Custom assistants and projects: setting AI up for your membership

If you use AI regularly, you have probably noticed how much of each conversation goes on explaining yourself. Who your members are. How you like things written. What the membership offers. Which words you never use. You paste the same background again and again, and when you forget, the results slide back to generic.

Custom assistants and projects solve this. They let you store instructions and reference files once, so every conversation starts already knowing your business. For a membership owner doing the same kinds of work each week, setting one up is one of the most practical steps you can take after learning the basics.

What custom assistants and projects are

The main assistants each offer a version of the same idea, under different names:

• ChatGPT offers custom GPTs, which are saved assistants with their own instructions and files, and Projects, which group related chats with shared files and instructions.
• Claude offers Projects, where you add instructions and a knowledge base of documents that every chat in the project can draw on.
• Gemini offers Gems, which are saved assistants with custom instructions, and depending on your plan, reference files.

The details differ, and these features change regularly, so check the vendor's help pages for how yours works. The principle is the same everywhere: stored instructions tell the AI how to behave, and reference files give it facts about your membership to work from.

When a setup is worth the effort

A custom setup earns its keep when you repeat a type of task with the same background. Good candidates include:

• A newsletter assistant that knows your voice, sections and regular features.
• A content helper that knows your course structure and lesson names.
• An events assistant that drafts announcements, reminders and follow-ups in a consistent format.
• A support drafting helper that knows your help articles and policies, used by you or your team to draft replies you then review.

Beatriz runs a community for freelance translators. She set up one project for her newsletter and another for course materials. Before, she spent the first few minutes of every session re-explaining that her members are experienced professionals who dislike being talked down to. Now that is simply built in.

Write the instructions like a job description

Stored instructions work best when they read like a clear brief to a new team member, covering background, audience, tone and rules. If you already have a standard briefing from briefing AI like a new assistant, it makes an excellent starting point. Here is a template you can adapt:

You help me create content for [membership name], a membership for [describe members]. Members join to [main reasons]. Write in a [describe tone] voice, using US spelling, short paragraphs and plain language. Never use these words or phrases: [list]. Never promise [results, refunds, response times or anything else you do not promise]. Use the attached files as your source for facts about the membership, including course names, event formats and policies. If a request needs information that is not in the files, say so and ask me rather than guessing. When you finish a draft, list anything I should double-check before publishing.

The last two lines matter most. They reduce invented details and remind you that a person reviews every draft before members see it.

Choose reference files carefully

The files you attach become the assistant's picture of your membership, so choose them with care:

• Include: your style guide, a short description of each level or program, a list of course and lesson titles, your public FAQ and policies, and two or three examples of your best writing.
• Keep current: an outdated price or retired course in a file will keep reappearing in drafts. Note the files in your prompt library or planner and update them when things change.
• Leave out: members' personal data, support histories, payment records and anything confidential. Stored files persist, which makes this even more important than in a single chat.

If your team will use the setup, a paid business plan usually offers stronger data commitments and admin controls; free or paid AI plans explains what changes.

Test it before you rely on it

Once your assistant or project is set up, test it with questions where you know the answers:

• Ask about a course or policy that is in the files, and check the answer is accurate.
• Ask about something that is not in the files, and check it says so instead of inventing an answer.
• Ask for a draft and check it follows your tone and rules.
• Try a tricky request, such as a refund question, and check it stays within the limits you set.

Beatriz found her first version kept inventing webinar dates. Adding “never state a date unless it appears in the files” fixed it. Expect a couple of rounds of adjustment.

Keep it internal, and keep a person in the loop

These setups are working tools for you and your team. Resist the temptation to share a custom assistant with members as a quick chatbot: a member-facing assistant needs much more testing, clear limits and a way to reach a person, and it deserves its own careful plan. Internally, the rule stays the same as for any AI output: someone reads and checks every draft before it reaches a member.

Set up your first one

1. Pick the one task you repeat most often with the same background.
2. Write instructions using the template above, and add your review rule.
3. Gather three to five reference files, with any personal data removed.
4. Test it with five questions you know the answers to, and adjust.
5. Save your instructions in your prompt library so you can rebuild the setup in another tool if you ever switch.

Keeping your source documents organized helps beyond AI, too. If your procedures and policies are scattered, an internal wiki gives you and your team one place for the knowledge your assistant draws on.