You never have to choose between "use AI" and "stay safe." You just choose how.
Workarounds that actually work
| Problem | Workaround |
|---|---|
| Long doc, need a summary | Summarise one section yourself; paste only that, plus "no names in this excerpt" |
| Real client scenario | Invent Example Ltd with fake numbers; say "fictional" in the prompt |
| Tone help on a sensitive email | Describe the situation without names: "email to a late-paying customer, firm but polite" |
| Worried about training | Turn training off in settings and avoid pasting red-tier data anyway |
| Genuinely private material | Don't use cloud at all — running AI on your own PC comes later in the course; until then keep it offline (local notes, pen and paper) |
Lab rules for Lessons 9 and 10
You're about to tour and compare the cloud tools. Treat it like a driving lesson on a quiet road — controlled conditions, nothing precious on the line.
| Do | Don't |
|---|---|
| Use green and anonymised amber prompts | Paste real client files, invoices or inbox exports |
| Use the same fake scenario across platforms for a fair comparison | Paste passwords "to test if it remembers" |
| Screenshot menus and features, not private messages | Upload your mate's medical letter "to see what it says" |
| Note which app felt better for which job | Treat free-tier labs as "private because I'm only learning" |
Comparison prompt example (safe):
You are helping a fictional tattoo studio "Example Ink" in Leeds. Write three social post ideas for a flash day — no real people, no prices over £200.
For the comparison lab I still use the same daft fake shop every single time. Boring for the AI; safe for me; and it gives me consistent results to compare across ChatGPT, Claude and Gemini.
If you've already pasted something red-tier
- Delete the conversation in that app, if you can.
- Rotate anything exposed — passwords, API keys.
- Don't panic. Learn the rule; perfection was never the goal.
Continue — lock in your three rules.