🎥 Watch the full episode here

📆 Published: 29th October 2025

🕒 Estimated Reading Time: 3 mins. Time saved: 25+ mins! 🔥

🤘 Stop Comparing AI Models—Start Using Them Right: The Practical Guide to GPT-5, Claude & Gemini

This week it’s another practical one. It’s also pretty punchy so easy to jump into and listen to the whole podcast if you need to.

If you’re anything like me, given the crazy pace of things you’re probably totally over trying to compare various AI models. If you search for some advice on this, most podcasts on the subject seem to be just endless model comparisons that don't tell you how to actually use them.

However, this recent one from Peter Yang is a great practical review on matching the right AI tool to your specific workflow, plus nine real-world projects to illustrate how you can save hours every week.

Whether you're a product leader drowning in document edits, a founder making critical decisions, or a marketer managing multiple content streams, this guide will help you leverage each model's strengths - and the game-changing feature (Projects) that most people completely overlook.

🏗️ The Model Matchup: Which AI Wins at What

The biggest mistake people make? Using the same AI for everything. Each model has distinct strengths, and the real productivity gain comes from matching the tool to the task.

Here's the breakdown across ten critical use cases:

Everyday Answers: GPT-5 takes the crown here. It's fast, concise, and perfect for quick queries. Claude is solid too, but the $20/month plan hits rate limits frustratingly fast. For rapid-fire questions, ChatGPT is your go-to.

Writing & Editing: Claude dominates this space—and it's not even close. If you feed Claude examples of your best writing, it learns your voice and style. ChatGPT tends to over-rely on bullet points and produces verbose, generic output. For the 80% of work that involves writing and editing, Claude is the clear winner.

Coding: This one's genuinely competitive. Both Claude and ChatGPT's CodeEx deserve stars here. Claude Code is faster for most tasks, but many engineering teams swear by CodeEx with GPT-5 High—it can fix gnarly bugs in just a few lines of code. OpenAI has caught up to Anthropic remarkably fast in this space.

Deep Research: Claude pulls ahead again. The critical difference? Claude produces 4–5 pages of actually readable research, while ChatGPT and Gemini dump 30–40 pages of information on you. For decision-making, concise synthesis beats data overload every time.

Web Search: Gemini wins on speed. When you need quick answers (nearby restaurants, current events), Gemini's AI mode is noticeably faster than waiting for ChatGPT to generate. Speed matters when you're searching, not researching.

Voice Chat: ChatGPT still has the most natural, authentic voices. That said, my suggestion is to use a dedicated tool like Whisper Flow because all three models interrupt too frequently during dictation.

Image Generation: Gemini's Nano Banana is the standout here. It generates faster than ChatGPT and produces superior results. If you're building image-heavy applications, Nano Banana is worth exploring.

Video Generation: ChatGPT's Sora takes this one, despite Gemini's VO 3.1 being arguably a better model. OpenAI hired heavily from Meta, so they understand social dynamics better. For family videos and social content, Sora feels more natural.

Live Camera & Computer Use: Honestly? Skip these for now. Computer use is clunky—watching AI click around and take screenshots takes forever. Neither feature has found its killer use case yet.

Atlas (OpenAI's New Browser): Interesting MVP, but not ready for prime time. It's slower than Chrome for basic searches and returns less relevant results. OpenAI is taking the right approach (launch MVP, iterate with users), but there's work to do.

Grok: Despite being a capable model, XAI's aggressive marketing of AI companions (including "waifus") undermines the product's credibility. The company needs real marketing expertise to appeal to mainstream users.

Key Takeaways:

• Match models to tasks: GPT-5 for speed, Claude for writing/research, Gemini for images

• Don't use one AI for everything—you're leaving productivity on the table

• Emerging features like Atlas and computer use aren't ready for critical workflows yet

• The real competitive advantage isn't the model—it's how you structure your prompts and context

🧠 The Secret Weapon Nobody Talks About: Projects

Here's what separates casual AI users from power users: Projects. While everyone obsesses over AI agents and browsers, Projects is the feature that actually saves hours every week.

All three major labs (OpenAI, Anthropic, Google) offer Projects, but Claude Projects edges ahead because it lets you attach Google Docs directly, paste plain text files, and maintain memory across conversations. This is the foundation for everything that follows.

