AI for Total Beginners: Learn the Absolute Basics of AI 2026
Explainer Articles
SHARE
Facebook X Pinterest WhatsApp

"AI for Total Beginners": How We Would Teach AI in 2026 (Full Guide with Timestamps from our Live Walkthrough)

We went live for two hours to walk through how to actually use AI in 2026, from absolute zero to scheduled automations. Here's the full companion guide with every key moment timestamped so you can jump to what matters most.

Written By
Grant Harvey
Grant Harvey
Apr 3, 2026
43 minute read

Most people watching our AI for Total Beginners livestream joined for one reason: they know they're not getting the most out of AI, but they don't know where to start fixing that. Two and a half hours later, we'd walked through every level of the framework from our 5-Level AI Proficiency Stack article, live on screen, with real demos and a flood of audience questions.

This companion guide covers everything from the stream, reorganized as a step-by-step walkthrough you can follow at your own pace. Every section links to the exact moment in the video. Watch the whole thing, or jump to the parts that matter for your workflow.

One more thing: someone in the chat called us the "Ben & Jerry of AI" (58:58), and we're never letting that go.

First up, the TL;DR

Here are the key moments and takeaways, timestamped so you can jump around:

Now let's walk through it step by step.

Before You Start: Pick Your Tool

The first decision is which AI to use. Grant and Corey have opposing daily drivers (20:27) and they're upfront about it. Grant uses Claude. Corey uses ChatGPT. But they agree on the recommendation:

If you're brand new, try these four (in order of recommendation):

  1. chat.com (ChatGPT)
  2. claude.ai (Claude)
  3. gemini.google.com (Gemini)
  4. grok.com (Grok)

Ask each one a couple questions. See which interface you like. Then just go with that one, because at any given moment, the top two or three models are very close in intelligence (22:30). You'll always be near state-of-the-art with either of the top two.

If you can only afford one $20/month subscription, both Grant and Corey agree (1:58:28): go with ChatGPT. OpenAI's Plus plan gives you the most features for the money. Corey says he rarely hits rate limits on the $20 tier. Claude's $20 plan can run out of usage quickly if you're doing serious work; the real unlock on Claude is the $100 tier (2:01:40).

One more option worth knowing about: Copilot gives you access to both ChatGPT's and Claude's best models under one subscription (21:24). They also just shipped a new feature where the models critique each other's answers, which nobody else is doing at scale yet. If you're already in the Microsoft 365 ecosystem, this might be the move.

What about Perplexity? Someone asked (23:30). Our take: it's excellent for research and its Discover feature (which writes personalized news articles based on your interests), but it's a supplement, not a daily driver. If you're willing to pay for Perplexity Computer, that's a different conversation; it's more of an OpenClaw replacement and is pretty powerful on its own.

⚡ Action step: Go to chat.com and claude.ai. Create free accounts on both. Ask each one the same question about something you're working on. Pick whichever one you prefer. You can always switch later.

Advertisement

Step 1: The Absolute Basics

If you've never opened an AI tool before, start here (10:05). Corey pulls up ChatGPT and walks through the whole thing. It's a chat window. You type a question. The AI responds. That's it.

But the key insight comes almost immediately. Corey doesn't just ask "what's the best EV on the market?" He asks "what's the best EV on the market right now for a new driver?" (11:11) That qualifier changes everything. The AI recommends a Hyundai Kona Electric, specifically because it's easy to park and beginner-friendly. Without the qualifier, you'd get a generic list.

This is the single biggest difference between AI and traditional search. With Google, more words in the search box usually makes results worse (13:01). With AI, more context almost always makes results better. Write in full sentences. Be specific about your situation.

Talk to It Out Loud

Corey also shows the voice input button (11:39). You hit the microphone icon, say what you want, and it transcribes your speech into text and sends it. He dictates "Hey, I've been seeing these new Lucid EVs. Tell me about those." If typing feels like a barrier, just talk. The AI doesn't care how the words get there.

Iterate: Your First Answer Won't Be Perfect

Right after the first demo, Corey shows the concept of iteration (14:04). He follows up by adding that they live in a midsize rural city of 25,000 people. Suddenly the AI starts factoring in charging infrastructure, range anxiety, and service availability. It even flags that Lucid EVs, while impressive, might be tough to service in a small town.

Don't expect to get what you want on the first try. Go back and forth. Add context. Correct it. This is a conversation, not a search query.

Advertisement

The Model Picker: Instant vs. Thinking vs. Pro

One thing that confuses beginners is the model selector (16:15). Here's the simplified version:

  • Instant is fast. Handles everyday questions. "What colors make purple?" doesn't need deep reasoning.
  • Thinking takes longer but produces smarter answers. Use this for anything complex.
  • Pro is for power users who need the highest intelligence available and are willing to wait minutes.

Corey's daily recommendation: just keep it on Thinking.

Search Your Past Chats

Here's a feature most people don't know exists. Corey demos the search function (17:23) by typing "Microsoft" into the sidebar search bar. Instantly, every conversation he's ever had mentioning Microsoft appears: subscriber pitching tips, AI integration for program management, security discussions. This is your personal knowledge base, and it's been building every time you chat.

Deep Research: When You Need It to Go Away and Think

Deep Research (18:18) is a tool that lets you ask a question and have the AI think about it for up to 30 minutes. Corey explains: if you're doing a big report and you need it to go through a spreadsheet, gather industry data from a government site, and accuracy is really important, that's where Deep Research earns its keep.

You don't have to just sit and stare at it while it works (19:14). Spin up two or three windows, start each task, go drink your coffee and check your email. The AI will notify you when it's done. "That's where the big productivity boost is. It's doing some leg work for me while I'm going through something else."

Advertisement

Download the Desktop App

One more important step before you move on (33:00): download the desktop app. Both ChatGPT and Claude have one. Grant recommends this because the desktop apps have capabilities the web versions don't, including Claude's Cowork and scheduled automations. Corey's practical reason: "Your AI isn't hiding in one of the 1,200 tabs you have open. It's its own program."

Set Your Privacy Settings (Do This Now)

Before uploading anything personal, lock down your data settings.

On Claude (56:42): Go to Settings. Turn off "Help improve Claude." This prevents your conversations from being used in training data.

On ChatGPT (1:06:41): Go to Settings, then Data Controls. Turn off "Improve the model for everyone." It defaults to off on many accounts, but check.

This is especially important if you're going to upload client information (56:28). Someone asked about uploading client GPA and swim times to find best-fit colleges. Grant's advice: you can absolutely do this, but generify the personal data first or make sure you're on a team or enterprise account where data handling is more controlled. You don't want to feed somebody else's personal information into AI training data.

