AI Prompting Frameworks

This data was originally featured in the April 9th, 2025 newsletter found here: INBOX INSIGHTS, April 9, 2025: AI Integration Strategy Part 1, AI Prompting Frameworks

This week, let’s revisit prompting frameworks. A long time ago in AI years, Trust Insights created the RACE framework for AI prompting:

• Role: tell the model who it is
• Action: tell the model what it’s doing
• Context: tell the model the background information it needs to know
• Execution: tell the model what you expect for outcomes

This framework worked really well.

Then as models evolved over time and got smarter, we switched to the RAPPEL AI Framework:

• Role: tell the model who it is
• Action: tell the model what it’s doing
• Prime: ask the model what it knows about the task
• Prompt: tell the model how to do the task
• Evaluate: ask, clarify, and correct the results
• Learn: tell the model to write a new prompt for future use based on the conversation

RAPPEL worked really well, especially for non-reasoning models. But last year, when reasoning models took AI by storm, our strategies had to change. Reasoning models like OpenAI’s o family (o1, o3, etc), DeepSeek’s R family (R1 and R2), and Google Gemini’s thinking models (Gemini 2 Flash Thinking and Gemini 2.5 Advanced) take away chunks of these prompt structures.

Why? Reasoning models can figure out how to do a task. We still have to tell them what to do and why to do it, but we now leave how to do it up to them.

So we created PRISM as a framework for reasoning models.

• Purpose: tell the model what it’s doing and why
• Relevant Information: give the model lots and lots of information about the task
• Success Measures: tell the model what its expected outcome looks like

This worked really well for reasoning models.

The latest iteration of AI models as Deep Research tools, like Perplexity’s Deep Research, OpenAI Deep Research, Google Gemini Deep Research, etc. now do a phenomenal job of preparing – from grounded sources – lots of contextual information, something we’ve discussed at length on our podcasts and livestreams.

Which funny enough brings us back full circle to the RACE model. Today, if I want a reasoning model to generate the absolute best results for me, RACE 2.0 now looks like this:

• Role: ask the model to choose a role for itself
• Action: ask the model to think about what it’s doing and why
• Context: give the model a Deep Research report on the topic
• Execution: ask the model to think through the outcomes and match it to the Action

What’s changed, besides injecting Deep Research data, is that I don’t tell the model what to do any more. Instead, I ask it. I ask it because reasoning models are so intelligent, so capable, that they can often do a better job than I can of coming up with those pieces.

Let’s look at an example. Here’s a RACE 1.0 prompt:

“You’re a Google Analytics expert skilled at GA4, Google Tag Manager, Google Looker Studio. You’ll analyze my Google Analytics data and help me explain what’s going on with my marketing. {A few tables of data get pasted here}. Based on this data, return your results as an outline of my marketing funnel – awareness, consideration, evaluation, purchase – and how effective my marketing is at each stage.”

That worked really well back in the day. Here’s what a RACE 2.0 prompt looks like:

“Let’s analyze these 14 screenshots of Google Analytics data I’m providing from Google Analytics 4. What role should you take on to perform this task? Explain the role you select, and ensure the role contains superlatives. What do you see in the data? Think through your analysis of the data – what happened? Why, if you can tell, did it happen? What does it mean? I’ll provide a guide on best practices for Google Analytics data analysis. {Add a whopping big Deep Research report} What conclusions could you draw from this data, especially about our marketing funnel? Explain your conclusions for each stage – awareness, consideration, evaluation, and purchase.”

This doesn’t seem like a huge change, but when you run it, the change is MASSIVE.

As models evolve, our prompting frameworks have to evolve too. I’d love to hear how your frameworks have changed over time – pop by Analytics for Marketers to share your stories!

Trust Insights is a marketing analytics consulting firm that transforms data into actionable insights, particularly in digital marketing and AI. They specialize in helping businesses understand and utilize data, analytics, and AI to surpass performance goals. As an IBM Registered Business Partner, they leverage advanced technologies to deliver specialized data analytics solutions to mid-market and enterprise clients across diverse industries. Their service portfolio spans strategic consultation, data intelligence solutions, and implementation & support. Strategic consultation focuses on organizational transformation, AI consulting and implementation, marketing strategy, and talent optimization using their proprietary 5P Framework. Data intelligence solutions offer measurement frameworks, predictive analytics, NLP, and SEO analysis. Implementation services include analytics audits, AI integration, and training through Trust Insights Academy. Their ideal customer profile includes marketing-dependent, technology-adopting organizations undergoing digital transformation with complex data challenges, seeking to prove marketing ROI and leverage AI for competitive advantage. Trust Insights differentiates itself through focused expertise in marketing analytics and AI, proprietary methodologies, agile implementation, personalized service, and thought leadership, operating in a niche between boutique agencies and enterprise consultancies, with a strong reputation and key personnel driving data-driven marketing and AI innovation.