Free to read · no account needed

20 min · start

Accounts, safety, and your first good prompts

Goal: Open ChatGPT, Claude, or Grok safely—2FA and revoke over-broad connectors before any send that moves real work.

Best for: Absolute beginners creating their first ChatGPT/Claude/Grok account this week.

Who this is for: first week with AI chat tools. No coding required.

Free to read: this page is free to read. No account on iswarp is required.

Why beginners skip this

Accounts ready but the safety list skipped feels fast—habits never land, accidents start week one.

1. Pick one tool for week one

Pick one primary tool for week one—do not open five accounts at once.

If you mostly need…Start with
Everyday drafting, brainstorming, mixed tasksChatGPT
Long documents, careful rewrite, code reviewClaude
Timely public topics, informal brainstorming, X-context questionsGrok

You can switch later. Habit beats shopping.

Official entry points (verify URLs yourself): chatgpt.com, claude.ai, grok.com / Grok inside X.

2. Lock the account before you paste work

Finish strong password + 2FA before you paste any work into the chat box.

1. Unique password in a password manager. Never reuse the password from email or banking. 2. Turn on MFA (authenticator app or passkey). SMS-only is better than nothing; app/passkey is stronger. 3. Review active sessions and sign out devices you do not recognize. 4. If the product offers stronger enrollment (for example OpenAI’s Advanced Account Security for eligible consumer accounts), read the recovery trade-offs before you opt in—losing recovery keys can lock you out.

If you suspect a compromise: change the password, sign out everywhere, revoke API keys if you ever created any, and contact the vendor’s support from their official site—not from a cold DM.

3. Decide what never goes in the chat box

Passwords, OTPs, live keys, and unredacted customer PII never go in the chat box.

  • Passwords, one-time codes, API keys, or wallet seeds
  • Full ID numbers, bank details, medical records, or unredacted contracts
  • Private customer lists, unpublished financials, or anything your employer forbids sharing with third-party tools

Safer pattern: redact names → “Client A in industry X”—ask for structure and language, then fill secrets yourself offline.

When a product offers temporary / incognito chat, use it for sensitive drafts so history and memory stay cleaner. Still assume the vendor may retain data for abuse monitoring per their policy—redaction is your first line of defense.

4. Check training, memory, and retention once

Open training/memory/retention controls once and set them to match company policy.

  • Model improvement / training opt-out — whether new chats can help train models
  • Memory / personalization — whether the assistant stores facts about you across chats
  • History export / delete — so you can remove old threads

Open settings once, set a default you are comfortable with, then move on. Revisit when your job changes (personal use vs. client work).

For workplace use, prefer a team / enterprise plan your org administers. Personal free accounts are fine for learning public skills—not for confidential company data.

5. Your first useful prompt (copy and adapt)

Write your first reusable prompt as Goal / Format / Boundaries and run one real task.

Better—use four lines: Goal · Context · Format · Boundaries.

Goal: Rewrite this email so the recipient knows the decision and the next step.
Context: We delayed the Friday launch by one week; engineering needs a freeze date.
Audience: Product manager I work with weekly; tone: clear, calm, no blame.
Format: 120 words max, subject line + body, one bullet list of next steps.
Boundaries: Do not invent dates or owners. Mark any assumption with [?].

Draft:
---
[paste your rough draft here]
---

Run it once. Then send one follow-up only:

Example

Filled prompt

Goal: Rewrite this email so the recipient knows the decision and the next step.
Context: We delayed the Friday launch by one week; engineering needs a freeze date.
Audience: Product manager I work with weekly; tone: clear, calm, no blame.
Format: 120 words max, subject line + body, one bullet list of next steps.
Boundaries: Do not invent dates or owners. Mark any assumption with [?].

Draft:
---
Launch might move, pls sync somehow, thx
---

What good output looks like

  • Subject + body ≤120 words with clear decision
  • Next steps as bullets; unknowns marked [?]
  • No invented freeze date or owner
Keep the same facts. Make the subject line more specific. Cut 20%.

That loop—clear first ask, then a small edit—beats rewriting the whole prompt from scratch.

Example

Filled prompt

Job: Finish account/safety step for “Accounts, safety, and your first good” before send scopes.
Boundaries: draft only; Unknowns [?]; no silent send; no invented metrics.
Return: checkable artifact + one-minute pass/fail for start-accounts-and-safety.

What good output looks like

  • Matches named constraints for “set up ChatGPT, Claude, or Grok safely, and send”
  • Unknowns labeled; no invented owners/metrics
  • Stops before irreversible actions
  • Pass/fail markable in ≤60s

6. A five-minute practice drill

Five minutes: revoke one over-broad connector, or add “do not invent” to one prompt. 2. Fill the Goal / Context / Format / Boundaries template. 3. Paste the result into a doc and edit by hand for five minutes. 4. Save the prompt that worked in your notes for next time.

You are not “learning AI.” You are building a reusable work habit.

Common pitfalls

PitfallFix
Pasting secrets “just this once”Redact first; use temporary chat when available
Trusting every citation or dateAsk for sources; open the links yourself
Switching models every hourFinish one week on one tool
One giant prompt for everythingSplit: research → outline → draft → polish
Leaving MFA offEnable it before uploading files

MFA today

Enable MFA before you explore fancy features. Account recovery is harder than prompt craft.

Example

Filled prompt

Tonight: 2FA on, privacy/training toggles reviewed, never-paste list of 5.
Do not ask chat for live prices.
Context: real workstream this week (Asia/Shanghai). Name actual files/people.
Constraints: no invented metrics/owners/prices; label Unknowns; stop before send/delete/pay.
Done check: I can mark pass/fail in one minute on the output.

What good output looks like

  • Accounts/safety checklist done for “set up ChatGPT, Claude, or Grok safely, and send o”
  • No send scopes before Done written
  • Kill-switch / password hygiene noted
  • Pass/fail: checklist dated

Red flag

Accounts ready with the safety list skipped leaves habits unlanded—pin the list first.

Save this card

Card: 2FA and revoke over-broad connectors before any send—the safety checklist is not optional. Keep this beside the tool; read before you start.

Before you start

Have password manager and 2FA ready. For “set up ChatGPT, Claude, or Grok safely, and send one prompt ,” finish identity before any send-capable connector Guide: start-accounts-and-safety Keep the live artifact visible the whole time. If you only have a blank box, stop and fetch the real job first

How to use this week

This week pin a never-paste list on the desk: keys, ID numbers, client names; run one fake-paste drill before work.

Practice now

Tonight pin a never-paste list on the desk: keys, ID numbers, client names; run one fake-paste drill.

What to do next

1. Finish account MFA + privacy defaults today. 2. Run the email (or outline) prompt on a real task. 3. Read How to choose a model when you are ready for a second tool. 4. Return to Learn for writing / research / code paths as they ship.

Close the loop

Write one line: next time I start work with the safety list skipped, I refuse.

One sentence takeaway

Pin the never-paste list first; fake-paste drill passes before work.

Next

Path: Your first 7 days · How to choose a model · Clear everyday prompts

← Back to home · Learn · Tools · Compare · Start