Personalization: The Setting Nobody Uses

This was one of the biggest "wait, I can do that?" moments. Corey walks through ChatGPT's personalization settings (25:25).

How to find it: Click your profile in the bottom left corner, then Personalization. You'll see options to set your communication style (quirky, professional, nerdy), add custom instructions that apply to every chat, and tell the AI facts about yourself.

Grant pulls it up on screen (25:50) and Corey explains how he uses it: his custom instructions tell the AI to challenge his assumptions and double-check if he's taken anything for granted as true. "I rush. I do a lot of things in a small amount of time every single day. And I need something that can be like, 'Slow down, Corey. You left the keys in the door.'"

This is free, takes five minutes, and dramatically changes the quality of every conversation.

⚡ Action step: Open your AI tool of choice. Go to Settings or Personalization. Write 2-3 sentences about who you are, what you do, and how you want the AI to communicate with you. Save it.

Advertisement

Step 2: Set Up Your First Project (Level 1)

This is the part that surprises people (29:49). Most beginners think Step 1 is learning to prompt. We put projects first because prompting without a project is like giving detailed instructions to a stranger every single time you need help, instead of briefing a coworker once and having them remember.

What is a project? Grant explains it simply (34:54): projects are organizing folders, like Google Drive for your AI chats, but more powerful. Inside a project, every new conversation inherits the same context, instructions, and uploaded files. You set it up once; the AI remembers forever.

The rule of two: If you're going to do something more than twice, it belongs in a project. (54:54)

Live Demo: Building a Job Search Manager in Claude

Grant builds this from scratch on stream. Follow along:

1. Create the project (34:54): In Claude, hit "Create project" from the sidebar. Give it a name ("Job Search Manager") and description. In ChatGPT, it's the project icon in the sidebar.

2. Upload your files (39:32): Go to the "Files" section and add reference documents. For the job search example, Grant uploads a resume. You can upload directly from your device, paste text content, or connect Google Drive. Every chat in this project can now reference your resume without you re-uploading it.

Pro tip: you can also send it your LinkedIn profile and say "build a resume based on this." (53:20)

3. Use XML tags for precision (41:01): If you have multiple files in a project and want the AI to reference a specific one, wrap the file name in XML tags. These are the pointy carrot brackets: "<" and then ">" that direct the AI's attention to exactly the right document. You just put your document name in between those two carrots and bam, you have an xml tag.

4. Write custom instructions (42:40): This is the part that turns a folder into a workspace. Grant writes instructions like "always use web search to look up the current date before drafting a cover letter" and "you are a resume builder whose job is to help me update my resume and write cover letters."

Corey explains what's happening under the hood (44:40): you're giving the project its own "system prompt" (the core set of instructions that shapes every response). Think of it as the AI's job description for this specific project.

5. The shortcut: have the AI write the instructions for you (43:30): Don't know what instructions to write? Ask the AI. Grant types: "Help me write the best set of custom instructions possible based on Claude's current best practice documentation." Claude searches the web, finds its own docs, and writes the instructions for him. It even asks a series of clarifying questions first (46:19): Where will this run? What should it help with? How should it behave? This is worth noting: AI models a year ago didn't stop and ask questions. The fact that it does now is a real improvement.

6. Paste the instructions and save (50:33): Copy the generated instructions into the project's custom instruction field and save. Every new chat in this project now inherits those rules automatically.

7. Set memory scope (1:00:00): Both Claude and ChatGPT let you choose whether a project has its own isolated memory or shares memory with your entire account. For work projects, keep it project-only.

8. Iterate on instructions over time (1:08:25): When the AI gives you a response format you don't like, update your instructions. Grant noticed the job search results didn't include direct URLs, so he added: "Whenever I request job postings, list the direct URLs to the direct job postings with the job description in a bullet point list so I can easily click to apply." Then he told Claude to re-read its instructions and try again. It updated in real time.

Advertisement

The Same Flow in ChatGPT

Grant shows the ChatGPT equivalent (1:00:34): create project, go to project settings (the dot-dot-dot in the top right), set instructions, upload files via "Sources." You can also connect Google Drive or Slack.

One key ChatGPT difference: when creating a project, click the gear icon and set memory to "Project only" instead of "Default" (1:01:15). You can't change it later, so decide upfront.

Sharing: You can share ChatGPT projects with other people on your team (1:02:25) so everyone has the same context and memory. Big unlock for organizations.

One more question from the chat: Should you move your account-wide ChatGPT custom instructions into project instructions instead? (59:36) Yes. Grant doesn't recommend having account-wide custom instructions unless it's something you absolutely need in every single conversation. They can impact results in unexpected ways. Keep instructions project-specific.

What About Gemini?

Gemini has Gems (1:02:39), which are more like custom GPTs than full projects. You can give a Gem a name, instructions, and knowledge files. Grant shows their header image creator Gem (1:03:40) that generates newsletter headers using sample images and specific design instructions. The live demo image was good enough that Grant used it for that day's actual newsletter.

One note: Gems don't organize your chats the way Claude and ChatGPT projects do (1:06:25). You can't see all your previous conversations inside a specific Gem. That's why we recommend Claude or ChatGPT as your daily driver and use Gemini for one-off use cases it excels at.

Also, Corey finds "fast" mode more reliable than "pro" mode for image generation (1:04:32) in this specific Gem. Pro messes up text rendering. A good reminder that "smarter model" doesn't always mean "better results" for every task.

Advertisement

Moving Existing Chats Into Projects

Already have a pile of conversations scattered everywhere? You can move them (1:29:31).

In ChatGPT: Hover over the chat in the sidebar, click the three dots, hit "Move to project," and select the destination.

In Claude: Click the three dots on a chat and choose "Change project."

One audience member flagged a good tip: be careful not to accidentally drop a chat into the wrong project.

ChatGPT's Library Feature

Corey briefly shows ChatGPT's Library (50:02), which displays all of your uploaded files, every image you've generated, and every video you've created, all in one place. "It's easier than finding things in my Google Drive."

⚡ Action step: Create one project right now. Pick the task you do most often with AI (writing, research, planning, whatever). Name the project, add your context documents, and write 3-5 sentences of custom instructions. Or just ask the AI: "Help me write the best custom instructions for this project."

Step 3: Learn the Prompting Basics (Level 2)

Once your project is set up, prompting gets much simpler (1:10:00) because half the context is already loaded. You don't need to say "I'm a marketing director at a fintech startup" every time; the project already knows.

Corey's advice is refreshingly honest: prompt engineering was a big deal a couple years ago (30:14), and while it's still valuable, it's AI 101, not a career unto itself. It was largely mocked by the technical community and heavily adopted by non-technicals. Watch a 15-minute YouTube video on basic patterns and you'll have enough to get started.

