You’re not imagining it, your ChatGPT-5 results probably have gotten worse. I was getting so frustrated because my trusty old prompts were suddenly spitting out shallow, unhelpful answers from what’s supposed to be a more powerful model. Then I found this fantastic video from an industry pro that laid out exactly why this is happening and, more importantly, how to fix it. The creator spent a month digging into OpenAI’s own guides to uncover the new rules of the road for prompting GPT-5.
The Core Problem: Why Your Old Prompts Are Broken
The big reveal from the creator is that OpenAI made two fundamental changes to GPT-5’s architecture. Understanding these is key to getting great results again.
First, they simplified the model choices in a process called model consolidation. Instead of a big menu of options, there’s now an “invisible router” that directs your prompt to the right tool. Think of it like a customer service operator. The problem is, this router isn’t perfect. If your prompt is too simple, it might send you to the fastest, but dumber, model to save OpenAI money on computing costs.
Second, GPT-5 is now incredibly good at following instructions. This sounds great, but it’s a double-edged sword. The model was trained with AI agents in mind, which require surgical precision. The good news is it will do exactly what you say. The bad news is it has lost the ability to guess what you mean if your prompt is vague. The old models were better at filling in the blanks for us, but GPT-5 takes you literally.
So, if we keep using our old, less-structured prompts, we’re basically telling a very literal-minded assistant to do a sloppy job. But this innovator shared five awesome techniques to adapt, and I’m going to break down the three most powerful concepts.
Deeper Dives & Key Insights
📌 Insight 1: Forcing a Smarter Response with “Nudge” & “Verbosity” Phrases
This is the easiest and fastest way to see an immediate improvement. The mind behind it discovered you can directly influence that “invisible router.” By adding simple phrases to your prompt, you can nudge the router toward using a higher-reasoning model and control how much text you get back.
For Deeper Reasoning: To prevent the router from defaulting to a weaker model, the creator suggests adding a “nudge phrase” to the end of your prompt. The most effective ones were:
- “think hard about this”
- “think deeply about this”
- “think carefully”
When you use one of these, you’ll often see the “thinking…” indicator appear in the interface, which is a sign that a more powerful reasoning process has been triggered. The resulting output is less about surface-level answers and more about identifying second-order effects and nuances you might have missed. It’s perfect for high-stakes tasks where overlooking a detail could be a problem.
For Controlling Length: The router also controls output length, or verbosity. The person who shared it found you can guide this with specific instructions:
- Low Verbosity: For quick summaries or Slack messages, use a phrase like, “give me the bottom line in 100 words or less.”
- Medium Verbosity: For explanations that need context, try “aim for a concise 3 to five paragraph explanation.”
- High Verbosity: For detailed documents, be explicit with “provide a comprehensive and detailed breakdown, 600 to 800 words.”
💡 Insight 2: Building an “XML Sandwich” for Flawless Structure
Because GPT-5 is so literal, providing clear structure is no longer optional, it’s critical. The talented creator highlighted a technique they call the “XML Sandwich.” This is all about labeling the different parts of your prompt so the AI knows exactly what to do with each piece of information. Instead of one giant wall of text, you’re putting your instructions into clearly labeled boxes.
Think of it this way:
- Old Way (Vague): “Here’s my resume and a job description. Help me prepare for the interview.”
- New Way (XML Sandwich):
<task>Act as a hiring manager. Based on the resume and job description provided, ask me three behavioral questions I am likely to face.</task>
<resume>[…paste your resume here…]</resume>
<job_description>[…paste the job description here…]</job_description>
The difference in output quality is huge. This structure eliminates ambiguity and helps the model execute the task with the precision it was designed for. A big takeaway for me was to create templates for these in a text expander app. The author suggests having a default template with tags like <task>, <context>, <format>, and even a pre-filled <tone> tag (e.g., <tone>user-friendly and conversational</tone>) to save time on recurring tasks.
✅ Insight 3: The “Perfection Loop” for World-Class Outputs
This one blew me away! The expert pointed out that GPT-5 is excellent at critiquing its own work, and we can use that to our advantage. The “Perfection Loop” is a technique where you instruct the model to define excellence for itself before it even starts writing. You tell it to create an internal rubric, grade its own draft against it, and iterate internally until it produces a top-scoring result.
Imagine hiring a writer who drafts a report, grades it a 6/10 against their own quality checklist, rewrites it to an 8/10, and keeps refining until it’s a perfect 10/10, all before showing you the final version. That’s what you’re making the AI do. The creator gives a couple of great examples:
- For a Report: “Write a market analysis report on the enterprise AI industry. Before you begin, develop an internal rubric for what constitutes a world-class market analysis report. Internally iterate and refine the draft until it scores top marks against your rubric.”
- For a Presentation Outline: “Draft an outline for my QBR. 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 is a high-effort technique best reserved for complex, zero-to-one tasks like creating finished documents or writing production-ready code. It ensures you get the absolute best first draft possible.
These techniques can be stacked on top of each other for even better results. You can use a nudge phrase within an XML-structured prompt that also includes a perfection loop. It’s a whole new way of thinking about prompting, and it’s essential for getting the most out of GPT-5.
Go check out the original post from this AI professional for the full breakdown, including a meta-prompt that turns ChatGPT into a free prompt optimizer!