Are You Prompting Wrong?
How should I prompt? It’s probably the most frequently asked question I receive at my workshops.
How should I prompt? It’s probably the most frequently asked question I receive at my workshops.
The internet is awash with gurus offering “secret rules” for generative AI prompting. Some say prompts must be poetic, others insist on magic numbers of words, and more than a few sell courses promising to make you a “prompt engineer” overnight.
I’ve tested dozens of these supposed hacks. Some work sporadically, many don’t, and almost all are overcomplicated. Strangely enough, what consistently works best comes from Google’s own recommendations in their Prompting Guide. And unlike the noise online, these rules are simple, flexible, and work across most large language models (LLMs).
Think of prompting as starting a conversation. The clearer you are, the better your AI partner responds. And at the heart of it, four elements matter most:
Persona
Task
Context
Format
Let’s unpack each — with examples of what to do (and what not to do).
1. Persona: Who’s Talking?
Telling the AI who it should be is surprisingly powerful. It shapes tone, depth, and perspective.
✅ Good practice:
“You are a travel journalist. Write a short, witty review of a boutique hotel in Barcelona aimed at food lovers.”
❌ Bad practice:
“Tell me about a hotel in Barcelona.”
The first gives a role and an audience. The second is vague — you’ll likely get generic TripAdvisor copy.
2. Task: What Should It Do?
The task is the verb — summarise, brainstorm, draft, critique. Without it, AI flails.
✅ Good practice:
“Draft an executive summary email highlighting three risks from this report. Keep it under 200 words.”
❌ Bad practice:
“Explain this report.”
The clearer the verb, the sharper the output.
3. Context: What’s the Situation?
Context is where AI stops guessing. A few details transform a vague answer into something relevant.
✅ Good practice:
“I am planning a three-day offsite in Manchester with 20 marketing managers. Suggest icebreakers that build trust, fit indoors, and take no more than 30 minutes.”
❌ Bad practice:
“Suggest team activities.”
Notice how the details (Manchester, team size, time, indoor) constrain the AI into something useful.
4. Format: How Should It Look?
Ask for a structure — bullets, tables, scripts, headlines — and AI will oblige.
✅ Good practice:
“Summarise this 20-page report in five bullet points and a one-line conclusion for executives.”
❌ Bad practice:
“Summarise this.”
If you don’t ask for the shape, you’ll get freeform paragraphs that might not be easy to use.
Putting It All Together
The real magic happens when you combine all four:
“You are an HR manager. Write a two-minute welcome speech for new hires based on this mission statement. Use a warm, professional tone and finish with a call to action.”
Here, persona, task, context, and format align. And it works — in Docs, ChatGPT, Gemini, Claude, you name it.
Good vs Bad Prompting Habits
Good practices:
Use natural language — write like you’d brief a colleague.
Be specific — names, numbers, goals help.
Iterate — refine with follow-up prompts.
Review — AI outputs need a human eye.
Bad practices:
One-word prompts (“Resume”).
Overcomplicating with jargon.
Expecting a perfect answer on the first try.
Forgetting the audience (who will read the result?).
Final Thought
Different LLMs have their own quirks — some prefer shorter prompts, while others thrive on longer briefs. But across the board, persona, task, context, and format are the bedrock. Forget the gimmicks. Stick to these four, and you’ll stop prompting wrong — and start prompting smart.
The Google Guide is available here:
