# Turn one repeatable task into an assistant: build your first Custom GPT in 30 minutes Source: https://getstartedwithai.net/tutorials/build-first-custom-gpt Site: Get Started With AI Tagline: Build AI automations that actually run your business. Audience: solo professionals, freelancers, consultants and AI implementers in training Contact: hello@getstartedwithai.co.uk This Markdown companion is generated from the prerendered page for AI crawlers. # Turn one repeatable task into an assistant: build your first Custom GPT in 30 minutes Pick a task you do every week with small variations and you finish this build with a published Custom GPT that does that task for you, your team or your clients. Discovery call summariser, weekly status report, proposal scaffold, the kind of work that takes 45 minutes and could take 5. Total time today: 30 minutes. Three real worked examples of input and output are what makes the build land. Those samples teach the GPT your shape and your voice without needing a 1,500 word system prompt. ## Pick the task before opening the builder A bad Custom GPT is a generic assistant trying to do five things. A useful Custom GPT does one repeated task very well. Choose before you build. - Email replies in your voice for a single client. - Proposal scaffolding for a single service line. - Meeting notes turned into action items in your standard task format. - Weekly client report drafted from raw notes. - Discovery call summary in your standard intake format. ## What you need before the build - ChatGPT Plus, Pro or Business. Custom GPT building is not on Free. - Three real examples of the task done well: input plus the output you would have shipped. - A short brief of the rules: tone, format, what to never include. ## Step 1: Open the builder (2 minutes) 1. Click the GPTs button in the sidebar, then My GPTs, then Create. 2. Two tabs appear: Create (a chat-based builder) and Configure (manual fields). Use Configure for a real build. The chat builder is fine for prototyping but produces vague instructions. 3. Name the GPT after the task, not the tool. "Discovery Call Summariser" beats "My Assistant". ## Step 2: Write the four-block instructions (10 minutes) The instructions field is the system prompt that runs invisibly before every conversation. Use four blocks in order: Role, Inputs, Process, Output. 1. Role: one paragraph. Who the GPT is, who it works for, what it never does. Example: "You are the discovery call summariser for a B2B operations consultant. You read raw notes and produce a structured intake brief in the writers voice." 2. Inputs: list what will be pasted in. Example: "User pastes raw meeting notes (5 to 30 minutes worth). Sometimes a transcript. Sometimes a list of bullet points." 3. Process: 3 to 6 numbered steps the GPT must follow on every input. Identify the prospect, pull out the top three pain points, list any tools mentioned, capture the budget cue, format the result using the Intake Brief template. 4. Output: the exact format. Section headings, length, sign-off. Be specific. ## Step 3: Upload three working samples (5 minutes) A Custom GPT learns your voice from examples. Three completed jobs are enough for the GPT to copy your shape, sentence length and word choice without being told. - Three completed examples (input plus your finished output). - A short style note (one page) listing words you use, words you avoid, sentence length target. - Any reference template the output must match (an Intake Brief, a Proposal Scaffold, a Status Report). ## Step 4: Configure the rest (3 minutes) - Conversation starters: three example prompts users can click. Make them realistic. - Capabilities: turn on Web Search only if needed; turn off Image Generation unless the task requires images; leave Code Interpreter and Data Analysis on for any data work. - Profile picture: any logo or icon. Visual matters less than instructions. ## Step 5: Test against three real cases (10 minutes) Do not save and forget. Run the GPT against three real inputs you have already done. The output should match what you would have shipped. 1. Open a chat in the GPT preview pane. 2. Paste real input number 1. Read the output. If it misses anything, edit the instructions and rerun. 3. Repeat for inputs 2 and 3. By case three the output should be 80 percent ready, with only minor edits before sending. 4. If the output is below 80 percent at case three, the instructions need a Process step or a clearer Output rule. Iterate before publishing. ## What you ship A published Custom GPT with its own URL. Three test runs that pass at 80 percent ready. A reusable Role, Inputs, Process, Output template you can spin up in 15 minutes for the next task. ## What to build next Pair the GPT with a Project (one client, one Custom GPT) for the strongest setup. Then pick the second most-repeated weekly task and run the same 30-minute build again. ## Related builds More tutorials in Core Assistants. Each one runs in 60 to 90 minutes and ships a finished asset. ### Replace 3 hours of admin a week: the 60-minute ChatGPT build 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/chatgpt-setup-solo-professionals) ### Stop losing client context: organise every engagement in ChatGPT Projects 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/chatgpt-projects-client-work) ### Cut research and drafting time in half: the Claude Project build 30 to 60 minutes. Solo Pro. [Open tutorial](/tutorials/claude-setup-solo-pro) ### Pick the right project container: Claude Projects vs ChatGPT Projects in practice 60 to 90 minutes. Solo Pro. [Open tutorial](/tutorials/claude-projects-vs-chatgpt-projects) ## Questions people ask before they start ### Can I share my Custom GPT with a client? Yes. Set sharing to Anyone with the link or Only people in my workspace. Clients need a ChatGPT Plus account to use it. For team-wide deployment, build inside ChatGPT Business. ### Will my uploaded files train the model? No. Files uploaded to a Custom GPT are not used to train OpenAI models. They are only available inside the GPT in conversation. On Plus, turn off all training in Settings, Data Controls. ### How is this different from a Project? A Project is a folder with files and instructions, used inside ChatGPT. A Custom GPT is a published tool with its own URL, shareable to others. Use Projects for your own client work; build Custom GPTs when the same task gets repeated by you or someone else. ### What does this sell for? £600 to £1,200 per GPT for a packaged build from the client own examples plus a 30-day check-in. Build the same shape for two or three clients in the same sector and the second build is half the time of the first.