The Only Formula You Need

Grant's version (1:11:35):

  • Task: What do you want done?
  • Context: What does the AI need to know? (If you have custom instructions in your project, a lot of this is handled already.)
  • Format: What should the output look like?

You can skip the "persona" part if your project instructions already define the AI's role. Grant's practical example (1:12:00): instead of "analyze the sales data," try "I have a board meeting in 3 hours and I need a spreadsheet analyzing all data across X, Y, Z. Format it exactly like this example." Then upload a previous report so the AI knows what you're aiming for.

Advertisement

The Copy-Paste Power Move

Corey's approach is even more direct (1:13:13): grab a well-formatted example, drop it at the bottom of the chat window, and say "Here's the blank form. Here's a good one. Go check this site. Fill out that form." No fancy prompt engineering required.

Corey also gives everyone explicit permission to copy-paste our articles (1:16:17) into a project and say "help me set up my AI setup exactly like this article says." Go for it.

One Reader's Trick We Love

Linda in the chat (1:26:22) said she uploads every Neuron newsletter to Claude. "Generally, he will thank me or say he didn't know." Grant's reaction: "I think the best way to read our newsletter is probably to just upload it to the AI and be like, 'How could this help me?'"

Two Rules That Matter More Than Any Framework

Both OpenAI and Anthropic agree on these:

  • Tell the AI what TO do, not what NOT to do. Instead of "Don't use jargon," say "Explain in plain language a non-technical reader would understand." Positive instructions consistently outperform negative ones.
  • Be specific about format. If you want bullet points, say so. If you want a one-page summary, say so. If you want it formatted like a specific example, upload the example.

Context Engineering: The Bigger Picture

There's one level deeper worth knowing about: the field is shifting from "prompt engineering" to "context engineering" because the prompt is just one piece of what the model sees. The documents, memories, conversation history, and tool definitions all matter. This is exactly why Level 1 (projects) comes before Level 2 (prompting). The context you set up in your project is the foundation. The prompt is just the question you ask inside it.

⚡ Action step: Go to your project from Step 2. Ask the AI a task-related question using the task + context + format formula. When the result isn't what you want, iterate. Add more context. Specify the format more precisely. Update your project instructions with any formatting rules you find yourself repeating.

Advertisement

Step 4: Connect Your AI to Your Tools (Connectors / MCP)

Before we get to Level 3, there's an important feature that makes everything else more powerful. Grant shows connectors (1:19:06), also called MCP (Model Context Protocol, which is now the industry standard governed by the Linux Foundation).

How to set them up in Claude:

  1. Go to Customize
  2. Click Connectors
  3. Hit the plus button
  4. Click "Browse Connectors"
  5. Find your app (Figma, Monday.com, Notion, Slack, Beehiiv, etc.)
  6. Click the plus sign and log in

Now Claude can use that app's functions on your behalf. Corey has Notion connected to both ChatGPT and Claude (1:20:22). From either tool, he can manage the Notion board that runs his website.

Grant demos Excalidraw creating a mind map (1:17:42) of the sample resume directly inside the Claude chat, all because the connector is plugged in. Want all the text in black? Just ask. Want circles instead of rectangles? Just ask. The key concept: treat the agent as a first-class citizen who can use the app the same way a human can (1:21:19). Any button you can click, the agent can click too.

Fun aside: Anthropic invented MCP, shared it with the world, and ChatGPT adopted it and called them "connectors." Now Claude has changed theirs to "connectors" too. (1:18:52) It's all still MCP under the hood.

⚡ Action step: Open Customize > Connectors in Claude (or the equivalent in ChatGPT). Browse the list. If you see an app you already use daily, connect it. Start with one.

Step 5: Create Your First Skill (Level 3)

This is where most people plateau (1:21:43), and it's where the real compounding gains start.

Here's the scenario: you've spent 15 minutes going back and forth with the AI on a specific task. You've refined the instructions, corrected the tone, specified the format. The result is perfect. Next week, you'll have to do it all over again.

Skills fix this. A skill packages everything the AI learned in that conversation into a reusable file it can call anytime. Think of it as the difference between cooking from scratch every night and meal-prepping. Same quality, fraction of the time.

The "if you're going to do it more than twice" rule applies here too. (1:29:01) If you're going to prompt the same task more than twice, it needs a skill.

Where skills are available:

Advertisement

Creating a Skill: Step by Step

1. Get the result you want first. Iterate with the AI until the output is exactly right. In the demo, Grant wrestles with Claude to get a job search workflow that finds relevant postings, generates custom cover letters, and provides direct application URLs. The AI even asks for specific information before it'll proceed (1:27:30), which Corey appreciates: "AI didn't do that a year ago. Stop and ask questions."

2. Ask it to create the skill. Grant's prompt (1:30:55): "Now can you use your skill creator skill to turn this conversation into a skill I can call anytime I need to look up jobs and create custom cover letters?" Claude's skill creator reads the conversation and packages it into a set of files.

Or, as Corey puts it: just say "make this a skill." (1:40:04) That works too.

3. Save it. Claude has a "Save skill" button (1:33:42) that installs the skill directly into your library. No downloading, no manual setup. One click.

4. Test it. Start a new chat in the same project. Grant says "Go look up the top jobs I can apply for today" (1:34:38) and the AI reads the skill file, runs the search, and finds postings. It works. Grant and Corey both note that the AI is relatively good at recognizing when to call a skill, but two times out of ten it doesn't, so they still tell it explicitly.

5. Iterate on the skill. The first version won't be perfect either. Grant notices the skill didn't auto-generate cover letters, so he says "Let's update this skill so it always automatically makes a cover letter for every job it finds." (1:37:04) Claude updates the skill, and Grant hits "Save skill" again, choosing "Upload and replace." Done.

6. Update your project instructions to call the skill. Grant copies his instructions into the chat (1:36:03) and asks: "Update my instructions for this project to use the newly created job search skill whenever I ask anything job search related." The AI writes updated instructions. Paste them back in.

Skills in ChatGPT

Same concept, different interface (1:37:36). Click your profile in the bottom left corner, go to Skills. You'll see a skill creator and options to create with chat, create with an editor, or upload from your computer. You can share skills across your company on team accounts (1:38:56).

Skills vs. Agents: What's the Difference?

Great question from the audience (1:34:53). A skill is the organizing principle: all the files, instructions, and tools the agent needs to accomplish a specific job. The agent is the thing that reasons and acts. Skills are the recipe; agents are the cook.

The Google Drive Workaround for Any Platform

Corey's clever tip (1:25:37): if you're using a platform without native skills (like Gemini), create a shared skill folder in Google Drive. Save your skill markdown files there. Connect Google Drive to your AI tool and have it reference the skills from the shared folder. Not perfect, but it means you can use the same skills across multiple platforms.

