How to use AI effectively: prompting, tools, and future-proofing

Executive overview

Most people give AI almost no context and expect great results. The quality of AI output is entirely determined by the quality of the input. A four-part prompt structure — role, context, command, format — closes the gap between generic and useful output.

The output will never exceed the quality of the input.

The four-part prompt structure

• Role: tell AI who to be (e.g. "act as a world-class conversion strategist for software")
• Context: give it all relevant background — documents, transcripts, specs
• Command: be explicit; make the implicit explicit
• Format: specify output format — bullets, table, CSV, PDF — or give it a template to follow

Pull vs push prompting

• Push prompting: you do 80% of the work, AI finishes the last 20%
• Pull prompting: give AI the outcome, let it figure out how to get there
• Pull structure: role + context → outcome-based objective → "ask me all the questions you need to create this"
• Answer its questions using voice-to-text; let AI do the heavy lifting
• Iterate: "ask me more questions to refine this"
• This is how developers use spec-driven development — now available to everyone

Picking and mastering one tool

• Switching between tools constantly is procrastination
• Masters go deep before going wide; learn one tool fully, the others become easier
• Claude: best for writing, deep thinking, code
• Gemini: best for research, up-to-date information, Google Workspace users
• ChatGPT: most integrations, widest adoption

Master prompts

• A master prompt is a manual about a role in your life (e.g. "Dan as CEO", "Dan as father")
• Gives AI persistent context so outputs are personalised, not generic
• Build one using pull prompting: "I want to create a master prompt for my role as X — ask me all the questions you need"
• Refine conversationally; save as a PDF for reuse across tools
• Every team member should have one

System prompts

• A system prompt is a reusable recipe: define the structure once, run it on repeat
• Build using pull prompting: "You're an expert AI engineer — create a system prompt that does X, ask me all the questions you need"
• Once refined, copy it into a custom GPT, Claude Project, or Gemini Gem
• Turns complex AI workflows into simple tools anyone on your team can use
• Real-world examples: leaked system prompts from Perplexity, Notion, Lovable — searchable on GitHub

Future-proofing: taste, vision, and care

• Machines optimise what exists; humans imagine what doesn't
• Taste: consume the best in your field — follow top creators, read newsletters, train your sense of quality
• Vision: schedule thinking blocks; visualise; read widely to expand your sense of what's possible
• Care: use AI to eliminate boring, dangerous, dirty work so you can invest time in people and relationships
• Human connection is irreplaceable — events are selling out because people crave it