It is 9:15 p.m. You paste a messy customer thread into ChatGPT, get a calm reply that sounds like your best support rep, and almost hit send. Then you notice three things at once: the model invented a policy refund that does not exist, the thread includes a phone number you should not have dropped into a personal Free account, and Legal has not approved this tool for customer PII. The draft was good enough to fool a tired human. That is the whole problem. ChatGPT is useful. It is not a license, a system of record, or a free pass around your company rules.
This is Part 8 of Learn ChatGPT from scratch, and it closes the series. Parts 1 through 7 covered what ChatGPT is, plans, first minutes, desktop modes, memory and Projects, multimodal inputs, and connectors/apps. Here we lock in judgment: privacy basics, work rules, when the product is the wrong tool, and a “good enough” checklist you can run in thirty seconds before you paste, upload, or connect anything. After this post, the path on AMS continues into a product map and deeper tutorials. The goal of this part is not fear. It is fewer expensive mistakes.
What you’ll learn
- How to think about privacy without pretending the UI is a contract lawyer
- Why personal plans and company workspaces are different risk stories
- Clear cases where ChatGPT is the wrong tool (and what to use instead)
- A practical “good enough” checklist for drafts, analysis, and agentic help
- How work policy, data class, and reversible stakes should gate every session
- Where this series ends and what to read next on AMS (product map, everyday, Work, Codex, Custom GPTs)
Privacy in plain English (what actually matters)
Privacy talk online often turns into rumor. Keep three separate questions, because people mash them into one:
- Who can see this content later? Chat history, shared links, workspace admins, compliance exports, third-party apps you connected.
- Is this content used to train models? Consumer defaults and business defaults differ. Settings can change. Re-check Data Controls.
- Does my employer allow this data class in this product? That answer lives in company policy, not in a shiny pricing page.
OpenAI’s public materials (Help Center and enterprise privacy pages) generally describe a split: for many business offerings (ChatGPT Business, Enterprise, Edu, and API customers under business terms), inputs and outputs are not used to train models by default. For many consumer / personal plans, content may be used to improve models depending on your Data Controls (often framed as “Improve the model for everyone”), and you can usually opt out. History retention, Memory, shared chats, and connected apps add extra paths. None of that replaces a lawyer or a DPA review for regulated industries. It is orientation so you stop treating “I have Plus” as “Legal said yes.”
Controls worth knowing by name
| Control | What it roughly does | Why you care |
|---|---|---|
| Data Controls / model improvement | Whether new chats may train models (consumer side) | Lower training exposure if you opt out; still not a work-policy waiver |
| Chat history | Whether conversations stick in your account | Easier revisit vs less residual content |
| Memory | Facts that can stick across chats | Helpful until stale or sensitive |
| Shared links | Other people can open a conversation view | Easy leak if you share the wrong URL |
| Apps / connectors | Third parties receive context under their terms | Your privacy story now includes their policies |
| Workspace admin settings | What members may use at work | The real gate for company data |
Practical habit: once a quarter, open Settings and screenshot (or write down) your Data Controls, Memory status, and connected apps. If you changed jobs, rotate. If you connected a CRM “just to try,” disconnect it.
Work rules beat product features
Companies care about data class, retention, logging, subprocessors, and who gets sued if something leaks. ChatGPT cares about being a useful product. Those are different jobs. When they conflict, work rules win if you want to keep your job and your customers.
Translate product language into policy language:
- “I can log in” is not “this data is approved.”
- “Business plan says no training by default” is not “customer contracts allow this SaaS.”
- “The model redacted the SSN in the summary” is not “the SSN never entered the system.”
- “We used a connector with Always ask” is not “write access is gone.”
If your org has an approved ChatGPT Business or Enterprise tenant, use it for work. If it does not, do not invent a shadow AI stack with personal subscriptions and shared customer exports. Escalate. The boring path is the professional one. Same story for schools and healthcare: Edu or specialized tools may exist; freestyle pasting into a personal account is still freestyle pasting.
A simple data-class ladder for sessions
| Data class (examples) | Personal Free/Plus habit | Approved company workspace |
|---|---|---|
| Public marketing copy, your own non-sensitive notes | Usually fine | Fine |
| Internal non-secret drafts (generic process notes) | Only if policy allows personal tools | Prefer company tenant |
| Customer names, tickets, contracts, pricing sheets | No, unless explicitly allowed (rare) | Only on approved paths and scopes |
| HR raw files, health, finance identifiers, secrets | No | Usually specialized systems or heavy review |
| Credentials, keys, full card numbers, government IDs | Never | Never in chat; use vaults and proper workflows |
Part 6’s upload ladder and Part 7’s connector ladder sit under this table. If the data class is forbidden, no amount of clever prompting makes the paste OK.
