Skip to content

Custom GPTs Course: Build Your Own Assistant

A custom GPT course should teach task choice, instructions, safe reference files and testing. How Truthshine's M1 builds GPTs, Gems and Claude Projects.

Blog · Prompting & AI Tools By Truthshine Editorial Team Updated 5 min read

Quick answer: A custom GPT course is worth it if it teaches you to choose the right task, write instructions that keep the assistant in scope, decide which files it may see and test it before colleagues rely on it. Truthshine's M1 spends its longest unit, 5 hours, on custom GPTs, Claude Projects, Gemini Gems and no-code workflows.

Key takeaways

  • The best tasks for a custom assistant are repeated, rule-based and checked by a person before anything goes out.
  • Build in the assistant your team already uses and is allowed to use: a custom GPT, a Claude Project or a Gemini Gem.
  • Good instructions have four parts: role, scope, tone and limits, including what the assistant must refuse.
  • Test with at least ten realistic questions, including awkward ones, and fix every off-topic reply before sharing.
  • M1's final project is a tested assistant or no-code workflow with a user guide, plus a 15-prompt library for your role.

Anyone can click through a GPT builder in ten minutes. Building a custom assistant that a team can rely on takes more: a well-chosen task, instructions that keep it in scope, reference files that are safe to share and a round of testing before anyone else uses it. Truthshine's M1, Prompting & AI Tools spends its longest unit, five hours, on exactly this, and the final project is an assistant you build and test for your own team.

Which tasks deserve a custom assistant?

Start here, because a good assistant on the wrong task still wastes time. The test is simple: is the task repeated, are the rules written down, and does a person check the output before it goes anywhere?

Good fit Poor fit
First drafts of replies to common customer questions, using an approved FAQ One-off tasks you will never repeat
Turning rough meeting notes into your team's standard minutes format Decisions about people, such as shortlisting candidates or approving credit
Product or service descriptions in a written brand voice Anything that needs live data the assistant cannot see, such as today's stock levels
Checking a draft against a house style guide Tasks that would need client or employee personal data to work
Explaining an internal process to new joiners from your SOP document Legal, medical or tax conclusions that a qualified person must make

M1's final unit begins with this choice, because the rest of the build depends on it.

Custom GPT, Claude Project or Gemini Gem: which should you build?

Build in the assistant your team already uses and is allowed to use. The skills are the same across all three, and switching later is mostly a matter of copying instructions across.

Option Lives in Choose it when
Custom GPT ChatGPT Your team works in ChatGPT and has a plan that allows building GPTs
Claude Project Claude Your team drafts long documents in Claude and wants shared instructions and files
Gemini Gem Gemini, in Google Workspace or a personal account Your organisation runs on Google Workspace
No-code workflow Zapier, with Google Sheets The task starts in a form or sheet and should end with an AI-drafted step in another app

The M1 FAQ notes that some features, such as building custom GPTs, need a paid plan. Plans and limits change often, so check the provider's current plan page before you pay for anything.

How do you write instructions that keep it on task?

Use four parts, and be specific in each. Here is a short example for a D2C brand's customer-reply assistant:

  • Role: You draft first replies to customer emails for a Jaipur-based handloom brand. A team member reviews every reply before it is sent.
  • Scope: Answer only questions about orders, delivery times, returns and fabric care, using the attached FAQ. For anything else, say that a colleague will reply within one working day.
  • Tone: Warm, brief and plain. Use the customer's first name. No sales pitch in a complaint reply.
  • Limits: Never promise a refund, a discount or a delivery date the FAQ does not state. Never ask for card, UPI or bank details. If the customer mentions a health issue or a legal threat, flag it for a manager.

Notice how much of the work sits in the limits. Most failures in custom assistants come from what the instructions forgot to rule out.

How do you test it before your team uses it?

Treat testing as part of the build, not an afterthought. In M1 you:

  1. Write at least ten sample questions, including three that should be refused and two that are deliberately vague.
  2. Run each one and record what the assistant did.
  3. Fix off-topic or overconfident replies by tightening the instructions, then test again.
  4. Write a short user guide: what the assistant is for, what it must not be used for, and who maintains it.
  5. Set sharing permissions so only the right people can use or edit it.

Upkeep matters too. When your FAQ or prices change, the assistant's files must change with them, and someone has to own that job.

What must an assistant never see?

Anything you would not put in a public document. Reference files should hold approved, non-confidential material: FAQs, style guides, published policies, product information. Client lists, employee records, unpublished financials and personal data do not belong there. M1 teaches this before you build, and India's Digital Personal Data Protection Act, 2023 is one reason the rule is strict.

What do you leave the course with?

M1's final project is an AI toolkit for your role: a tested library of at least 15 prompts for work you do every week, and one custom GPT, Claude Project, Gemini Gem or no-code workflow for a task your team repeats. Each comes with its purpose, example outputs, a checking step and notes on the data it must never see. Faculty evaluate it against a published rubric. You can also use your own business, or your employer with written permission, as the setting for the project.

When should you go further?

When you want the assistant to act, not only answer. Agents that read a sheet, draft a reply and wait for approval are taught in M5, Email, WhatsApp, CRM & AI Agents, a five-week module using n8n and Zapier with approval steps, limits and logs. The no-code AI apps course guide sets out what you can build at each level, and this guide to marketing automation and AI agents covers M5 in more depth.

What does it cost?

M1 is ₹7,999 including GST for 20 live hours over four weeks, starting 16 November 2026. You can spread it as ₹2,667 a month for three months on 0% interest EMI, subject to approval. A certificate awarded by Medhavi Skills University (UGC-recognised) follows when you pass, and a full refund is available if you ask before the cohort starts or within 7 days of the start date. Details are on the fees page.

What to do next

  1. List three tasks you repeat each week and score each against the good-fit test above.
  2. Read the custom assistants unit on the Prompting & AI Tools page.
  3. Join the free masterclass AI Agents for Marketers: A Live Build on 10 December 2026 to see where assistants end and agents begin.

Frequently asked questions.

Answers from our published policies and syllabi.

Building custom GPTs needs a paid plan, and plans change often, so check OpenAI's current plan page. In M1 you can build the same kind of assistant as a Claude Project or a Gemini Gem if that is what your workplace uses.

No. A custom assistant answers within instructions you set. An agent plans steps and uses tools, such as reading a sheet and sending a message, which needs approval steps and logs. Agents are taught in M5 with n8n and Zapier.

Only files your organisation allows and that contain nothing confidential or personal. M1 teaches you to decide what reference files an assistant may use and what data it must never see, before you build.

No. Custom GPTs, Claude Projects and Gemini Gems are set up with plain-English instructions, and the Zapier workflow in M1 is no-code.

Keep deciding

The future runs on AI. So should your career.

Learn live with practitioner faculty, build real work, and earn a certificate awarded by Medhavi Skills University.

0% interest EMI · 7-day refund window · 5-Year Future-Proof Pass