The RTF Prompting Framework: How Marketers Get Reliable Results From Claude and ChatGPT

When a marketer tells me "AI gives me generic output", I ask to see the prompt. It is almost always one line: "write a caption for my new product". The model did exactly what it was asked, which was not much. The fix is not a better model. It is a better brief, and the simplest brief that works every time has three parts: Role, Task, Format. We call it RTF and teach it in the AI week of the Professional Digital Marketing Associate Diploma. In the AI Marketing Professional Certificate we extend it to RCTF, adding Context as its own step, and then show how to stop writing most of it by hand.

Key takeaways

• Every reliable prompt answers three questions: who is the model acting as, what exactly should it do, and what should the output look like.
• Context is the fourth ingredient. Paste the brand voice, the audience and the facts. The model cannot know your business.
• Ask for options and constraints, not a single perfect answer. Then edit. Editing a good draft is fast; fixing a generic one is slow.
• Save prompts that work as reusable templates, one per recurring task.

The framework

Role. Tell the model who it is for this task. "You are a senior copywriter for a Beirut skincare brand that sells to women aged 25 to 40." Role sets the expertise and the vocabulary.

Task. State exactly what to produce and what to avoid. "Write five Instagram caption options for the launch of a vitamin C serum. Each under 40 words. No exclamation marks. Do not mention discounts."

Format. Say what the output should look like. "Number the options. After each, one line on the angle it takes." Format makes the output usable without reformatting.

Put together:

You are a senior copywriter for a Beirut skincare brand that sells to women aged 25 to 40. Write five Instagram caption options for the launch of a vitamin C serum. Each under 40 words. No exclamation marks. Do not mention discounts. Number the options and, after each, add one line on the angle it takes.

That prompt produces five distinct, on-brief captions. "Write a caption for my new product" produces one forgettable paragraph.

Add context, and the output stops being generic

RTF gives structure. Context gives specificity. That is why the AIMP version of the framework is RCTF: Role, Context, Task, Format. Before the task, give the model what it cannot guess:

• Brand voice: three adjectives and two example posts you like.
• Audience: who they are and what they worry about.
• Facts: the product details, the price, the offer, the claims you are allowed to make.
• Constraints: words to avoid, legal lines, length.

In Claude, put this material in a Project so every conversation starts with it. In ChatGPT, use custom instructions or a saved prompt. Either way, you write the context once.

Once that is set up, most of the framework disappears from your typing. The Project holds the Role and the Context, a Skill holds the Format, and all you write is the Task. That is the point of learning the full structure first: you know what the tools are doing for you.

Worked examples

A welcome email

Role: You are an email marketer for a boutique gym in Dubai. Task: Write the first email of a welcome sequence for people who just booked a free trial. Goal: get them to show up. Warm, direct, no hype. Under 150 words. Format: Subject line, preview text, body, one button label.

Ad copy variations

Role: You are a performance marketer running Meta ads for an online course. Task: Write eight primary-text variations, each using a different angle: problem, outcome, social proof, curiosity, objection, urgency without discounts, story, comparison. Format: A table with columns Angle, Text, Suggested headline.

A content brief

Role: You are a content strategist. Task: Turn the following customer questions into a brief for a 1,200-word article: target reader, working title, H2 outline, key takeaways, and the one action the reader should take at the end. Format: Markdown headings.

Three habits that make RTF work

1. Ask for multiples. Five options and you pick, rather than one and you argue with it.
2. Iterate in the same conversation. "Option 3, shorter, and make the first line a question." The model keeps the context.
3. Judge, do not accept. The output is a draft from a very fast junior. Your job is the taste and the facts.

Where this leads

Once RTF is automatic, the next level is turning prompts into reusable assets: Projects that hold your role and context, Skills that hold your formats and procedures, and Connectors that put the model inside the tools you already use. That progression, from a single good prompt to a personal AI stack, is the spine of our AI Marketing Professional Certificate.

Frequently asked questions

Does RTF work the same in Claude and ChatGPT? Yes. The framework is about the brief, not the model. Interfaces differ in where you store context, the prompt structure is identical.

How long should a prompt be? As long as the brief needs. A good prompt for a campaign can be a page. A good prompt for a subject line can be two sentences with the context already loaded.

What about image and video models? Same idea, different vocabulary: subject, style, composition, constraints. We cover that separately in the visual modules.