When ChatGPT is the wrong tool
Using ChatGPT for everything is how teams get fluent and reckless at the same time. These are high-signal “stop” cases. They are not anti-AI slogans. They are job-shape mismatches.
1. You need ground truth from systems
Revenue this week, inventory on hand, who owns account X, whether invoice 4412 is paid: those live in databases, warehouses, CRMs, and finance systems. ChatGPT can help you write the SQL, explain a schema, or draft a ticket. It cannot be the source of truth. If you ask “what was Q2 ARR?” without a connected, trusted system (and verification), you will get a confident number with no lineage. Query the system. Then use the model to help interpret.
2. You need licensed professional advice
Law, medicine, tax, regulated financial advice, and many compliance judgments require a licensed human who owns the risk. The model can explain concepts, outline questions to ask a professional, and help you organize documents. It should not be the final medical plan, legal strategy, or tax filing position you “just go with.” If stakes are personal health, immigration, criminal exposure, or material money, buy real advice.
3. You need offline, private-only handling
Some data must not leave a controlled network or device. Air-gapped research, certain government work, some M&A clean rooms, some clinical environments: cloud chat is the wrong shape. Use local tools, approved VDI, or vendor products that match the control package. “I turned off training” does not equal “never transmitted.”
4. Policy bans personal AI (or this vendor)
If the handbook, InfoSec page, or your manager says no personal generative AI for company data, that is the answer. Debating the model quality misses the point. Use the company-approved stack or get a written exception. Shadow tools create unlogged data flows that Security cannot defend in an audit.
5. You need to ship code without review
Codex-style coding help (and any agent that edits repos) can accelerate work. It does not replace tests, code review, threat modeling for sensitive paths, or ownership of the change. “Green tests the model wrote” is not a merge policy. If you would not merge a junior engineer’s PR unreviewed, do not merge an agent’s. Part of the upcoming Codex path on AMS will drill this; the judgment starts here: no blind ship.
6. You need hard multiplayer, real-time collaboration as the primary job
ChatGPT can draft an agenda and a doc. It is a weak substitute for a live negotiation, a design workshop with power dynamics, or a multi-party contract redline where relationship and authority matter. Use shared docs and meetings for the multiplayer layer. Use the model as a prep coach, not as the other party.
The “good enough” checklist
Most days you do not need a full risk committee. You need a short gate you will actually use. If you cannot check these boxes, slow down or switch tools.
- Draft or exploration, not final licensed advice. You are not asking the model to be your lawyer, clinician, or CPA of record.
- You can verify claims against sources you control. Files, dashboards, tickets, or humans who own the fact. If verification is impossible, the output is entertainment or brainstorming only.
- Stakes are reversible or low. A bad internal outline is cheap. A wrong public statement, wire instruction, medical step, or production deploy is not.
- You will edit before send. Your name is on the email, PR, or deck. Read it like you expect a skeptical manager to.
- Policy allows the tool and the data class. Correct workspace, correct connectors, no forbidden PII or secrets.
Optional sixth box for agentic or write-capable sessions: You know how to undo or stop the action (recall message, revert PR, disconnect app, kill the run). If undo is folklore, you are not ready for “Never ask” permissions.
Worked examples: pass vs fail the gate
Example A: rewrite a blog outline (usually pass)
You paste your own outline for a public AMS-style post. No customer data. You will edit every claim. Stakes are low. Policy allows personal or company ChatGPT for marketing drafts. Good enough. Use ChatGPT. Still fact-check anything that looks like a statistic.
Example B: “summarize this customer dispute and draft a refund email” (often fail)
The thread has names, order IDs, and payment hints. Your personal Free account is not the approved channel. The draft may invent a refund policy. Stakes include money and trust. Fail until you use the approved workspace, strip identifiers or follow the ticket tool’s AI features if any, and a human who owns refunds reviews send.
Example C: “what was last month’s churn?” with no data attached (fail)
No ground truth. The model will invent a plausible chart in prose. Fail. Pull the metric from the warehouse or BI tool first. Then ask ChatGPT to help explain drivers using numbers you pasted from that system, and still treat explanations as hypotheses.
Example D: coding agent opens a PR (conditional pass)
Pass only if tests run, a human reviews the diff, secrets stay out of the prompt, and the change is in a branch you can roll back. Fail if the plan is “merge what Codex did because it looked confident.”
