AI & The Art of Starting Over

The Puzzle That Keeps Changing

In my work with the Content Engineering students at Berghs, I see the same pattern repeat itself. They master AI tools, refine their prompting techniques, and apply the latest frameworks to optimize results. Then, suddenly, a new model update rolls out, and those once-groundbreaking techniques become obsolete overnight. Time to throw it all away and start fresh.

A prime example? The recent launch of deep research assistants from Perplexity.ai and OpenAI’s Deep Research. These tools have revealed a critical truth: traditional prompting methods are no longer enough when tackling complex research questions. We need a new framework for how to extract the best possible answers from AI.

Deep Research Demands a New Way of Thinking

Most people are used to AI responding to direct, simple prompts. But when it comes to deep research—sophisticated data analysis, contextual understanding, and critical insights—those methods fall short. Here’s how we need to rethink our approach:

Multi-Step Prompting is Essential

Instead of asking AI a single question and expecting a perfect answer, we must break down the inquiry into multiple steps:

• Step 1: Ask AI to generate a structured breakdown of key subtopics within your research area.
• Step 2: Dive deeper into each subtopic by requesting relevant studies, sources, and emerging trends.
• Step 3: Compare perspectives, analyze contradictions, and synthesize insights across different sources.

This process, known as multi-step reasoning, is key to avoiding shallow, generic responses.

Contextual Prompting is the New Standard

Most AI prompts exist in isolation. But for deep research, context is everything. To get better results, we must do the following:

• Instruct the AI to maintain a specific research angle rather than shifting directions.
• Provide custom datasets and reference materials (via RAG or vector-based databases).
• Utilize advanced AI tools like ChatGPT’s Advanced Data Analysis to refine, cross-check, and interpret findings.

Fact-Checking is Non-Negotiable

Unlike casual AI-generated text, hallucinations (fabricated information) can be disastrous in research. To mitigate this, do the following:

• Always request sources and citations.
• Push AI to critically evaluate its own answers by asking, “What are the counterarguments?”
• Cross-reference AI-generated insights with verified external data.

But how do we put all of this into practice? A structured approach makes all the difference. Enter R.A.G.E.—a framework designed to optimize AI-driven deep research.

R.A.G.E. – A Framework for AI-Driven Deep Research

• Reframe – Break down the question into smaller parts and explore multiple angles.
• Anchor – Provide AI with contextual data to reduce hallucinations.
• Guide – Use multi-step prompting and iterative questioning for deeper analysis.
• Evaluate – Cross-check AI’s responses against external, credible sources.

Learn this. Master it. But be prepared—in a month, a new methodology might surpass it. And suddenly, we’ll have to learn all over again.

Becoming experts at being beginners

The real challenge isn’t that AI is taking over jobs—it’s that we risk getting stuck in outdated ways of working. The ability to embrace continuous change is a superpower. Those who will thrive in the new era will be the ones who see themselves as lifelong learners, constantly diving into the unknown with curiosity that outweighs the fear of being wrong.

So, the next time you feel like you’ve finally mastered an AI workflow, only to have it rendered obsolete—congratulations. You’re exactly where you need to be.

Welcome to a world where your ability to start over, again and again, is your most valuable skill.

Written by Nina Amjadi