|
Listen to this article
Getting your Trinity Audio player ready...
|
💡Key Takeaways :: Better AI outputs come from better inputs. The framework Google teaches is called T.C.R.E.I. — Task, Context, References, Evaluate, Iterate. Apply it and you’ll get dramatically more useful results from every AI tool you use.
What Is the T.C.R.E.I. Framework?
It stands for Task, Context, References, Evaluate, Iterate. Remember it as: Thoughtfully Create Really Excellent Inputs.
Taught as part of Google’s AI training programme, this framework applies across every generative AI tool — ChatGPT, Google Gemini, Claude, Microsoft Copilot, and beyond. Here’s how to use it right now.
Step 1: Specify the Task (+ Persona + Format)
Do this: Tell the AI exactly what to produce, who to act as, and what format you want.
- ❌ “Write me a speech”
- ✅ “Act as a professional speechwriter and give me a 30 minute speech on AI tools for my marketing presentation.”
The persona filters the AI’s expertise toward what’s relevant. The format shapes how the output is delivered. Both together mean far less back-and-forth.
Real example from Google’s training:
“You’re a movie critic that specialises in Italian film. Create a table containing the greatest Italian films of the 1970s, separated into genres like thrillers, dramas, and comedies. Provide a 100-word summary of each film plus director and release year.”
Notice how task, persona, and format are all in one prompt. That’s the standard to aim for.
Step 2: Add Context
Do this: Include the details that actually change the answer.
- ❌ “Give me presentation topic ideas”
- ✅ “Give me 5 presentation topics that will resonate with an audience of CPG digital marketers”
Context rounds out the AI’s understanding of what you’re really after – your goals, your audience, your constraints, and anything you’ve already tried that didn’t work.
Real example from Google’s training:
Without context:
“How was DNA discovered?”
With context:
“You’re a science expert developing a new curriculum at a local college. Tell me in a couple of engaging paragraphs how DNA was discovered and what impact it had on the world. Write it in a way that people unfamiliar with science would understand. Previous students found this course dry and unintelligible, so make sure it grabs attention and makes a great first impression.”
Same question. Completely different — and far more useful — output. Context can be the longest part of your prompt, and that’s a good thing. Don’t rush it.
Step 3: Provide References
Do this: Give the AI examples to work from — past speakers, best rated presentations, trending news, topical tools, presentation styles, or even content you admire. Two to five solid examples is enough.
But simply pasting in a reference isn’t enough. Structure it so the AI knows exactly what to do with it. Google’s training recommends three approaches:
Use transitional phrases to signal where your reference begins:
“Refer to these materials…” or “Base your response on the following examples…”
The more concrete your references, the closer the output lands on the first try.
Step 4: Evaluate the Output
Do this: Before using anything the AI generates, ask — is this accurate, sourced, relevant, unbiased and consistent?
Different AI models are trained on different data, and the same prompt can produce different results on different runs. That makes critical evaluation non-negotiable. A quick check now prevents real problems later.
If the output isn’t right – that’s completely fine. Because that brings us to the most important step.
Step 5: Iterate
Do this: If the output misses the mark, refine the prompt – don’t abandon it.
Google’s training puts it simply: Always Be Iterating (ABI). Add more detail, clarify the task, swap in a better reference. Prompting is less like giving an order and more like a conversation that sharpens with each round. This is where the magic happens.
Quick-Reference Cheat Sheet
| Step | What to Do |
|---|---|
| Task | State what you need + persona + format |
| Context | Add goals, audience, constraints + background |
| References | Include 2/5 examples, clearly labeled |
| Evaluate | Check for accuracy, relevance + bias |
| Iterate | Refine until the output is exactly right |
The order matters less than the intention. Be deliberate about what you put in — and what you get back will surprise you.
Based on the prompting framework taught in Google’s official AI Prompting Essentials training programme.