⚡ Action step: Think of a task you do regularly. Go to Claude's skills library or ChatGPT's skills page. Browse what's already available. Then go to one of your best recent conversations and say: "Turn this into a reusable skill I can call anytime."

Advertisement

Step 6: Schedule Your First Automation (Level 4)

Once you have skills that run reliably, you can schedule them. (1:41:00) This is where your AI goes from "a tool I pick up" to "a worker that runs independently."

Where automations live:

Important: Automations are not in the web UIs of ChatGPT or Claude (1:42:33). You need the desktop apps. But don't be intimidated by Codex just because it has "code" in the name. Corey uses it for writing, banking, and cooking recommendations. (1:42:00) "Don't be intimidated by it."

Live Demo: Scheduling a Daily Job Search in Cowork

Grant shows the full flow (1:43:00):

1. Open your project in Cowork. Click "Start a new task in Cowork" and give it permission.

2. Type: "Every day at 6am, I want you to run the job search skill and find all the new jobs that have been posted today."

3. Add restrictions: "It needs to be above $80,000 in the greater San Francisco area. It needs to offer a comprehensive health insurance package."

4. Cowork reads the project context, checks the resume, loads the relevant skill, and creates a scheduled task (1:45:03).

5. Want to change the frequency? Just ask: "Can you actually run that every 15 minutes?" (1:45:15) Done. You can even run tasks every minute (though you'll burn through tokens fast on lower-tier plans).

6. Check your scheduled tasks in the Scheduled panel on the left sidebar (1:48:02). You'll see status, next run time, and a "Run now" button to trigger manually.

7. To pause without deleting, turn off repeats (1:56:22). The task stays in your panel marked as "skipped" so you can reactivate it anytime.

The results? The automation found real jobs at Chime, Salesforce, Flex, Prelim, Cargo, and Gusto (1:53:46), and generated custom cover letters for each one. "This is the job seeker's revenge on applicant tracking software."

The Same Thing in Codex

Grant shows the Codex equivalent (1:45:39): create a project folder, describe what you want ("let's create a skill to track job postings related to my resume"), and Codex scaffolds the whole thing. Same concept, different interface. And Codex is free to download.

Rate Limits: What to Expect

On the $200 Claude plan, Grant rarely hits usage limits (1:55:00). On the $20 plan, Corey notes he mostly hits limits. The $100 Claude plan is widely considered the best value. On ChatGPT's $20 plan, rate limits are higher because OpenAI has raised more money and can afford to burn it on infrastructure. "If you want to do anything serious with AI, you probably don't want to be on the $20 Claude plan."

⚡ Action step: Download the Claude desktop app or Codex. Pick one recurring task you do weekly. Set up a scheduled automation for it. Start simple: a weekly summary, a daily search, or a recurring report.

Advertisement

Step 7: Understand Agents (Level 5)

Grant keeps this brief (1:49:36) because agents deserve their own session. But here's the core distinction:

Automations run tasks on a schedule. You decide what runs and when. Agents run toward a goal. They reason about what needs to happen, pick tools, act, check if it worked, and loop until the job is done. You decide the outcome; the AI decides the steps.

The twist? If you followed along in Step 6, you've already been working with an agent. (1:50:11) The Cowork automation that searched for jobs, pre-qualified them, filtered by salary and location, and wrote cover letters? That IS an agent. It made decisions, called tools, and iterated on its own. You gave it a goal and it figured out the steps.

Corey makes a great observation (1:50:26): these automations are exactly what companies were spending months building as "AI agents" two years ago. "We talked about it a number of times, that soon you're just going to say 'hey, build me an agent that does this.' And now that's what we have."

A quick story that shows what Level 5 really looks like (1:17:04): Corey was at a Mexican restaurant when OpenAI dropped a research paper. He sent it to his OpenClaw agent via Telegram. The agent read the paper, decided it was useful, implemented the findings on his home computer, and finished before he finished dinner. That's Level 5.

For those who want to go deeper: OpenClaw, Claude Code, and Codex are where this lives. But start with Cowork or Codex before you try to build custom agents from scratch.

The Claude Code Leak

One timely topic that came up (1:51:18): Anthropic's Claude Code system prompt leaked the day before the stream. Grant explains why it matters: Anthropic puts enormous effort into the "harness" (the orchestration layer that makes Claude Code good at its job), and now competitors are reverse-engineering it. Within hours, people had replicated it in Python, Rust, and even built a blue-themed Codex version running on OpenAI's models. Both of Claude Code's main competitors (Codex and Gemini CLI) are open source (1:52:20), so the competitive dynamics here are fascinating.

And speaking of Codex: there are no em dashes in code (1:52:46). Corey, a former journalist who used em dashes in every article he ever wrote, finds this deeply personal. Now he uses a coding agent for writing. Fascinating!

Advertisement

Which Tool Is Best for What: Our Current Rankings

This is Part 2 of the stream (2:03:36), and it's the question we get asked the most.

ChatGPT is best for: one-off web searches, image generation, intensive research and strategy (especially Deep Research), and math. Their reasoning models have now solved 18 previously unsolved mathematical problems (2:04:29), 12 of which are confirmed full solves. Terrence Tao explained the situation with a great analogy (2:06:08): imagine hiking in the dark past cliffs of varying heights. You don't know which cliffs are short and which are tall. AI is like a jumping machine that can clear the short ones but has a limit on how high it can reach. Right now, we're finding out that a lot of "unsolved" problems were shorter cliffs than we thought. And the jumping machine keeps getting better.

Gemini is best for: accessing YouTube links (2:06:36), and this is a hard exclusive; no other model can watch and summarize full YouTube videos natively because Google owns YouTube. Also great for creating images with Nano Banana, building automations with Opal, and Deep Research for pulling sources. Caveat: the link quality from Gemini's Deep Research is excellent; the report text itself tends to be sloppy. Use it for the sources, then rewrite.

Claude is best for: deep research for niche links (Claude can't access as many mainstream websites but finds "diamond in the rough" sources (2:08:27) other models miss), working with files on your computer via Cowork, scheduling recurring automations, and writing. Grant says it's the strongest writer. Corey disagrees. They've agreed to disagree. Claude's Skills system is the most mature implementation of reusable AI capabilities.

Grok is best for: reading X (Twitter) links (2:07:00). Grok can access posts, threads, and profiles that other models can't fetch. On a free X account: 20 links per chat, about eight chats. That's 160 links a day, for free. "You don't have to pay Elon if you don't want to."

