Phase 1 · ChatGPT · Level 3 · Power User
ChatGPT: Power User Quiz · 10 min
Why it matters A quick check of your judgement before you move on. There's no penalty for a low score, and you can retake it, or take it first to test out of a level you already know.
This level was about wiring ChatGPT into real work: custom GPTs, scheduled Tasks, agents, connectors and the security to hold it all safely. These questions are the judgement calls that decide whether an automated, connected setup actually helps you or quietly works against you. None are trivia; each is a decision you'll face once ChatGPT is doing real jobs on your behalf. Pick the best option, and you'll get an explanation after each.
Passing this checkpoint: work through the level first, then score 70% or more here.
Skipping this level: already confident? Take this cold and score 80% or more to test out and jump ahead. Below that, nothing is lost, you just study the level as normal.
1. You're building a custom GPT to draft replies to customer feedback. You have a two-page document of response rules (what you can offer, standard timeframes, tone). Where should the rules go? Upload the whole document as a knowledge file; that's what knowledge is for. Put the actual rules, tone and workflow in the GPT's instructions, and use knowledge only for reference material it draws on. Paste the rules into every chat with the GPT as you go. Split them randomly between instructions and knowledge so both are covered. 2. Your new feedback-responder GPT wrote a lovely reply in your test. Before you share it with the team, what's the most important next step? Share it; a good reply proves it works. Test it on awkward cases (ambiguous, out-of-scope, hostile) to see whether it holds its boundaries or invents commitments. Add more example replies to the knowledge file until it's perfect. Rewrite the instructions to be longer and more detailed. 3. You set up a recurring Task to send you a Monday prep note. On a quiet week with little new, it produces a confident-looking note full of items that are actually just restated old business. What's the underlying problem? The Task is broken and should be deleted. The instruction lacks an honesty guardrail, so on a quiet week it pads rather than saying 'nothing new to flag', and you should fix the instruction. Tasks can't tell busy weeks from quiet ones, so this is unavoidable. You should run the Task more often so it has more to report. 4. You give Agent Mode a task: 'compare these three suppliers and put the pricing in a table.' Which briefing detail matters most for keeping it safe? Telling it to work as fast as possible. Adding a stop rule: don't submit forms, create accounts or buy anything; pause and ask before any consequential action. Asking it to use as many sources as it can find. Letting it decide the table columns itself to save you time. 5. Midway through an agent run that's browsing the web, the agent proposes an action you never asked for: emailing someone, or entering details into a site. What's the most likely explanation and the right response? It's being helpfully proactive; let it continue. It may have hit a hidden instruction on a page (prompt injection); treat the unexpected action as a red flag, stop, and don't approve it. The agent is malfunctioning and you should restart your computer. It's normal behaviour; agents always send emails as part of research. 6. You want to use Record Mode to capture your weekly team meeting so you can extract the actions. What must you do first? Nothing; you're a participant, so you can record freely. Tell everyone they're being recorded and get their agreement, and check your organisation's policy and local law. Record quietly so you don't disrupt the meeting. Get consent only from the most senior person present. 7. Record Mode transcribes your meeting and produces a clean, well-written summary. You're about to send a follow-up email based on it to the whole team. What's the sensible check? Send it; the transcript is automatic, so it's accurate. Verify the load-bearing specifics (names, numbers, dates, and any 'not' that flips a meaning) against the transcript before sending. Trust it because the summary reads smoothly and professionally. Ask ChatGPT whether the summary is correct and trust its answer. 8. You connect ChatGPT to your Google Drive and ask it to summarise 'the project brief'. It confidently summarises a document. Before acting on the summary, what's the key habit? Trust it; the answer came from your own Drive, so it's reliable. Have it confirm the exact file name and last-modified date, since a connector can grab a similarly-named or outdated file. Assume it picked the newest file automatically. Disconnect the Drive so it can't make mistakes. 9. You use a personal ChatGPT account for work. You turn off 'Improve the model for everyone'. Does that mean you can now paste confidential client data in freely? Yes; with training off, your data is private and safe to share. No; turning off training reduces how your content is used, but you're still sending it to an external service, so Phase 0 privacy rules and your employer's policy still apply. Yes, as long as you also use Temporary Chat. It doesn't matter either way; the setting has no real effect. 10. You've wired ChatGPT deeply into your work: custom GPTs, connectors into your Drive and email, months of context in memory. What single security step protects all of that most? Choosing a longer password. Turning on multi-factor authentication, so a stolen or leaked password alone can't get anyone in. Logging out after every session. Turning off memory so there's nothing to steal. Answered 0 of 10.
Answer every question to see your score Take the one-page cheat sheet with you Everything from this phase, distilled to a single page you can print or keep open while you work.
Open the ChatGPT cheat sheet