Summary of "95% of People STILL Prompt ChatGPT-5 Wrong"

Summary of “95% of People STILL Prompt ChatGPT-5 Wrong”

The video analyzes why many users experience worse results with ChatGPT-5 (GPT5) despite not changing their prompts. It highlights fundamental architectural changes in GPT5 that render old prompting techniques less effective. The creator spent a month testing GPT5 and shares five practical tips to significantly improve outputs.

Key Technological Concepts and Product Features

GPT5 Architecture Changes

Model Consolidation: OpenAI reduced multiple models into three variants: GPT5, GPT5 thinking mini, and GPT5 thinking. An invisible router assigns user requests to different models. This router is imperfect and tends to favor cheaper, less powerful models unless nudged.
Improved Instruction Following: GPT5 is trained to follow instructions very precisely, especially for AI agents. This makes vague prompts less effective. Precise, well-constructed prompts are now essential.

Invisible Router

Routes requests to different GPT5 models based on prompt cues.
Can lead to inconsistent output quality if prompts are not optimized.

Prompt Optimization Tool

OpenAI offers an official prompt optimizer that restructures and improves prompts by adding clarity, structure, and error handling.
Requires a separate developer account and payment method but can be mimicked with a meta prompt in ChatGPT-5.

Five Tips to Improve GPT5 Prompting

Router Nudge Phrases (Low Effort) Adding explicit phrases like “think hard about this,” “think deeply about this,” or “think carefully” at the end of prompts forces the router to select higher reasoning models. This triggers deeper reasoning and more nuanced answers, especially useful for high-stakes or complex tasks.
Verbosity Control (Low Effort) Use specific phrases to control output length and detail:
- Low verbosity: “Give me the bottom line in 100 words or less.”
- Medium verbosity: “Aim for a concise 3 to 5 paragraph explanation.”
- High verbosity: “Provide a comprehensive and detailed breakdown, 600 to 800 words.”

This helps tailor responses for different contexts like Slack messages, team meetings, or detailed reports.

Use OpenAI’s Prompt Optimizer or Meta Prompt (Medium Effort) Use the official prompt optimizer tool or a meta prompt that asks GPT5 to critique and improve your prompt. This enhances prompt clarity, structure, and error handling, leading to better outputs.

Meta prompt example: instruct GPT5 as an expert prompt engineer to improve your prompt.

Create an XML Sandwich (Medium Effort) Structure prompts with explicit XML-like tags to separate context, task, and output format. This labeled structure helps GPT5 understand and execute complex instructions more accurately.

Recommended for recurring tasks and custom instructions, especially in custom GPTs or projects.

Perfection Loop (High Effort) Instruct GPT5 to create its own internal rubric for excellence, grade its output against it, and iteratively improve until it achieves a perfect score.

Ideal for complex, zero-to-one tasks like writing finished documents or production-ready code. Can be added as a universal suffix to prompts for consistent self-improvement.

Additional Notes

Tips are cumulative and can be combined for better results.
The video recommends HubSpot’s free playbook on prompt engineering for deeper learning.
The creator offers a free AI toolkit and text expander app suggestions for managing prompt templates and phrases.

Main Speaker / Source

The video is presented by an AI content creator named Jeff (implied from the transcript), who conducted extensive testing and shared practical prompting strategies for GPT5 users.
HubSpot is mentioned as a sponsor and resource provider for prompt engineering playbooks.
OpenAI is referenced as the developer of GPT5 and the official prompt optimizer tool.