Codex is best for: scheduled automations, organizing coding work, and building software. Corey's take (2:10:31): use Codex if you want it done right and you're not in a rush. Use Claude Code if you need speed. Grant's recommendation: if you're a software engineer, start with Claude Code. If you're not, start with Codex. (2:10:16)

For organizing your work, Corey recommends Notion (1:20:02) and Grant mentions Obsidian (2:11:23) as a local-first alternative.

There's also a full list of 50+ professional AI tools for specific verticals in the original blog post, covering everything from legal (Harvey) to robotics (Figure AI, Skild AI) to cybersecurity (Wiz) to education (NotebookLM, Oboe).

Gemini Opal: Bonus Demo (and a Cautionary Tale)

Grant shows Opal (2:14:03), Google's automation builder, at the end of the stream. He asks it to create an automation that reviews YouTube videos daily. Opal builds a multi-step workflow that fetches the channel, analyzes recent videos, and generates an HTML report.

But here's the reality check: the output hallucinated a video title that doesn't exist (2:19:39). The report claimed a video called "OpenAI's New Model is Here and It's Incredible" was the most recent upload. Grant checked. No such video. "You completely made that up."

This is exactly why you always review AI output, and why Opal requires more manual correction than Cowork or Codex (2:20:00). There's a lot of functionality in Opal if you're willing to tinker, but the harness controlling it isn't as sophisticated as Claude or ChatGPT's tools yet.

If you want a full Opal deep dive (with Megan Li, the product manager who built it), check out our separate episode with her on the YT channel here.

Advertisement

Your Action Plan for This Week

You don't need to reach Level 5 this week. Here's what to do right now:

  • Today: Pick your daily driver (chat.com or claude.ai). Create a free account. Set your personalization and privacy settings.
  • Tomorrow: Create your first project. Upload one reference document. Write custom instructions (or have the AI write them for you).
  • This week: Ask the AI 5-10 questions inside your project. Iterate when the answers aren't perfect. Update your instructions based on what you learn.
  • Next week: Take your best conversation and turn it into a skill. If you have the desktop app, try scheduling your first automation.

Each level compounds on the last. Projects are the onboarding. Skills are the training. Automations are the job to be done every day. And agents are the coworker who does it all while you drink your coffee.

BONUS Q&A: Your Questions, Answered

We went back and pulled every question from the live chat. A lot of them were answered during the stream (and are covered in the guide above). But a bunch weren't, and we promised we'd circle back. Here's the rest.

We went back and pulled every question from the live chat. A lot of them were answered during the stream (and are covered in the guide above). But a bunch weren't, and we promised we'd circle back. Here's the rest.We pulled every question from the live chat. Many were answered during the stream and are covered in the guide above. Here's the full accounting so nothing falls through the cracks.

Advertisement

Quick Answers (Covered in the Guide Above)

These questions are answered in full in the main body of this article. Jump to the linked section for details.

  • "Where does Gemini rank against Claude and ChatGPT?" (@trumpetguy0912) → See Which Tool Is Best for What
  • "For automation and agents, do you need the more expensive version?" (@katiekinkead1962) → See Rate Limits: What to Expect. Short answer: you can do projects, prompting, and skills on the $20 tier. Automations require the desktop app (free to download). You'll hit rate limits faster on cheaper plans.
  • "How long is good to wait in Thinking mode?" (@lexymartin5153) → See The Model Picker. Can be a few seconds to a few minutes. You don't have to stare at it; go do something else.
  • "Does all this work in ChatGPT?" (@bithammer2512) → Yes. See The Same Flow in ChatGPT.
  • "Is that feasible in Gemini?" (@thomasvareilhes2861) → Partially. See What About Gemini?. Gems handle custom instructions and knowledge, but don't organize chats the way projects do.
  • "How do we move prior standalone chats into a project folder?" (@joematurojr) → See Moving Existing Chats Into Projects. Three dots menu > Move to project.
  • "Can I keep client info in projects?" (@americancollegeconnectionACC) → Yes, but see Set Your Privacy Settings. Generify PII or use a team/enterprise account.
  • "Can Claude create concept/mind maps?" (@DebSchiano) → Yes! See the Connectors section where Grant demos Excalidraw creating one live.
  • "Are skills the same as an agent?" (@ManageforGrowth) → No. See Skills vs. Agents. Skills are the recipe; agents are the cook.
  • "Do you prefer Claude Code or Codex?" (@InsideJWsMind) → See tool rankings. Grant prefers Claude Code, Corey prefers Codex.
  • "If all you can afford is $20/month, which one?" (@ninedesign108) → ChatGPT. See Before You Start: Pick Your Tool.
  • "Can you create a task that regularly looks for new research on a topic?" (@DebSchiano) → Yes, that's exactly what scheduled automations do. See Step 6.
  • "Do you have control over what model ChatGPT is using?" (@davidhead9543) → Yes. See The Model Picker.
  • "Best way to use AI tools for a job search?" (@mkhan_mvp) → See the full job search manager demo that runs through the entire stream.
  • "Can you customize ChatGPT the same way as Claude?" (@lisaeatonwright8715) → Yes. See The Same Flow in ChatGPT.
  • "Would you suggest moving ChatGPT instructions to Project instructions?" (@debbytomlinson6220) → Yes. See the discussion in Step 2. Grant recommends project-specific instructions over account-wide ones.
  • "Can you post links to prompt engineering videos and tools?" (@trumpetguy0912) → See All Resources and Links at the bottom.
  • "I'd like to understand more about data use and privacy when sharing documents." (@KellyH-q8z) → See Set Your Privacy Settings and the discussion about turning off training data sharing.
  • "Is a Perplexity subscription worth it?" (@DebSchiano) → See Before You Start. Great supplement for research, not a daily driver replacement. Perplexity Computer is a different story.
  • "You suggest paying for ChatGPT, what about Claude or Perplexity?" (@randycampbell-sp3fw) → See Before You Start for our full recommendation.