Three Pro Tips for Projects:

Tip 1: Attach Google Docs Directly. This ensures your project always knows the latest version of your documents. Add plain text files and deep research outputs as context. Your AI becomes smarter with every addition.

Tip 2: Include Your Output Style in the Prompt. Specify exactly how you want the AI to format responses. For example: "Use short bold stems followed by 2–3 short sentences so the entire output is easy to skim." This consistency saves iteration cycles.

Tip 3: Get AI to Iterate on Your Prompt. After a long conversation where you finally get the output you want, ask: "Can you update the project prompt to produce this output in one shot next time?" Save the updated prompt. Next time you use the project, you'll get to the desired output instantly.

Key Takeaways:

• Projects are more valuable than AI agents or browsers for most workflows

• Attach templates, examples, and context upfront—don't repeat yourself each conversation

• Refine your prompts iteratively, then save them for reuse

• Claude Projects has the best feature set for document-heavy work

📋 The Strategy Document Workflow: From Chaos to Clarity

Product managers spend weeks editing strategy documents. Here's how Projects transforms that time-sink into a streamlined process.

The Traditional Approach (2+ weeks): Think hard → Talk to customers → Get stakeholder feedback → Flood of Google comments → Iterate → More feedback → Repeat. You're essentially a document editor, not a strategist.

The Projects Approach (Days, not weeks):

Start by creating a project with three key inputs:

1. A Template (e.g., Amazon's Working Backwards framework). This shows AI what good looks like.

2. Deep Research Output (e.g., a 7-page report on your market). This provides strategic context.

3. Your Rough Draft (voice-dictated notes, initial thoughts, raw ideas). This captures your unique perspective.

Now ask AI to synthesise these into a compelling strategy document. For example, if you're building an AI coding startup competing in a crowded space, you might ask: "Create an Amazon-style PR FAQ for our product using these three documents as context."

AI drafts the PR FAQ. You give feedback: "I think the money is in enterprise. How do we compete with Figma and Make?" AI iterates. You refine. Within days, you have a document that would normally take weeks of back-and-forth.

The key behaviour change: Use AI as your thought partner, not your final answer. Give it context, get a first draft, iterate with feedback, then share with stakeholders. The document is better, and you've saved 10+ hours.

Key Takeaways:

• Upload templates, research, and rough drafts to your project upfront

• Use AI to create first drafts, not final outputs

• Iterate with feedback loops before sharing with stakeholders

• This workflow cuts strategy document timelines by 70%+

🎬 Nine AI Projects That Actually Save Time

The real power of Projects isn't theoretical—it's in the specific workflows you build. Here are three standout examples:

Project 1: Show Notes Generator Project

After recording a podcast interview, paste the raw transcript into this project. It automatically generates:

• Top quotes for intro reels

• Moments to cut (audio issues, repetitive chatter)

• YouTube titles and thumbnail patterns

• Social media posts with timestamps

• Show notes with proper formatting

Time saved: 3–5 hours per episode. The prompt is detailed (available in the Peter Yang's newsletter), but once set up, it's a one-tap operation. You get a solid first draft for all assets, then iterate to make them perfect.

Project 2: Family Trip Planner

This project captures your family's preferences: favourite destinations, trip types (half-day, full-day, weekend), availability constraints (e.g., "daughter has taekwondo Saturday mornings"), and past experiences.

Ask it: "What family-friendly events are happening this weekend?" Because it has all your context, it finds relevant, age-appropriate activities you'd actually enjoy. No generic suggestions—just personalised recommendations.

Project 3: Product Strategy & PRD Hub

Upload your working backwards template, market research, and initial thoughts. Use it to draft PRDs, strategy documents, and competitive analyses. The project becomes your strategic thinking partner, always aware of your company's context and constraints.

Key Takeaways:

• Projects multiply the value of AI by 10x through context and memory

• The best projects are specific to your workflow (podcast, family, product, etc.)

• Each project saves 3–10 hours per week once optimised

• Start with one project, then expand as you see the pattern

🔮 Deep Research: The Extended Thinking Mode for Hard Decisions

Most people use Deep Research wrong. They type a single question and wonder why the output is mediocre. Here's how to actually use it.

