I was wasting Claude until I learned the RCCF method - MakeUseOf

AI has gotten dumb. That's the consensus these days, but I find this sentiment incredibly preposterous. Instead of using AI chatbots as our assistants, we've become overly reliant on them, stunting our thinking by feeding lazy prompts. Combined with the fact that security concerns have made these models much more reserved, the answers we get are rarely satisfactory.

Take Claude, for instance — I find it the most intelligent AI chatbot right now, but people still claim they're getting lackluster responses from it. The problem isn't with Claude itself, but with what we've been feeding it and expecting it to solve our problems, even though we're not providing it with the correct context and structure. That is, until I came across the RCCF method on Reddit, and it turned out that I had been using Claude wrong all along.

What is the RCCF method?

Structure integral to AI chatbots

AI models lack true cognitive abilities like those of humans, such as consciousness, goals, and personal experiences. Asking them questions or telling them to solve a problem without a clear intent will leave you overwhelmed with a cluster of raw, unstructured information. The problem arises when we expect AI to think for us, when it is really incapable of doing so. It can't conjure an original voice on your behalf without guidance, and if anything, it should be used to steer you in the right direction for your work.

This is where the RCCF method comes into play, and I think of it as a sort of role-playing with Claude, commanding it to act in a specific manner. Here's what it involves.

• Role: Define the function Claude should take up.
• Context: Explain the relevant information to the chatbot.
• Command: Instruct it with an order of what you might want.
• Format: Specify the layout in which you would want the answer.

Applying the RCCF method to Claude

The push method

Through the RCCF method, I assigned Claude the role of a SaaS product marketer for Notion. The context I described was to market to freelancers who often juggle several apps and meetings, so they have a platform to organize tasks, journal, and capture thoughts. As for the command, I asked Claude to write a positioning paragraph explaining how Notion would help with project tracking, automating mundane tasks, and keeping the workflow as simple as possible. Lastly, the format I asked Claude was to list it in around 130 words in a simple paragraph without bullet points.

Conversely, I sent Claude another prompt, this time plain as one would normally do.

Here's the result without the RCCF method. The difference is day and night, and you'll notice that without RCCF, Claude doesn't have a clear path and just lists generic and unfocused descriptions.

The pull method

Right above was an example of a push prompt in which the goals and the intent of my actions were clearly defined to Claude. However that may not be the case all the time. Creative blocks happen to me all the time, and sometimes I just get lost, not being sure of what to achieve with a specific task. In this case, I used the RCCF method, but this time with a pull prompt that nudges the AI to define the goals for me.

For the pull prompt, I asked Claude to once again write a short copy of Notion in a product marketing style, but for the command, I asked the chatbot to ask me key questions about the product, audience pain points, and also gaps in features, as well as competitor advantages, to paint a better picture for the end result I needed.

Claude then came back to me, asking me the tone I wanted for the paragraph, the target audience, and more questions that I hadn't thought of on my own. The biggest advantage you get with the pull prompt is that you understand your own work better, reduce the risks of an inaccurate response, and you won't get a generic answer.

How I use RCCF in my daily workflow

The RCCF method has helped me significantly throughout college, whether it was breaking down case studies a night before my exams or helping plan a workout split based on my goals. With each passing prompt, it becomes harder for AI chatbots to hold the original intent of the context drift. Without the RCCF method and aimlessly asking Claude to perform a task or help me out, the chatbot was just as confused as I was. Every follow-up prompt was me attempting to reestablish the original context, and that resulted in my efforts and token being wasted. The RCCF method takes just a single prompt for Claude to understand the purpose of my task, and if the answer still doesn't seem satisfactory, don't start again. Instead, use instructions to guide accordingly, or terms like "think deeply" and "research" to trigger a more thorough response.