Projects and Memory

  • "If you have 2 separate projects, can one project know what is in the other?" (@petercarnevale6764) — No. Projects are isolated by design. A job search project can't see what's inside your writing project. If you need information shared across projects, upload the same reference files to both, or use account-wide memory (instead of project-only memory) so the AI remembers key facts about you everywhere.
  • "Can you set instructions later, after creating the project?" (@amyknapp-p4m) — Yes, absolutely. You can add, edit, or completely rewrite your project instructions at any time. Grant did this live on stream when he added formatting rules after his first results came back wrong. Just go to your project settings and update them.
  • "Does Projects retain context as we build out several script files for a coding project?" (@karthiksuri) — Yes, within a project the AI can reference previous chats and uploaded files. But for serious coding work with multiple files, Claude Code or Codex are better suited because they work directly with your local file system and track an entire codebase, not just what's in the chat window.
  • "Can the projects see the work done in Code?" (@ksmimages) — Not directly. Claude projects (chat interface) and Claude Code (terminal tool) are separate environments. If you want a project to reference code you've written, upload the relevant files to the project's knowledge base.
  • "Does Gemini Pro have project memory?" (@ninedesign108) — Gemini Gems let you set persistent instructions and upload knowledge files, but they don't organize your chat history the way Claude and ChatGPT projects do. You can't see all your previous Gem conversations in one place. If that organizational feature matters, Claude or ChatGPT is the better pick.
  • "If you make a table in a project, can you have the agent remember it?" (@katlinkavangh5873) — Yes. If the table is in your project knowledge (uploaded as a file or pasted as text content), the AI will reference it in every chat. If you created the table inside a conversation, it lives in that chat's history. On Claude, it can even search across past chats in the same project.
  • "How thorough do Project Instructions need to be before they drift? How do you prevent drift?" (@edoardorojas276) — Long conversations will drift no matter what. Three practical fixes: keep instructions focused and concrete (not vague), start a new chat when the conversation gets long (project instructions reset with each new chat), and periodically paste in a reminder like "re-read the project instructions and make sure you're still following them." Break large tasks into shorter chats rather than running one mega-conversation.
  • "Token usage: adding files to Projects versus adding files in the chat window. Is there a difference?" (@MarcusStone-sj8jq) — Grant addressed this on stream: token usage is the same either way. Whether a file lives in the project knowledge or is uploaded directly into a chat, it uses the same amount of context. The difference is convenience; project files persist across every chat so you don't re-upload them.
  • "Would you say Projects in Gemini is NotebookLM for a more controlled experience?" (@scottS5239) — Different tools. Gemini Gems are custom AI assistants with persistent instructions and uploaded knowledge, similar to ChatGPT custom GPTs. NotebookLM is for deep analysis of specific source documents (it generates podcasts, study guides, and Q&A from your uploads). Use Gems when you need a reusable assistant. Use NotebookLM when you need to deeply understand a set of documents.
Advertisement

Privacy and Data

  • "I'm nervous about creating any AI account due to privacy. It needs access to all personal information. Thoughts?" (@hilarmy-g4q) — You don't have to share everything. Start with a free account and don't upload anything sensitive while you get comfortable. The privacy settings we covered (turning off training data sharing) are your first line of defense. For serious work with sensitive data, team and enterprise accounts offer stronger protections. If privacy is your top concern, tools like LM Studio run entirely on your computer with zero data leaving your machine.
  • "If I create anything of value on any AI platform, do I own it?" (@seandeadman6000) — Generally yes. Both OpenAI and Anthropic's terms of service assign ownership of output to the user. But this is a legal question that varies by jurisdiction, and we're not lawyers. If you're building something commercially valuable, get proper legal advice.

Migrating Between Tools

  • "How do I move individual ChatGPT chats to a Claude Project?" (@joematurojr) — No direct one-click migration between platforms. The practical workaround: export your ChatGPT data (Settings > Data Controls > Export), then upload the relevant conversation files to your Claude project's knowledge base. Or just copy-paste the key content from your best conversations into Claude as text files. It's manual, but it works.
  • "I've created a prompt engineering system in ChatGPT. Thinking of switching to Claude." (@edoardorojas276) — Copy your GPT instructions and paste them into Claude project instructions. The syntax is slightly different (Claude responds well to XML tags for structuring complex instructions), but the core logic transfers. You might need to adjust tone and formatting preferences, but the foundational instructions carry over.
  • "I have several GPTs from ChatGPT. Would I copy the GPT instructions into the Project instructions in Claude Desktop?" (@debbytomlinson6220) — Yes, that's exactly right. Open each GPT in ChatGPT, copy its instructions, and paste them into a new Claude project's custom instructions field. You'll also want to re-upload any knowledge files the GPT relied on. The behavior won't be identical (different models have different strengths), but the core workflow transfers cleanly.
Advertisement

Specific Use Cases

  • "What tool or platform is best for statistics/data pattern detection?" (@rebeccaprosise3534) — For general data analysis, ChatGPT's Code Interpreter (built into the paid tier) is excellent; you upload a dataset and it writes Python code to analyze it on the fly. For more specialized stats work, Databricks and Glean are worth exploring. For academic research, Claude is strong at interpreting complex statistical findings in plain language.
  • "Is Perplexity a good option for marketing analysis?" (@hilarmy-g4q) — Great for competitive research and trend monitoring because of its search capabilities. For actual marketing strategy and content creation, ChatGPT or Claude give you more depth. Use Perplexity to gather the raw research, then bring those findings into a ChatGPT or Claude project for analysis and strategy.
  • "How to use AI to auto fill leases and related documents?" (@BillMcKee) — Both ChatGPT and Claude can read document templates and fill in fields based on information you provide. Upload your lease template, provide the data, and ask it to fill it in. One important caution from another viewer (@AriG-y9j): be very careful with legal documents. AI models sometimes produce output designed to seem helpful rather than be accurate. Always have a human review any legal document before signing or sending.
  • "I am looking for a way to organize a massive amount of photos. Can I upload pics and ask to organize by xyz?" (@hilarmy-g4q) — ChatGPT's Library feature helps with photos generated inside ChatGPT. For organizing existing photo libraries, AI-powered tools like Google Photos or Adobe Lightroom's AI features are better suited than a chat-based AI. You could upload a sample of photos to Claude or ChatGPT and ask it to suggest an organizational system or tagging structure.
  • "I make the same posters each year with new images and text. What's the best way?" (@ConfusedaboutAIAdmin_Help) — Perfect use case for a Gemini Gem or a Claude project with a skill. Upload your previous poster examples as reference images. Write instructions describing the layout, colors, fonts, and style. Then each year, just say "make this year's poster with [new text] and [new image]." This is exactly the workflow Grant showed with the header image creator Gem.
  • "Can you use AI to set reminders?" (@katlinkavangh5873) — Yes, through scheduled automations. In Claude's Cowork, you can set a recurring task that runs at a specific time. Codex can do the same. These aren't simple calendar reminders; they're full AI tasks that run on a schedule. For basic reminders, your phone is still easier. For "remind me by doing this complex task every Monday at 9am," that's where AI automations shine.
  • "What file type do you need for transcription?" (@DebSchiano) — Most common formats work. ChatGPT and Claude both accept audio files (MP3, WAV, M4A, MP4) for transcription on paid tiers. For text transcription from images or PDFs, just upload the file directly. Claude's vision capabilities can also transcribe text from screenshots and photos, though accuracy varies.
  • "Is Claude better or best for job search?" (@tosha50000) — One viewer (@edoardorojas276) gave a great breakdown in the chat: use ChatGPT to strategize which companies to target (where, why, how), Perplexity to research each company and role, and Claude to update your resume and write tailored cover letters. Our full demo in this guide uses Claude for the end-to-end workflow, but the best approach uses multiple tools for what each does best.

