What this is

Fifty ways to actually use ChatGPT in the GPT-5 / agent-mode / Sora 2 era — each one with the exact prompt and a copy button. It's the companion library to my "50 Ways" video: the demo below is the library, embedded live. Grab any prompt, change the one thing that matters to you, paste it in.

It's organized as seven category acts plus a wildcard closer, and every entry carries a badge for where it came from:

  • SHOW — ways I already use it in my own work (transcript-to-agent specs, styled decks, second-brain notes, single-file apps, the blindspot pre-mortem), recast as ChatGPT prompts.
  • PROMPT — the ones worth running yourself, including the prompts OpenAI itself pitches in its docs and launch posts (deep research, study mode, agent mode, Sora's cinematographer prompt).
  • CLIP — how other people used it, credited to the creator, with the exact prompt they shared (the AI-found Linux zero-day, the hospital-bill negotiation, Mollick's one-prompt startup, the Sora 2 bodycam genre).

Why it's built this way

Most "ways to use ChatGPT" lists are useless because they stop at the idea. "Use it to study" tells you nothing; the study-mode prompt a real student used to run a full game-theory curriculum tells you everything. So every one of the fifty ships the literal, paste-ready text — no paraphrasing.

And every one is sourced. The ones I use, I built. The ones OpenAI pitches, I pulled from their own docs and launches. The viral ones are credited to the person who made them, with the prompt they shared. You always know how much to trust it and where to go deeper.

A few that land every time

  • "Tell me one thing about myself I don't know." The memory-era party trick — Sam Altman quote-tweeted it. The answer is uniquely about you.
  • Agent mode does the grocery run. Hand it a recipe, watch it fill an Instacart cart on its own. The clearest "it does things now instead of just saying things" moment.
  • Red-team your own idea. The antidote to a model that just agrees with you — make it attack the plan before the world does.
  • The 95%-confidence prompt. One line that makes ChatGPT interview you before answering, which kills most hallucinated assumptions.
  • GeoGuessr from any photo. Jaw-dropping — and a built-in reminder to be careful what you post.

Use it, don't just watch it

The whole point is the copy button. Open the library, find the row for the thing you actually need to do this week, copy the prompt, swap in your details, and run it. If one saves you an hour, it earned its place on the list.

Dated July 2026 — the frontier moves fast. Half of these didn't exist a year ago, and some feature names will keep shifting. The prompts are the durable part.