How to Prompt Like a Pro: Lessons from Google’s AI Training

AI / By

Most people think using AI is as simple as typing a question and waiting for an answer. That’s true—AI will give you something back. But if you really want to squeeze the highest quality respones out of AI, the difference really comes down to one thing: prompting. 

 

I recently combed through an eight-hour Google training on prompt engineering so you don’t have to. The course was dense, but packed with practical frameworks, tips, and tricks to get better results out of AI every time you use it. And here’s the good news: you don’t need to be a developer or data scientist to apply these lessons. Once you understand the structure, anyone can learn to “talk to AI” in a way that feels like you’ve unlocked cheat codes. 

The Five-Step Prompt Framework

The foundation of Google’s training is a simple five-step framework: Task, Context, References, Evaluate, and Iterate. Think of this as your roadmap for every AI interaction. 

  1. Task – Clearly define what you want the AI to do. Vague prompts like “help me with marketing” aren’t enough. Instead, say, “create a social media post about our new weekend hours.” 

  2. Context – The more detail you provide, the better the results. Instead of just asking for a post, you might add that the business is a local café, the audience is young professionals, and the vibe should be upbeat. 

  3. References – Examples are fuel for AI. If you want a specific tone, provide a sample post you liked in the past. If you want visuals, link or describe them. 

  4. Evaluate – Don’t accept the first draft blindly. Ask yourself: does this actually fit what I need? 

  5. Iterate – Great prompts are rarely “one and done.” Adjust, refine, and test until you get what you want. Google calls this ABI—Always Be Iterating. 

Mnemonic devices make this easier. The course had its own, but I prefer a playful version: Turtle Carry Really Enormous Igloos. Silly? Yes. Memorable? Definitely. 

Four Ways to Fix a Bad Prompt

Sometimes, even when you follow the framework, AI still misses the mark. That’s where iteration comes in. Google teaches four proven methods to get better results: 

  1. Revisit the Framework – Add missing elements like persona, context, or references. For example, instead of asking for a “sales pitch,” you might ask, “act as a luxury car sales expert pitching to a young professional audience. Keep it short and persuasive.”

  2. Simplify – Long-winded prompts can overwhelm AI. Break them into smaller steps. For instance, first ask for a headline, then ask for supporting copy.

  3. Try Analogies – If the results feel flat, switch perspectives. Instead of “write a marketing plan,” ask “write a story about how a customer discovers and loves our product.” Storytelling prompts often bring more life into the output.

  4. Add Constraints – Unlimited freedom leads to generic results. Boundaries force creativity. For example, instead of “make me a playlist,” say “make me a playlist of only 90s rock songs under three minutes.” 

 Together, these strategies make iteration less frustrating and more productive. 

Beyond Text: Multimodal Prompting

Prompting isn’t limited to text anymore. Many AI tools can handle images, audio, video, and even code. This is called multimodal prompting. 

Here’s how it works: 

  • Upload a product photo and ask AI to write a social post around it. 
  • Share a chart and ask for insights in plain English. 
  • Provide your logo and colors, then ask AI to design a flyer in your brand style. 

 The trick? Be explicit about the input and the desired output. “Here’s an image of our new menu item. Write a playful Instagram caption that highlights it as a limited-time special.” That level of specificity helps AI nail the assignment. 

Handling AI’s Weak Spots

As powerful as AI is, it’s not flawless. Google emphasized two big pitfalls: 

  1. Hallucinations – AI sometimes makes things up. Always verify facts. 

  2. Biases – Because models are trained on human content, they can reflect human biases. 

The best practice is to keep a “human in the loop.” Use AI for ideas, drafts, or analysis, but double-check before publishing or presenting. 

Everyday Use Cases That Save Time

The course walked through dozens of everyday work tasks that AI can handle. Here are a few adapted examples: 

  • Emails: Instead of spending 15 minutes drafting a policy update for your staff, let AI create a polished draft in under a minute. Add your edits, hit send. 
  • Spreadsheets: Not great with formulas? Ask AI: “How do I calculate average sales per customer in this dataset?” It can generate the formula and explain how it works. 
  • Presentations: Feed AI your notes and ask it to generate slide headlines, talking points, or even design ideas. 

Each of these examples frees up mental bandwidth for the work that actually requires your creativity and judgment. 

Advanced Prompting: Chaining, Thought Processes, and Trees

Once you master the basics, you can level up with advanced techniques: 

  • Prompt Chaining – Break a big project into a sequence of smaller prompts that build on each other. For example, start by asking for a summary of customer survey responses, then use that summary to generate marketing ideas, then refine those ideas into a campaign outline. 
  • Chain of Thought – Ask the AI to show its reasoning step by step. This is helpful for decision-making or problem-solving tasks, like asking it to walk through the logic of a pricing strategy. 
  • Tree of Thought – Instead of one straight path, ask AI to explore multiple branches of ideas at once. For example, “brainstorm three different ad campaign directions, each with a unique theme and audience approach.” 

 The magic happens when you combine these methods. Imagine asking AI to generate multiple campaign options (tree of thought), explain the reasoning behind each (chain of thought), and then refine the best one through multiple iterations (chaining). That’s pro-level prompting. 

Creating Your Own AI Agents

The final module in Google’s course introduced AI agents—customized setups where you define a persona, context, rules, and stop conditions. Think of it as building a specialized consultant inside your AI tool. 

 

Examples: 

  • A training agent that roleplays interviews with interns and gives feedback. 
  • A feedback agent that critiques your pitch as if it were a skeptical client. 

Designing an agent follows the same five-step framework, but with more emphasis on persona and context. It’s like giving AI a job title and clear boundaries before asking it to work.