Tools and Features

  • "Can Cowork access emails?" (@DebSchiano) — Yes. Claude has a native Gmail connector that lets you search and read emails, draft replies (though it can't send on your behalf; you review and send manually), and use email content as context for other tasks. To set it up: go to Customize > Connectors, find Gmail, connect your Google account. Cowork also supports Outlook via third-party MCP connectors.
  • "Claude Cowork is only available on Mac, not on Windows!" (@Achillus009) — This was true until recently. As of about three weeks before this stream (mid-March 2026), Claude's desktop app (including Cowork) is available on Windows. Download it here.
  • "I'm not finding Skills on ChatGPT!" (@pischdoa) — Skills in ChatGPT are currently available on Business accounts and through the Codex desktop app. They haven't rolled out to the standard consumer Plus plan yet, but should be coming soon. In the meantime, you can use Codex (free to download, connects to your ChatGPT account) to create and manage skills. Or if you just want to try making one, click here for a one-click prompt that walks you through it.
  • "Do we need to set up skills? I use Claude for genealogy research and have had good success." (@lindabright4609) — No, skills aren't required. If your current workflow is working well without them, that's fine. Skills become valuable when you find yourself repeating the same complex prompt or workflow. If you're doing the same type of genealogy query over and over (searching census records, formatting family trees, analyzing DNA matches), that's when packaging it into a skill saves real time. But if each query is unique, you may not need one.
  • "Have you noticed the frontier LLMs have slowed down the last week?" (@pauloyler1975) — Grant addressed this on stream: there is definitely a compute shortage right now, particularly impacting Anthropic. When demand outpaces server capacity, all users experience slower responses. Paid users get priority, but even they notice it during peak hours. This is a temporary infrastructure issue, not a permanent degradation of the models.
  • "What is hallucination and how do you keep that from happening?" (@philgood1960) — Hallucination is when the AI confidently generates information that isn't true. It'll cite fake sources, invent statistics, or claim something happened that never did (we saw this live with the Opal demo hallucinating a video title). To reduce hallucinations: turn on web search so the AI is pulling from real sources, ask it to cite its sources so you can verify, and always fact-check anything important. Using Thinking mode also helps because the AI reasons more carefully before responding.
  • "How do you get ChatGPT to stop answering in bullet points?" (@rrr2rrr) — Add this to your personalization settings or project instructions: "Write in natural paragraphs. Do not use bullet points or numbered lists unless I specifically ask for them." You can also say this in individual prompts. The AI defaults to lists because most people find them scannable, but it'll follow your formatting preferences if you tell it.
  • "What is the advantage of Cowork over regular chat for most Claude tasks?" (@TheIainMorrison) — Cowork can access files on your actual computer (not just files uploaded to the chat), schedule recurring tasks, and use computer use (clicking through apps on your behalf). Regular chat with connectors handles most day-to-day tasks. Think of it this way: if you need Claude to work with files you haven't uploaded, run tasks while you're away, or interact with apps beyond what connectors support, use Cowork. For everything else, regular chat is fine.
  • "Can skills include instructions to call other tools to perform specific tasks?" (@Alex-t6t5t) — Yes. Skills can include tool calls, connector usage, and multi-step workflows. A skill could tell Claude to search the web, pull data from a connected app, process it, and format the output, all in sequence. The skill file defines the full workflow including which tools to use and when.
  • "How would you compare Claude Projects vs NotebookLM?" (@frotastic6251) — Different tools for different jobs. NotebookLM is for deep analysis of specific documents; you upload sources and it becomes an expert on those sources (great for research, studying, and podcast generation). Claude Projects are a persistent workspace for ongoing tasks across many conversations. Use NotebookLM when you need to deeply understand a set of documents. Use Claude Projects when you need an AI coworker who remembers your context and can take actions for you.
  • "Why don't you use or speak to DeepSeek?" (@Donalee-Visionary) — Grant addressed this briefly on stream: it's behind the curve right now relative to the top four. DeepSeek made a big splash earlier this year with impressive open-source models, and they're worth watching. But for the daily driver recommendation we're making to beginners, the app experience and consistency matter, and that's where ChatGPT, Claude, Gemini, and Grok are ahead.
  • "Is the download for Projects only available in paid plans?" (@davidbaines4136) — As of 2026, ChatGPT Projects are available on all plans, including free. OpenAI expanded access so that free users can create projects, add instructions, and organize chats. The desktop apps for both ChatGPT and Claude are also free to download. Where the paid tiers matter is rate limits (how much you can use it before getting throttled), access to the most intelligent models, and advanced features like skills, Thinking mode, and Deep Research.
  • "Is Opal free?" (@DebSchiano) — Opal is accessible through Google accounts, but Grant noted on stream that the models powering it may require API calls or a Google Workspace account for higher limits. The basic functionality appears to be free with a Google account, but you may hit usage caps quickly on the free tier.
  • "What's the best platform for getting certified in the AI space?" (@jnwatts) — There's no single "gold standard" AI certification yet. Google offers a Generative AI learning path, Coursera has the Andrew Ng courses (widely respected), not to mention Andrew Ng's own Deep learning platform, and IBM has AI certifications through Coursera as well. For practical skills rather than certifications, honestly, following along with guides like this one and building your own projects will teach you more than most courses. And stay tuned; we might have something to announce on this front soon.
Advertisement

For Copilot Users

Several people in the chat asked about Copilot equivalents, so let's address this properly.

