ChatGPT App & GPTs
Learning Outcomes
- Install and sign in to the ChatGPT desktop app on your platform
- Compare the subscription tiers and choose the right one for your needs
- Switch between models using the model picker for different task types
- Manage ChatGPT's memory so it remembers what you want and forgets what you don't
- Discover, evaluate, and use custom GPTs from the GPT Store
Lesson Plan
| Segment | Duration | Topic |
|---|---|---|
| Intro | 3 min | The ChatGPT ecosystem at a glance |
| Demo 1 | 6 min | Installing the desktop app and signing in |
| Explain | 6 min | Subscription tiers compared |
| Demo 2 | 6 min | The model picker and choosing a model |
| Demo 3 | 6 min | Managing memory and chat history |
| Demo 4 | 7 min | The GPT Store and using custom GPTs |
| Explain | 3 min | When a GPT beats a plain prompt |
| Wrap-up | 3 min | Key takeaways, preview Lesson 4 |
Before You Begin
Pre-work:
- Have an OpenAI account (free to create at chatgpt.com)
- Complete Lesson 1: Claude Desktop Setup so you can compare the two apps side by side
- Skim the feature evaluation reference — you'll apply that judgement when picking GPTs
Shopping List:
- A computer (macOS or Windows) with internet access
- An email or Google/Microsoft/Apple account for sign-up
- A real task in mind — a document to draft, a topic to research, or data to think through
- Optional: a payment method to trial a paid tier
ChatGPT lives in the web app (chatgpt.com), the desktop app, the mobile app, and inside custom GPTs. The desktop app is the power-user home base — it launches with a hotkey and runs tool-heavy tasks more smoothly than a browser tab. Download the installer from chatgpt.com/download, then sign in.
The install differs slightly by platform:
Open the .dmg and drag ChatGPT into Applications. The Mac build is native — voice, trackpad haptics — and needs a recent macOS on Apple Silicon. Summon it with Option + Space.
Run the installer (or grab it from the Microsoft Store) and summon it with Alt + Space. The Windows app is a focused wrapper around the web experience.
ChatGPT sells several tiers, from a free plan up through individual paid plans and team/enterprise plans. Names and prices shift often, so anchor on what each tier unlocks.
| Tier | Who it's for | What you gain over Free |
|---|---|---|
| Free | Light use | Default model, basic memory, can use GPTs |
| Individual paid | Daily power users | Higher limits, full model picker, advanced tools, can build GPTs |
| Higher individual | Heavy/professional use | Largest limits and context window, top reasoning model |
| Team / Business | Small teams | Shared workspace, admin controls, no training on your data by default |
| Enterprise | Large organisations | SSO, advanced security, custom data controls, seat minimums |
The free tier is genuinely useful, but it withholds two big things: higher usage limits and the ability to create your own GPTs — both arrive on the first paid individual tier.
ChatGPT no longer runs one model. The current generation is a small family: a fast "Instant" model (the default for all logged-in users), a slower "Thinking" model for hard reasoning, and a top "Pro" model for the most demanding work. On paid tiers a model picker sits above the message box — click the name to switch.
| If your task is... | Reach for... |
|---|---|
| Quick answers, drafting, rewriting, everyday Q&A | The default fast / Instant model |
| Multi-step logic, tricky maths, careful planning | The Thinking / reasoning model |
| The hardest, long-running, high-stakes work | The top Pro model (highest tiers only) |
Feel the difference — send this to two models and compare. The fast model answers quickly; the reasoning model visibly works through the constraints and produces a tighter plan:
> Plan a three-day, two-person trip to Lisbon on a £600 budget excluding flights. Account for travel between sights, opening days, and one rest afternoon. Show the reasoning behind each choice.
ChatGPT's memory has two distinct layers, and the difference separates a casual user from a power user:
- Saved memories — an explicit, editable list of facts ChatGPT chose to keep (your name, your job, your preferred spelling). You can read it and delete any entry.
- Reference chat history — an implicit recall of patterns from past chats. Helpful, but not directly inspectable.
Control both from your profile picture, then Settings, then Personalization. Memory is on by default across the main tiers; you can toggle either layer off, delete individual memories, or clear them all.
Steer memory deliberately — tell it what to keep, and what to drop ("Forget what you saved about my holiday plans"):
> Remember that I write in British English, prefer concise bullet points over prose, and work in financial services so I need a compliance-aware tone.
For a clean, un-personalised conversation — researching a gift, or testing a prompt with no baggage — start a Temporary Chat. It neither reads nor writes memory.
A GPT is a packaged ChatGPT pre-configured for one job: fixed instructions, optional knowledge files, specific tools, sometimes connections to outside services. The GPT Store is the searchable directory of GPTs built by OpenAI and the community — open it from the left sidebar (GPTs or Explore GPTs) for a search bar and categories like Writing, Productivity, and Research. Anyone, including free-tier users, can use public GPTs; only paid subscribers can build them.
Work through this discovery routine:
- Search a category that matches a real task ("research", "spreadsheet", "logo").
- Open a GPT's profile and check its description and example prompts (what it's tuned for), rating and conversation count (a rough quality proxy), builder (official and verified are safer), and whether it asks to connect an external service.
- Click Start Chat and give it a genuine task; log the keepers in the cheat sheet.
Try three specialised GPTs against the same brief to feel the variation:
> I have a 12-page market report I need to turn into a five-slide executive summary. Walk me through what you'd extract and how you'd structure each slide.
When a GPT genuinely saves you time, it stays pinned in your left sidebar for one-click reuse.
A GPT is just instructions plus knowledge plus tools, bundled so you don't retype them — so it's worth it only when you'd otherwise repeat the same setup over and over.
| Use a custom GPT when... | A plain prompt is fine when... |
|---|---|
| You do the same task with a fixed format weekly | It's a one-off question |
| You need it to reference the same documents | The context lives in this single chat |
| You'll share the tool with colleagues | It's only ever for you, once |
The honest trade-off: a well-written one-shot prompt frequently beats a mediocre GPT, because a GPT's value is entirely in its hidden instructions. The prompt below does in one message what a poorly-built "Proofreader GPT" might do worse:
> Act as a meticulous proofreader for British English. Fix grammar and spelling only — never change my meaning or voice. Return the corrected text, then a short bullet list of every change and why.
Questions & Answers
Key Takeaways
- The desktop app is your base — the global hotkey beats a browser tab, and every account feature is identical across web, desktop, and mobile.
- Buy capability, not a price tag — choose a plan by what it unlocks (limits, the model picker, building GPTs) and re-check pricing before you pay.
- The model picker is a choice — default fast model for most work, the reasoning model for hard multi-step tasks, accepting its tighter limits.
- Memory has two layers — an editable saved-memories list and a quieter reference-to-past-chats layer; steer both, and use Temporary Chat for a clean slate.
- GPTs are packaged prompts — use them freely, but a GPT only beats a sharp one-shot prompt when its hidden instructions are genuinely good.
- Judgement over hype — the real skill is deciding when a task is repetitive enough to deserve a custom GPT versus a well-crafted prompt.
Next Steps: Lesson 4: ChatGPT Canvas & Code Interpreter