Deep Research is for difficult decisions, not quick answers. Use it when you need to evaluate trade-offs, choose between options, or understand a complex market. All three providers (OpenAI, Anthropic, Google) offer it, but Claude produces the most concise, readable reports (7 pages vs. 30–40 pages from competitors).

Five Steps to Deep Research Mastery:

Step 1: Start with a Clear Goal. Don't ask "What's the AI coding market?" Ask "I'm deciding whether to build an AI coding tool or acquire one. What should I consider?" Specificity drives better research.

Step 2: Share Relevant Context (This is the biggest lever). Provide a full page of context: documentation, voice-dictated notes, past decisions, constraints. The more personalised context you provide, the more personalised and useful the research becomes.

Step 3: Specify Sources. Tell it where to look. "Research professional reviews on TechCrunch and user reviews on Reddit." Point it to the right sources, not generic web search.

Step 4: Ask It to Ask You Questions First. Include in your prompt: "Ask me clarifying questions before you begin research." This ensures the AI understands your specific situation before diving in.

Step 5: Integrate with Projects. Once you get the research report, add it to your project. Now your project has even more context for future decisions.

Real Example: Finding the Right Piano App

Instead of asking "What's the best piano app?", provide context:

• "I have a Kawai piano, I'm an advanced beginner"

• "I've been learning from YouTube videos with floating notes"

• "I want to learn sheet music reading and proper technique"

• "My budget is $X per month"

• "Ask me questions before researching"

Deep Research asks clarifying questions, then produces a report recommending Playground Sessions (not Simply Piano, which you might have guessed). The recommendation includes why it matches your specific needs. This is expert-level advice, personalised to you.

Another Example: Multi-Generational Family Trip

Your family trip project has context on preferences, past destinations, and constraints. You ask Deep Research: "What are great places to go in December for a multi-generational trip under $10,000?"

It asks clarifying questions (weather preferences, activity types, etc.), then recommends New Orleans, Mexico's Riviera Maya, and Costa Rica—not generic tourist destinations, but places that match your family's specific profile.

Key Takeaways:

• Deep Research is for decisions, not quick answers

• Personalised context is the biggest lever—provide a full page upfront

• Ask it to ask you questions before researching

• Claude's concise reports are better than competitors' data dumps

• Integrate research reports back into Projects for compounding context

🚀 Getting Started: The Three-Step Framework for AI Mastery

You don't need to master all three models or all nine projects. Start here:

Step 1: Use Different Models for Different Tasks

• ChatGPT for everyday answers

• Claude for writing, editing, and deep research

• Gemini for images and multimodal work

• CodeEx or Claude Code for coding (depending on your preference)

Step 2: Create One Project for Your Main Work Stream Pick your biggest time sink (strategy documents, content creation, customer research, whatever). Build a project with templates, examples, and context. Use it as your thought partner. Watch the iteration cycles compress.

Step 3: Use Deep Research for One Important Decision Pick a decision you're facing (product direction, hire/build/buy, market entry, etc.). Provide full context. Let Deep Research ask questions. Get a concise report. Add it to your project. Repeat for future decisions.

The Compounding Effect: Each project gets smarter as you add context. Each decision informed by Deep Research becomes part of your project's knowledge base. Within weeks, you've built a personalised AI system that understands your business, your preferences, and your constraints.

This isn't about using AI to replace thinking—it's about using AI to accelerate thinking. The models are tools. The real skill is structuring your prompts, providing context, and iterating with feedback.

Key Takeaways:

• Start with one model per task, not one model for everything

• Build one project that matches your workflow

• Use Deep Research for decisions, not questions

• The system compounds—each iteration makes the next one faster

• The real productivity gain comes from structure, not from the model itself

🔗 Links Referenced

• Claude Projects

• ChatGPT Projects

• Google Gemini

• Playground Sessions (Piano App)

• Amazon Working Backwards Framework

• Nano Banana Image Generation

What's your biggest time sink right now? Start there. Build a project. Watch your productivity multiply. The models are ready—the question is whether you're ready to use them strategically.

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know what you liked or want changing! 🚀👋

Alastair 🍽️.