  • "Will this be relevant to people forced to use Copilot for work?" / "What's the Copilot equivalent of a Project?" / "Are Projects the same as Notebooks in Copilot?" (@trumpetguy0912, @ManageforGrowth)
  • Yes, nearly everything we covered translates to Copilot, though the interface and naming are different. Here's the mapping:
  • Copilot Agents (formerly part of Copilot Studio) are closest to ChatGPT and Claude projects. You create a persistent AI assistant with custom instructions, uploaded knowledge, and specific behaviors. This is where you'd build something like the job search manager we demoed.
  • Copilot Notebooks are workspaces for saving context, code, and documents. Think of them as the "files and knowledge" side of a project.
  • What Copilot does that ChatGPT and Claude can't: Copilot is deeply integrated into Microsoft 365 through something called Microsoft Graph, which means it automatically knows who you work with, what meetings you attended, which files are relevant, and what projects are in flight. You don't have to upload context manually; it can pull from your Outlook, Teams, SharePoint, and OneDrive automatically. That's a huge advantage for enterprise workflows.
  • What Copilot can't do as well: It's weaker at free-form reasoning, creative tasks, and long-form writing compared to ChatGPT or Claude. Starting with Wave 3 (March 2026), Microsoft added multi-model architecture where Copilot can use Claude for critique and verification alongside GPT models, which should help.
  • The pricing catch: The consumer version of Copilot is free and basic. The real power is in Microsoft 365 Copilot, which requires a Microsoft 365 license on top of the Copilot subscription. The real cost is $50-80+ per user per month combined. ChatGPT Plus is $20/month standalone with no additional license needed. So if your company already pays for M365, Copilot is an incremental add. If they don't, ChatGPT or Claude is significantly cheaper.
  • Our recommendation for Copilot users: Use Copilot for tasks inside your Microsoft workflow (email summaries, Teams recaps, document drafting in Word, Excel formulas). Use ChatGPT or Claude for everything else (research, creative work, coding, building skills and automations). Many organizations use both.

Free vs. Paid: What Do You Actually Get?

"What is the difference between free and monthly subscription?" / "What more do subscriptions offer?" (@DebSchiano)

This came up multiple times, so here's the real breakdown for each platform as of April 2026:

ChatGPT:

  • Free ($0): Access to the same base model (GPT-5.2) as paid users, but capped at roughly 10 messages every 5 hours. After that, you're switched to a less capable "mini" model. No Thinking mode, no Deep Research, no custom GPTs, no Codex agent. Limited file uploads (3 per day). May get throttled during peak hours. OpenAI started testing ads on the free tier in 2026.
  • Go ($8/month): 10x the free limits on messages, image generations, and file uploads. Unlimited access to the base model. But still no Thinking mode, no Deep Research, no custom GPTs, no agent features. May include ads.
  • Plus ($20/month): 160 messages per 3 hours with the standard model. 3,000 Thinking-mode messages per week. Deep Research, custom GPTs, Codex agent, voice with video, image generation, and Projects. This is the sweet spot for most people.
  • Pro ($200/month): Unlimited access to the most powerful model variant. No rate limits. Maximum context window. For professionals who use it all day.

Claude:

  • Free ($0): Access to Claude Sonnet 4.5 (the mid-tier model, not the flagship Opus). Roughly 30-100 messages per day depending on demand, with 4-8 hour resets. Projects and Artifacts are included (this changed in 2026). No Claude Code. No priority access during peak hours. Searching past chats is limited to paid plans.
  • Pro ($20/month): 5x the usage of free. Access to Claude Opus 4.6 (the flagship model). Priority access during high traffic. Claude Code (the terminal coding agent). Extended thinking. All connectors and skills.
  • Max ($100/month): 5x Pro's usage. Higher priority. More Claude Code capacity. This is what Grant uses and what he recommends for serious work.
  • Max 20x ($200/month): 20x Pro's usage. Essentially unlimited for most workflows.

Gemini:

  • Free ($0): Access to Gemini models with usage limits. Gems (custom assistants). Basic search.
  • Advanced ($20/month): Higher limits. Access to the most powerful Gemini model. Deep Research. NotebookLM integration. 2TB Google storage.
  • Ultra ($250/month): The most powerful tier with the highest model intelligence. Deep Think reasoning. Maximum limits.

The bottom line: On every platform, free gives you the same base intelligence with strict usage caps and no advanced features. The $20 tier on any platform removes most caps and unlocks the tools that make AI genuinely useful for work (projects, skills, thinking mode, agent features). The difference between free and $20 is enormous. The difference between $20 and $100+ is mostly about how often you hit rate limits.

Advertisement

Migrating Off OpenAI: Does Organizing Into Projects Help?

"Before migrating off OpenAI, is it best to organize all your chats clearly into Projects?" (@lexymartin5153)

We looked into this, and the honest answer is: organizing into projects doesn't change the export format at all. When you export your ChatGPT data (Settings > Data Controls > Export Data), you get a ZIP file containing one big conversations.json file and a chat.html file. Every conversation is dumped into that single JSON file regardless of which project it lives in. There's no project-level export, no way to export just one project's conversations, and no special organization in the export based on your project structure.

So what should you do instead?

  • Step 1: Export your data from ChatGPT. Go to Settings > Data Controls > Export Data. You'll get an email with a download link (can take up to 7 days, but usually arrives within hours).
  • Step 2: Use a tool to organize the messy export. There's a free open-source tool on GitHub called chatgpt-to-claude that parses the messy conversations.json and generates one clean Markdown file per conversation, organized into monthly folders. You run one Python command and get a folder structure like 2026-01/001_My_first_topic.md. No external libraries needed.
  • Step 3: Upload the organized files to your new platform. Take the Markdown files that matter most and upload them to a Claude project's knowledge base (or whatever platform you're moving to).

That said, organizing your chats into projects before you migrate is still smart; not for the export, but because it forces you to identify which conversations actually matter. Most people have hundreds of throwaway chats mixed in with the 20-30 conversations that contain real value. Sorting into projects helps you figure out which ones to bring over and which ones to leave behind.

Have a question we still didn't cover? Submit it through our AI Skill of the Day request form and we'll answer it in the newsletter or a future livestream.

One More Thing

We're launching a robotics newsletter (8:05) in the next few weeks. If AI stuff is interesting to you and you want to keep track of how robot progress is going (and how soon they'll be able to replace your entire physical body; lol JK WE HOPE), sign up here for early access.

We're also starting a new Monday morning show, AI News and (something goofy that changes every week) (2:22:18). We won't have a Thursday night live next week, but keep an eye out.

And there may be a cool announcement about beginner AI education coming in a few weeks. We can't say more than that yet. Stay tuned.

Everything referenced in the stream and this guide:

The original article:

Getting started:

Skills and prompt engineering:

Comparing models:

Running models locally (we didn't talk about this, but we talk about it a lot; this is an option if you want free / local AI access that doesn't go over the cloud):

Have a question we didn't answer? Submit it through our AI Skill of the Day request form and we'll cover it in an upcoming newsletter or livestream.

Want to stay in the loop? Subscribe to The Neuron for daily AI news, tools, and tips delivered to 700,000+ inboxes every morning.
Grant Harvey

Grant Harvey is the Lead Writer of The Neuron, where he continues to lead the publication's daily coverage of AI news, tools, and trends.

The Neuron Logo

Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.

linkedin
instagram
x
facebook
youtube
rss

Company

Contact us Advertise with us Newsletter Free ChatGPT Course Write Anything (with AI)

Categories

Explainers

Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved

Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.

Terms of Service Privacy Policy California - Do Not Sell My Information