I recently watched Jeff Su’s video about ChatGPT5, and it was eye-opening. Like millions of others, I’d been frustrated with seemingly worse results from what should be a more powerful model. The revelation? We’ve been using outdated prompting techniques with a fundamentally changed AI architecture.
As someone who relies on AI tools daily for my speaking and consulting work, I needed to understand these changes. After testing Jeff’s recommendations, I can confirm they work remarkably well. Here’s my take on the five most effective techniques to dramatically improve your ChatGPT5 outputs.
Understanding What Changed
First, we need to grasp two major changes to GPT5. OpenAI consolidated multiple models into just three options (GPT5, GPT5 Thinking Mini, and GPT5 Thinking) and added an invisible router that decides which model handles your request. This router often defaults to faster but less capable options unless specifically prompted otherwise.
Second, GPT5 follows instructions with extreme precision. While this sounds positive, it means the AI no longer compensates for vague prompts by guessing what we want. It needs clear direction to deliver quality results.
Router Nudge Phrases: The Four-Word Fix
The simplest improvement comes from adding just four words to your prompts: “think hard about this.” This phrase forces the invisible router to select a higher reasoning model.
I tested this with financial planning questions and saw dramatic differences. Without the nudge phrase, I received generic information about index funds versus money market accounts. With it, I got comprehensive analysis including second-order effects I hadn’t considered.
Three phrases consistently trigger deeper reasoning:
- “Think hard about this”
- “Think deeply about this”
- “Think carefully”
When you see the “thinking” indicator appear, you know you’ve successfully triggered the more powerful reasoning capabilities.
Verbosity Control: Getting the Right Length
Another game-changer is controlling output length with specific phrases. After testing dozens of variations, I found three reliable options:
- Low verbosity: “Give me the bottom line in 100 words or less. Use markdown for clarity and structure.”
- Medium verbosity: “Aim for concise 3-5 paragraph explanation.”
- High verbosity: “Provide a comprehensive and detailed breakdown, 600-800 words.”
I’ve added these to my text expander app so I can quickly access them when needed. The difference in output quality is substantial – especially when communicating with executives who need concise information.
Prompt Optimization: Let AI Improve Your Prompts
OpenAI offers an official prompt optimizer tool that rewrites prompts specifically for GPT5. While it requires a separate developer account, there’s a free workaround: ask ChatGPT5 itself to optimize your prompts.
Simply use a meta-prompt like: “You are an expert prompt engineer specializing in creating prompts for ChatGPT5. Take my prompt and make it better.” Then paste your initial prompt.
The optimizer consistently makes three improvements: adding structure, eliminating vagueness, and incorporating error handling. This approach teaches you to write better prompts yourself over time.
XML Sandwich: Structure Matters More Than Ever
With GPT5’s precision-focused approach, organizing your prompts with XML tags dramatically improves results. Instead of dumping everything into one paragraph, explicitly label each component:
<task>Act as a hiring manager and ask me three questions based on my resume and job description</task>
<resume>[paste resume here]</resume>
<job_description>[paste job description here]</job_description>
This structure helps GPT5 better comprehend its task, leading to more relevant outputs. I’ve created templates in my text expander for common scenarios, which saves significant time.
The Perfection Loop: Self-Improvement in Action
The most powerful technique leverages GPT5’s ability to critique itself. Instead of accepting its first response and manually requesting improvements, tell it upfront to create its own definition of excellence, grade its work, and iterate until it achieves the best result.
For example: “Draft an outline for my quarterly business review presentation. Before you begin, create an internal rubric with five criteria for a perfect QBR. Then use that rubric to internally iterate the outline until your response scores 10 out of 10.”
This approach works best for complex zero-to-one tasks like creating finished documents or writing production-ready code.
Combining These Techniques for Maximum Impact
What I love most about these techniques is they’re stackable. You can use nudge phrases with verbosity control, XML sandwiches, and the perfection loop simultaneously.
Since implementing these approaches, my AI outputs have improved dramatically. I’m getting more thoughtful, better structured, and more accurate responses – exactly what we should expect from a more powerful model.
The key takeaway? AI tools evolve, and so must our techniques for using them. By adapting our prompting strategies to match GPT5’s new architecture, we can unlock its full potential and get significantly better results with minimal extra effort.
