I’m willing to bet that almost everyone, myself included, has been using ChatGPT inefficiently. We often just throw prompts at the default model and hope for the best, but it turns out there’s a much smarter way to work.
I just stumbled upon this incredible video that completely changed how I approach ChatGPT’s features. The author, a savvy AI professional, cuts through all the confusing jargon and shows exactly when to use each tool. The mind behind it has a simple but powerful philosophy: the model you choose should depend on the complexity of the task, not the type of task. That insight alone is a huge unlock!
This expert’s core message is all about being strategic. Instead of using one feature for everything, we need to match the tool to the job. The video provides clear rules for when to use a powerful “reasoning” model versus a basic “chat” model, when to use ChatGPT’s web search over Google, and when to fire up advanced features like Deep Research and Canvas. It’s all about getting better results with less effort.
The Key Idea 🧠
The creator explains that OpenAI is rolling out updates so fast that it’s easy to miss the most important shifts in strategy. His guide is a masterclass in efficiency, showing how to move beyond simple prompts and start leveraging the platform’s specialized functions for complex work. By understanding the distinct purpose of each feature, you can stop wasting time and supercharge your outputs.
Here’s a deeper dive into the most actionable insights:
📌 Choosing Your Model: It’s About Complexity, Not Category.
The creator makes a fantastic distinction between the two main types of models. Reasoning Models (the latest ones with clean names, like the current GPT-4o) are your go-to for anything important or difficult. Think of them as the specialists you call in when the stakes are high. The contributor gives great examples, like asking it to “act as a nutritionist and create a vegetarian breakfast with at least 15g of fiber and 20g of protein.” This requires the model to think through multiple constraints. Another example is uploading a long, messy email thread and asking for a diplomatic reply. You’re trading a little speed for a much higher quality answer.
Basic Chat Models (older versions like GPT-3.5 or models with extra terms in their names) are perfect for quick, low-stakes tasks where speed is more important than nuance. For instance, asking “which fruits have the most fiber?” or turning a few bullet points into a simple email. The answer doesn’t need to be perfect. The key takeaway here is that defaulting to the most powerful reasoning model is usually the best bet unless you specifically need a lightning-fast response for a simple query.
Pro-Tips from the creator:
- Use Delimiters: Help the model by structuring your prompt. Put your instructions under a ### Task heading and the text you want it to analyze under a ### Document heading. This clarifies what to do versus what to analyze.
- Ditch “Think Step-by-Step”: This phrase helps basic models but actually hurts the performance of advanced reasoning models. They already think methodically, so adding it is counterproductive.
- Examples are Optional: Reasoning models are great with zero-shot prompts (no examples needed). Only add examples if you find the initial results are off the mark.
💡 Search Smarter: ChatGPT Search vs. Google.
This was a huge clarification for me. The creator’s rule of thumb is simple and effective: if you need a single, standalone fact (like a stock price or today’s weather), just use Google. It’s faster. However, if you need a fact plus context, synthesis, or a quick explanation, use ChatGPT’s web search.
Practical applications:
- For a stock price, Google “NVDA stock.” For an analysis, ask ChatGPT search, “When was Nvidia’s latest earnings call, did the stock price go up or down, and why?” The model will research and synthesize the answer for you.
- For formatting, this feature is awesome. Instead of just searching for data, you can ask ChatGPT search to find “global vaccination rates over the past 5 years in table format.” It will do the research and structure the data for you, saving a ton of manual work.
✅ Unleash Power Features: Deep Research & Canvas.
Here’s where things get really interesting. The creator dives into two of the most powerful, yet underused, features.
Deep Research is like having a dedicated research assistant. You give it a complex topic, and it acts as an AI agent, spending 10-20 minutes browsing dozens of sources to produce a detailed report. The person who shared it gives a brilliant business use case: “Analyze and compare the AI chip roadmaps for Nvidia, AMD, and Intel based on their latest earnings calls.” This is something that would take a human analyst hours. You can even connect it to your Google Drive to have it analyze your private company data alongside public industry reports.
Canvas is an interactive, editable workspace for your AI-generated text. The innovator’s rule here is perfect: “toggle this on if you know you’re going to edit and build upon ChatGPT’s response more than once.” His example of preparing for a performance review is spot-on. You can upload your company’s rubric, have the AI draft an outline, then go in to make inline edits, add your specific achievements, and finally instruct it to generate an executive summary based on all the content in the canvas. It turns a static response into a living document.
This was an incredibly practical breakdown, and I’ve only scratched the surface of the examples and pro-tips the creator shares. This new mental model for using ChatGPT is already making my work so much more efficient.
I highly recommend checking out the full video from this industry pro to see these features in action!