGPT Project: a workspace. Best for an ongoing piece of work, such as a client, a book or a product launch, where you want chats, files and instructions grouped together.
• Claude Project: a workspace built around a knowledge base. Best when your work depends on long documents (policies, research, contracts, style guides) and you want every chat to draw on them.
• Gemini Gem: a quick saved persona. Best if you live in Google Workspace and want lightweight helpers you can create in a minute.

Plans, file limits and sharing options change often, so check each tool’s current help pages before you rely on a specific limit. The decision logic below stays the same.

Side-by-side comparison

How to decide: four questions

1. Will other people use it? If yes, and they are outside your team, a Custom GPT is the easiest to share. For teammates, any of them can work on a team plan.
2. Is it a task or a piece of work? A task (turn meeting notes into action items) suits a Custom GPT or Gem. A piece of work (everything about the Smith account) suits a Project.
3. How much reading does it need? If the answers depend on long documents, Claude Projects tend to be the most comfortable home, because the knowledge base is central to how they work.
4. Where do you already work? If your team lives in Gmail, Docs and Drive, a Gem keeps you in that world. If you already pay for one assistant, start there.

Build the same assistant in all three

The best way to choose is to build one real assistant three times. Use a job you do every week. Here we will use a “Proposal First Draft” helper for a small web design business.

Step 1: Write the instructions once

Write them in a document first, so you can paste the same text into each tool:

ROLE: You draft first versions of client proposals for Brightside Web Studio.
INPUT: I will paste the client's enquiry and my call notes.
OUTPUT: A proposal with these headings: Summary, Goals, Scope,
Timeline, Price, Next steps. Under 600 words.
RULES:
- Use prices only from pricing_2026.pdf. Never invent a price.
- If scope is unclear, list questions to ask the client instead of guessing.
- Friendly, confident, no jargon. UK spelling.
FINISH: List any assumptions you made.

Step 2: Prepare one or two clean files

For example, a one-page price list and one past proposal you are proud of. Remove anything confidential you would not want shared, such as other clients’ names.

Step 3: Set it up in each tool

• ChatGPT Custom GPT: go to GPTs, choose Create, then the Configure tab. Paste your instructions, add the files under Knowledge, add two or three conversation starters (“Draft a proposal from these notes”) and choose who can access it.
• Claude Project: open Projects and create a new project. Paste your instructions into the project instructions and upload the files to the project knowledge. Every new chat inside the project uses them.
• Gemini Gem: open the Gem manager and create a new Gem. Give it a name, paste the instructions and add the files if your plan allows it. Use the preview pane to test before saving.

Step 4: Test with the same five requests

Give each assistant the same five real enquiries, including one tricky one: a client asking for something that is not on your price list. Score each output on accuracy, tone and whether it followed the rules. Most people find a clear winner for their task within half an hour, which is far more useful than any general ranking. Our lesson on testing prompts for consistent results shows how to do this properly.

Common mistakes with saved assistants

• Putting secrets in shared assistants. Assume anyone who can use a shared assistant may be able to coax out its instructions and file contents. Keep confidential data out of anything shared widely. See our lesson on privacy and what not to paste.
• Too many files. Five clean, current files beat fifty loosely related ones. Outdated files produce outdated answers.
• Vague instructions. “Be helpful and professional” does nothing. Give a structure, rules and examples of good output.
• Never updating it. Put a date in your file names and review each assistant monthly, especially after price or policy changes.
• Trusting it blindly. Saved assistants still make mistakes. Check facts, prices and names before anything goes to a client. Our guide on how to fact-check AI answers gives a five-minute routine.

Five saved assistants worth building first

1. Email reply drafter with your tone and common answers. Our prompt templates for small business emails make a good starting point.
2. Meeting notes to action items: decisions, owners, deadlines, risks.
3. Social post helper with your brand voice and platform rules.
4. Customer FAQ assistant loaded with your policies, for drafting support replies a human checks.
5. Weekly report summariser that turns a spreadsheet export into five bullet points.

When you are ready to connect these assistants to your other apps, so new emails or form entries trigger them automatically, read our comparison of Zapier vs Make vs n8n for beginners.

Learn it step by step

Our Prompt Writing That Works course includes a full lesson on Custom GPTs, Claude Projects and Gemini Gems and another on reusable templates and custom instructions. To apply them to real business tasks, continue with AI for Business and Marketing.

Building a support assistant? A saved assistant works well for customer emails. See our guide on how to use AI for customer support replies, with a support brief template and a safe review workflow.

Frequently asked questions

What is the difference between a Custom GPT and a ChatGPT Project?

A Custom GPT is a reusable assistant built for one job that you can share. A Project is a workspace that groups your own chats, files and instructions around one ongoing piece of work.

Do I need a paid plan?

It depends on the tool and changes over time. Using shared assistants is often possible on free plans, while creating them, adding more files or sharing with a team may need a paid plan. Check each tool’s current plan page.

Which is best for long documents?

Claude Projects are designed around a knowledge base, so they are a strong choice for document-heavy work. ChatGPT and Gemini also accept files, so test with your own documents.

Can I move an assistant from one tool to another?

Yes, if you keep your instructions and files in your own documents. Paste the same instructions and upload the same files into the other tool, then test again.

Are saved assistants private?

Private ones are visible only to you, but shared ones can leak their instructions and file contents to anyone who uses them. Never put confidential information in an assistant you share outside your team.