Guide: Mastering GPT-5

Understanding GPT-5 and Model Evolution

Note: I provided an in-depth overview of GPT-5's capabilities and initial impressions in Friday's issue. This guide focuses on practical implementation techniques.

GPT-5 represents a significant advancement in AI capabilities, offering improved reasoning, better long-context understanding, and more sophisticated problem-solving abilities. Unlike earlier models, GPT-5 excels at complex planning, multi-step reasoning, and self-evaluation—making it particularly valuable for professional work that requires strategic thinking and quality control.

However, there's an important caveat: OpenAI has removed access to many older models, forcing users to adapt to the new system whether they're ready or not. This "upgrade or lose access" approach means you can't simply stick with what was working before. While this pushes everyone toward more capable tools, it also eliminates the choice to use simpler, faster models when they might have been perfectly adequate for basic tasks.

This guide focuses on practical techniques for getting the most value from these advanced AI capabilities, treating GPT-5 not just as a writing tool, but as an intelligent collaborator that can plan, execute, and evaluate work alongside you.

Note on Legacy Models

If you prefer to use older models like GPT-4o, you can still access them:

1. Go to Settings in ChatGPT

2. Toggle "Show Legacy Models"

3. Select your preferred older model from the dropdown

Quick Start Guide

Before diving into detailed techniques, here are the essential practices that deliver immediate results:

Choose the Right Model

• GPT-5: Default choice for most professional tasks

• GPT-5 Thinking: Use for complex analysis, novel coding, or strategic planning

• Thinking Pro: For the most demanding tasks requiring extensive reasoning (if you’re paying)

Set Clear Parameters

Always establish upfront:

• Objective and success criteria

• Target audience and tone

• Expected format and deliverables

• Quality standards

Use the Plan-Then-Execute Approach

1. Ask GPT-5 to propose a plan before starting work

2. Review and approve the approach

3. Let it execute while maintaining quality checkpoints

4. Iterate based on results

Core Setup for Maximum Effectiveness

Model Selection Strategy

• Start with GPT-5 for general professional tasks

• Upgrade to GPT-5 Thinking when you need:

  • Novel code solutions

  • Complex data analysis

  • Strategic planning

  • Creative problem-solving

• Set reasoning expectations: "Spend up to 60 seconds planning; use 10-15 key reasoning steps"

Personalization Setup

• Configure Custom Instructions with your role, typical audiences, preferred tone, citation style, and common constraints

• Create project-specific guidelines for recurring work types

• Maintain updated source materials like style guides, data dictionaries, and current project briefs

Workspace Organization

• Use Projects for multi-file work to keep context, files, and conversations together

• Name files descriptively for easy reference

• Keep source documents current and well-organized

The Mental Model: Planner + Executor

The most effective approach treats GPT-5 as both a strategic planner and skilled executor:

1. Frame the Work

• Define clear objectives and success criteria

• Specify constraints and requirements

• Identify target audience and context

2. Plan Together

• Request a brief plan with milestones and potential risks

• Ask for a scoring rubric to evaluate success

• Review and refine the approach before execution

3. Execute with Oversight

• Generate multiple options when appropriate

• Use the established rubric for evaluation

• Maintain quality checkpoints throughout

Prompt Patterns

The "Deep Thinking" Approach

Take the time needed; prioritize accuracy over speed. 
Sketch your plan briefly, then execute methodically.

Variations:

• "Think deliberately through each step"

• "Reason systematically; show your key decision points"

• "Use up to [X] seconds for planning; aim for thorough analysis"

Plan-Then-Execute Pattern

You are my [specific role]. 

Goal: [clear objective]
Constraints: [limitations and requirements]
Deliverables: [expected format and scope]

Process:
1) Propose a 5-8 step plan with potential risks and mitigations
2) Wait for my approval OR proceed if I say "go"
3) While working, maintain a brief changelog and list any assumptions

Options-Evaluation-Improvement Cycle

Generate 3 diverse approaches to this challenge. 

Then create a 6-8 point evaluation rubric with weights.

Score each approach using the rubric, justify your scores, and select the best option.

Improve the winning approach to version 2.0 and show what changed.

Style Capture and Application

Here are 2-3 samples of my writing style. 

Analyze and create a style guide covering:
- Tone and voice characteristics
- Sentence structure preferences  
- Common phrases and terminology
- Words or approaches to avoid

Apply this style guide to all future responses in this conversation.

Quality Assurance Integration

Before finalizing any response, run a quality check:
- Fact-check key claims against reliable sources
- Flag any weak evidence or unsupported statements
- Mark assumptions clearly
- List the top 3 potential failure modes and how to address them
- Cite all sources used

Research and Analysis Framework

Research task: [specific topic or question]

Deliverables:
- 1-page executive summary
- Key findings with source citations
- Areas of uncertainty or conflicting information
- Recommended next steps or experiments
- 5-slide presentation deck

Prioritize reputable primary sources and note any limitations.

Advanced Techniques and "Power User" Tips

Depth and Quality Enhancement

• Request deeper thinking: Explicitly ask GPT-5 to "take more time" and "show the planning process" for better results

• Use evaluation rubrics: Have GPT-5 create scoring criteria before judging its own work

• Set creative constraints: Provide specific requirements and examples to follow or avoid

Efficiency Multipliers

• Template development: Save your best prompt patterns for reuse

• Multi-file workflows: Upload all relevant materials at once with clear instructions about relevance and citation

• Self-audit capabilities: Request assumption registers, risk assessments, and fact-checking before finalizing work

Team Implementation and Scaling

Custom GPT Development

Create specialized assistants for recurring needs:

• Domain-specific experts with pre-loaded knowledge and processes

• Style and brand enforcers for consistent communications

• Quality assurance specialists for review and compliance checking

Template for Custom GPTs:

This GPT serves as our [domain] specialist. It:
- Enforces [specific style/policy guidelines]
- Accepts and processes relevant file types
- Proposes logical next steps
- Creates appropriate artifacts (documents, presentations, data analysis)
- Runs final quality assurance checks

This is only a small part of the full guide. Here is the full thing, a free gift for you.

• OpenAI quietly tested letting public ChatGPT chats appear in Google search, but pulled the feature after privacy concerns over accidental personal info leaks.

• Google’s AI Mode gets a major boost with image/PDF uploads, dynamic Canvas planning, and live video input via Lens, blending real-world context into AI-powered search.

• Yelp now uses AI to stitch user photos, videos, and reviews into TikTok-style videos for restaurants, blending OpenAI scripts, ElevenLabs narration, and Amazon captions.

• Anthropic’s Claude Code lets devs spawn specialized AI agents

• Microsoft Ads adds bulk website exclusion lists across Search and Audience Ads, letting brands block up to 10k sites per list for better control, safety, and efficiency.

• Bing inches up in market share as Google dips, boosted by AI integrations in ChatGPT and Meta AI, plus Copilot Search innovations challenging the search giant.

• Google will roll out more ads in AI Mode before Q4

• The ‘playbook’ has changed and you have to adapt to it, your brands digital footprint is not just a feed!

• Character.AI launches the first AI-native social feed

📹 Outset - An AI tool that interviews users/candidates with video.

📅 Agenda Hero Magic - Creates AI-generated calendar events

📸 Kondor for iOS- ChatGPT for your phone camera

⚠️ Norm AI - Eliminate compliance risks for your business at scale with AI

🎨 Visual Electric - An image generator built for the creative process.

😍It was great, all I needed

🙂Good, not great

🙄It sucked