Custom GPTs and Claude Projects for Your Team

What custom GPTs, Claude Projects and Gemini Gems are

Custom GPTs are tailored versions of ChatGPT with instructions, optional knowledge files and chosen capabilities; OpenAI explains how to create and edit GPTs.
Claude Projects are workspaces with project knowledge and instructions, which teams on business plans can share; see Anthropic's overview of Projects.
Gemini Gems are saved instructions for a recurring task, with optional files; Google describes them in its Gems help.

Labels and limits differ by plan and change often, but the thinking is the same in all three.

Pick the right first task

Start with something common, document-heavy and low risk because a person reviews the result: an onboarding helper, a proposal first-drafter, a policy question answerer, a meeting-note formatter or a customer email drafter.

Clear source material: the answers exist in approved documents.
Repeatable questions: people ask the same things often enough to justify setup.
Human review: a person checks anything important before it leaves the business.

Build it in six steps

Give it one job in one sentence: "This assistant helps office managers answer onboarding questions using the employee handbook and benefits summary, with short answers that name the source file."
Add a few approved files: current versions only, clear names, no stale drafts or comments, and a short note saying which file wins if two disagree. Keep sensitive data out unless your plan and permissions really allow it.
Write the instructions: "Answer only from the approved files. If the answer is not there, say you do not know and name who to ask. Use plain English and bullets. Separate the answer from suggested next steps."
Add three to six example answers showing the right length, tone and a polite refusal.
Set refusal and escalation rules: no legal, tax, medical or financial advice; no binding offers, refunds or employment decisions; sensitive HR, payroll, security and customer disputes go to a named person.
Build it in your business workspace, not a personal account, and share it only with the people who need it.

Test before you share it

Ask it 20 real questions: ten it should answer, five that are ambiguous and five it should refuse. Check every answer against the source file, and confirm it says "I don't know" instead of guessing. Our AI hallucinations guide explains why that matters.

Then give it to two or three people for a week, collect the questions it got wrong and fix the files or instructions rather than the symptoms.

Privacy and plan settings to check

Use a business or team plan with admin controls, and check whether your files and chats are used for model training under its terms.
Control sharing: who can use, edit and copy the assistant, and whether it can be shared outside the company.
Do not upload files people should not all see; everyone who can use the assistant can usually get answers from its files.
Know where files are stored and how to delete them.

Our list of what never to paste into ChatGPT applies to uploaded files too.

Keep it current and owned

Name an owner for each assistant, write down its job, sources and rules, and review it monthly or whenever a policy changes. Replace old files rather than adding new ones next to them, re-run your test questions after every change and retire assistants nobody uses.

When you need more than a custom GPT

Custom GPTs and Projects are great for small, low-risk jobs. You have outgrown them when you need different people to see different documents, answers must always cite the exact source, customers will use it, it must connect to live systems such as your CRM or ticketing, or you need logs of every question and answer. That is the point to look at a proper AI knowledge base or custom AI development.

Where ASL fits

ASL Help Engine answers staff questions from reviewed help content, cites its sources and only shows each person the content their role allows, the things a custom GPT cannot enforce. For customer-facing answers, Viavitna does the same from content you approve. Contact us to talk through your first assistant.

Frequently asked questions

What is a custom GPT?

A version of ChatGPT set up for one purpose, with its own instructions, optional knowledge files and capabilities, that your team can reuse instead of rewriting the same prompt.

What is the difference between a custom GPT and a Claude Project?

Both give an assistant instructions and reference files for a recurring job. A GPT is a reusable assistant in ChatGPT; a Claude Project is a shared workspace in Claude. Gemini Gems do the same in Gemini.

What files should I give a team assistant?

A small set of current, approved documents with clear names, plus a note on which file wins if two disagree. Leave out drafts, comments and anything not everyone using it should see.

How do I stop it from making things up?

Tell it to answer only from the files and to say when it does not know, add example refusals, and test it with questions it should refuse before sharing it.

When should we use something more than a custom GPT?

When you need per-person permissions, exact source citations, customer-facing answers, connections to live systems or logs of every question and answer.

Related guides

Explore the complete product

Read features, current access and limitations

Custom GPTs and Claude Projects: Build an Assistant for Your Team