95% of People STILL Prompt ChatGPT-5 Wrong

ChatGPT-5 requires new prompting techniques to get better results, as old methods perform worse due to architectural changes and the model's enhanced instruction following capabilities.

• New prompting techniques are essential for ChatGPT-5 due to architectural changes and its enhanced instruction-following capability.

• Use router nudge phrases to trigger deeper reasoning and verbosity control phrases to manage output length.

• Structure prompts with XML tags, optimize them with meta-prompts, and employ the 'perfection loop' for iterative self-improvement on complex tasks.

Following the launch of ChatGPT-5, many users experienced worse results because OpenAI made fundamental architectural changes, making old prompting techniques inefficient. The new model consolidates previous versions and is much better at following precise instructions, meaning vague prompts yield poor outcomes. Five simple tips, including router nudge phrases, verbosity control, prompt optimization, XML sandwiching, and the 'perfection loop,' can drastically improve GPT-5 outputs by forcing deeper reasoning, controlling output length, structuring prompts, and enabling self-correction.

GPT-5 Architectural Changes

• 00:00:36 ChatGPT-5 introduced two major changes: model consolidation and enhanced instruction following. Model consolidation means a hidden router now selects which model handles a request, often defaulting to less powerful options. The model's improved ability to follow instructions, designed for AI agents, means it struggles with vague or poorly constructed prompts, leading to worse results if old prompting techniques are used.

Router Nudge Phrases

• 00:02:07 Router nudge phrases, such as 'think hard about this,' can force ChatGPT-5's invisible router to select a higher reasoning model, leading to more nuanced and detailed outputs. Activating deeper reasoning often reveals 'second-order effects' not initially considered, which is crucial for high-stakes tasks where overlooking such details could be detrimental. Phrases like 'think hard about this,' 'think deeply about this,' and 'think carefully' reliably trigger this deeper reasoning.

Verbosity Control

• 00:04:30 Specific phrases can control ChatGPT-5's output length, a feature handled by the invisible router's verbosity setting. Low verbosity phrases (e.g., 'give me the bottom line in 100 words or less') are ideal for critical information. Medium verbosity (e.g., 'aim for a concise three to five paragraph explanation') suits key takeaways with context, while high verbosity (e.g., 'provide a comprehensive and detailed breakdown, 600 to 800 words') is best for comprehensive documents.

Prompt Optimization

• 00:06:21 OpenAI offers an official prompt optimizer tool that rewrites prompts for ChatGPT-5 by adding structure, eliminating vagueness, and incorporating error handling. A free workaround using a meta-prompt (e.g., 'You are an expert prompt engineer... take my prompt and make it better') allows ChatGPT-5 to critique and improve its own instructions, providing similar benefits without requiring a separate developer account or payment method.

The Perfection Loop

• 00:09:54 The 'perfection loop' exploits ChatGPT-5's ability to critique itself, instructing the model to first create its own definition of excellence for a task. It then grades its initial output against this internal rubric and iteratively refines its work until it achieves top marks. This method is highly effective for complex, 'zero-to-one' tasks like generating finished documents or production-ready code, and can be combined with other prompting tips for enhanced results.