Using AI Best Practice
by JimC · 40 things on NewTwos
- *How to prompt effectively**
- Give AI as much context as possible before asking your question; the more input, the better the output
- Structure every prompt with four elements:
- **Role** (who should AI act as),
- **Context** (background info and documents),
- **Command** (explicit instruction), and
- **Format** (how you want the output delivered)
- *Picking your tool**
- Choose one AI tool and master it before switching; jumping between tools is just procrastination
- Claude for writing, deep thinking, and code; Gemini for research and Google-integrated workflows; ChatGPT for integrations and broad use
- *Advanced prompting**
- Move from "push prompting" (you do the work, AI finishes) to "pull prompting" (you give the destination, AI finds the route)
- For pull prompting: give the role and context, state your outcome, then say "ask me all the questions you need to complete this" and answer them
- Use voice-to-text to answer AI's questions quickly rather than typing everything out
- *Master prompts**
- Build a "master prompt" for each major role in your life (e.g. pastor, COO, husband) covering everything relevant about that context
- Tell AI: "I want to create a master prompt for my role as [role name] - ask me all the questions you need to create it for me"
- Let AI generate its questions, then answer them conversationally using voice-to-text rather than typing
- Once AI gives you the draft master prompt, review it and refine anything that feels off - just talk to it naturally ("update this," "you missed that," "add this")
- Save the finished version as a PDF named something like "Master Prompt - Pastor" or "Master Prompt - COO"
- Upload that PDF at the start of any AI conversation so it has full context on you immediately
- Save it as a PDF so you can upload it to any AI tool instantly, giving it full personal context from the start
- *System prompts**
- When you get an output you love, build a system prompt around it so you can replicate that result consistently
- Use pull prompting to build the system prompt itself, then paste it into a Claude Project, custom GPT, or Gemini Gem
- Details:
- Master Prompt = who you are
- It's personal context about you, your roles, your life, your organisation. It helps AI understand you so answers feel tailored rather than generic. You upload it at the start of a conversation.
- System Prompt = how a specific task should always be done
- It's a set of instructions that defines a repeatable workflow. It's less about you and more about the job. You paste it into a Claude Project (or custom GPT etc.) so that every time someone uses that project, it already knows exactly what to do and how to do it.
- A concrete example for you:
- Your master prompt for your pastoral role might include who you are, your church context, your theology, your preaching style, your congregation in Barlanark.
- A system prompt might be something like: "Every time I give you a passage, produce a structured preaching outline following Explain/Illustrate/Apply, in a conversational Scottish pastoral voice, aimed at a working-class housing scheme congregation, around 4,200 words." You'd paste that into a Claude Project once, and every new conversation in that project starts with those rules already baked in - no re-explaining needed.
- *Future-proofing yourself**
- Develop your **taste** by consuming the best work in your field consistently
- Protect time for **vision** through thinking blocks, doodling, and reading broadly
- Invest in **care** by using AI to handle routine work so you can focus on genuine human connection
- *Getting started**
- Pick one daily task and hand it to AI today; 10 minutes of action beats 10 hours of tutorials