Stop Guessing: The CLEAR Framework for Better AI Prompts (The Definitive Guide) - Digital Compliance Academy

I see it every day in our workshops. Smart, capable professionals staring at ChatGPT, Claude, or Gemini, typing something like:

“Write a marketing email for our new product.”

And then they look surprised when the AI spits out something generic, boring, and frankly, a bit rubbish. They sigh, edit it heavily (or delete it entirely), and tell their friends “Yeah, I tried using AI, but it wasn’t very good.”

The problem isn’t the AI. The problem is how you’re asking.

We assume that because these tools use natural language, we can just chat to them like we would a colleague. But your human colleague has shared context. They know your brand voice, they know the project history, and they know that “marketing email” actually means “a short, punchy teaser for the Q3 newsletter.”

The AI knows none of this. Unless you tell it.

At the Digital Compliance Academy, we teach a structured approach to prompting called the CLEAR Framework. We didn’t invent it (it’s a widely recognised standard used by experts across the industry), but we believe it is the most effective way to make sure you’re giving the AI everything it needs to do a good job.

This isn’t just a list of tips. It’s a reliable engineering structure for language.

Why Prompts Fail (The “Lazy Human” Theory)

Prompts usually fail for one of three reasons:

1. Vagueness: You assumed the AI knew what you meant.
2. Over-politeness: You padded the instruction with “pleases” and “could you possiblys” (see our post on stopping this).
3. Lack of Constraints: You didn’t tell it what not to do.

The CLEAR Framework solves all three.

---

C - Context

Start by telling the AI who it is, and what the background situation is. Don’t just ask for a task; set the scene. This is often called “Persona Priming.”

• Bad: “Write an email.”
• Better: “Act as a marketer.”
• The CLEAR Way: “You are an expert internal communications manager for a UK logistics firm. We are currently undergoing a digital transformation project that is causing some anxiety among staff. We need to communicate a new training initiative.”

Why this works: It switches the “mode” of the response from generic encyclopaedia to specific professional. It sets the vocabulary, the tone, and the assumed knowledge level.

L - Logic (and Layout)

Explain the steps you want it to follow. If you were delegating a task to a junior intern, you wouldn’t just say “do this.” You’d say “first read this, then draft a summary, then format it as a bulleted list.”

Most complex tasks fail because the AI tries to do the thinking and the writing at the same time. Split them up.

• The CLEAR Way: “Step 1: Summarise the key benefits of the training from the attached PDF. Step 2: Identify the top 3 concerns staff might have. Step 3: Draft an email that addresses those concerns using the benefits.”

Why this works: It forces “Chain of Thought” reasoning. By making the AI show its working, you reduce hallucinations and improve logical flow.

E - Expectations

This is where 90% of prompts fail. You need to be incredibly specific about the Output. What does the result looks like?

• Format: “Output as a Markdown table,” “Output as a Python script,” “Output as a JSON object.”
• Tone: “Professional but reassuring,” “Cynical and witty,” “Academic and dry.”
• Length: “Under 200 words,” “3 sentences max.”
• Language: “Use UK English spelling (e.g. ‘organise’, ‘programme’).”

The “Negative Constraint”: Crucially, tell it what NOT to do.

• “Do not use emojis.”
• “Do not use the words ‘delve’ or ‘cutting-edge’.”
• “Do not include a preamble or ‘Certainly! Here is the…’ text.”

A - Adjust

Here is the reality check: Your first prompt will rarely be perfect.

Treat AI like a collaborative process, not a vending machine. When you get the first result, don’t just accept it or reject it. Read it, see what’s wrong, and ask for an adjustment.

• Refinement: “That’s good, but the tone is too corporate. Rewrite the second paragraph to be more human. And you missed the point about the deadline—add that back in bold.”

This iteration loop is where the quality happens. Good prompters don’t write one perfect prompt; they have a good conversation.

R - Refine (The Human Layer)

The final step is the human layer. Never copy and paste directly from an AI tool.

You are responsible for every word you publish or send. AI is a tool for drafting and ideation, not for final execution. Take the AI’s best effort, and then:

1. Fact Check: Did it hallucinate a date or policy?
2. Tone Check: Does it sound like you?
3. Flow Check: Does the logic hold together?

---

Putting it all together: A Live Example

Let’s look at that original bad prompt again: “Write a marketing email for our new product.”

Here is the CLEAR version for a B2B SaaS launch:

(Context) You are a senior product marketer for ‘SecureFlow’, a B2B cybersecurity software company. We are launching a new ‘AI Safety’ module for enterprise clients.

(Logic) Please write a launch email to our existing customer base. Start by acknowledging the rising threat of “Shadow AI” in the workplace. Then introduce our module as the solution that enables safety without blocking innovation.

(Expectations)

• Tone: Helpful, authoritative, urgent (but not fear-mongering).
• Format: HTML-ready email with a subject line.
• Style: Short paragraphs. UK English. No buzzwords like “game-changer”.
• Call to Action: “Book a Demo”.

(Adjust) [After first output] “Make the opening punchier. It takes too long to get to the point.”

The difference in quality between the two outputs will be night and day.

Using the CLEAR framework doesn’t add hours to your work. It saves you hours of editing bad AI copy. It forces you to think clearly about what you actually want before you start typing.

And that—thinking clearly—is the one thing AI can’t do for you.