I’ve been cataloguing AI tools for a while, and ChatGPT is the one where the gap between marketing claims and daily reality is widest.
Here’s how I’d group what it’s actually used for:
Strong fit
- Drafting and rewriting first drafts, tone shifts, summarising long documents
- Code explanation and debugging reading unfamiliar code, spotting errors, writing tests
- Structured thinking outlines, decision trees, turning messy notes into a plan
- Language work translation, tone matching, learning a language through conversation
Works with supervision
- Research good at finding angles, bad at citing accurately without browsing on
- Data work fine on small clean datasets, unreliable on anything large
- Customer support drafting solid templates, poor at edge cases
Weak fit (people try anyway)
- Factual lookup without search recall is unreliable
- Anything needing current data the cutoff is a hard wall
- High-stakes judgement legal, medical, financial
Two questions for people here:
- What use case surprised you by working better than expected?
- Where have you hit a wall you couldn’t prompt your way around?