Hallucinations, overconfidence, and your reputation
The model is trained to continue text that looks helpful. Helpful-looking text can be wrong about citations, APIs, case law, drug interactions, and your internal process. Overconfidence is a feature of the interface, not proof of truth. Your career risk is not “AI made a mistake.” It is “you forwarded a mistake with your name on it.”
Habits that cut the worst failures:
- Ask for uncertainty: “List what you are guessing vs what is in the file.”
- Force empty answers: “If the source does not contain X, write ‘not in sources.’”
- Separate draft from decision: never let the first generation be the last step on high stakes.
- Prefer primary sources for numbers: systems you query, PDFs you opened, humans who own the metric.
- Watch for “policy fiction”: invented HR rules, refund policies, or security exceptions that sound official.
Common mistakes (series-end edition)
| Mistake | Why it hurts | Replace with |
|---|---|---|
| Treating personal Plus as company approval | Shadow IT, training/retention mismatch | Company workspace or written exception |
| Pasting secrets “just this once” | Credentials live in history and maybe more | Vaults, redaction, never paste keys |
| Skipping verification because tone sounds senior | Confident wrong answers travel far | Good enough checklist every time |
| Using chat as the warehouse | No lineage, no grain, no audit | Query systems; chat for interpretation only |
| Enabling write connectors for convenience | Hard-to-undo actions | Read first; approvals; least privilege |
| Ignoring Memory and shared links | Stale or leaked context | Quarterly hygiene review |
| Shipping agent code unreviewed | Silent bugs and security holes | PR + tests + human owner |
A 15-minute privacy and policy drill
- Open ChatGPT settings. Note plan type (personal vs workspace).
- Find Data Controls. Set model improvement to match your risk comfort on personal accounts. Do not use that as permission for forbidden work data.
- Review Memory. Delete anything that should not stick.
- Review Apps/connectors. Disconnect unused ones. Confirm write permissions are not “Never ask” by accident.
- Find your company AI policy (or ask IT where it lives). Write three allowed and three forbidden examples for your role.
- Print or pin the good enough checklist near your monitor for a week until it is muscle memory.
What this series covered (and what it did not)
Learn ChatGPT from scratch was orientation, not mastery of every surface.
| Part | Focus |
|---|---|
| 1 | What ChatGPT is (and is not) in plain English |
| 2 | Free, Plus, Pro, Business, Enterprise without hype |
| 3 | Account, first minutes, safe early habits |
| 4 | Desktop modes: Chat, Work, Codex at map level |
| 5 | Memory, custom instructions, Projects |
| 6 | Voice, images, files, upload judgment |
| 7 | Connectors, apps, write risk, IT policy |
| 8 (this part) | Privacy, work rules, when not to use it |
We did not turn you into a prompt engineer influencer, an OpenAI admin, or a production ML engineer. We gave you a stable mental model: product vs model vs company, climb complexity only when the job needs it, verify against sources you control, and let policy win when stakes are real.
What to read next on AMS
When you finish this series, keep going in this order unless your job forces a different path:
- ChatGPT product map: Chat vs Work vs Codex, Custom GPTs vs plain chat, models chooser, light API/builder notes. Use this when you keep mixing product names and surfaces.
- ChatGPT everyday tutorial: first useful weeks of writing, planning, learning, structure, email and workplace writing without the agentic blast radius.
- ChatGPT Work tutorial: agentic office work with approvals, multi-step jobs, and when Work beats plain Chat.
- ChatGPT Codex / coding tutorial: repos, safe exploration, review, tests, git-friendly habits (no shipping blind).
- Custom GPTs tutorial: build a simple GPT for a repeating task; instructions, knowledge files, and actions at a light depth.
Related AMS paths still apply: Learn for the wider curriculum, Practical AI and data quality series when you care about pipelines and verification culture, and Claude/Grok product paths if your workplace standardizes on another vendor. The judgment in this post transfers: wrong tool is wrong tool, regardless of logo.
Quick recap
- Separate “who can see this,” “does it train models,” and “does policy allow this data class.”
- Personal plans are not company approval. Use the right workspace.
- Wrong-tool cases: ground truth systems, licensed advice, offline private only, policy bans, shipping code without review, and multiplayer jobs that need humans in the room.
- Good enough: draft not licensed final, verify against sources you control, reversible stakes, edit before send, policy allows tool and data.
- This series ends here. Next: product map, then everyday / Work / Codex / Custom GPTs tutorials.
Sources
Privacy defaults, plan names, and admin screens change. Re-check official pages before you write policy or train a team.
- OpenAI Help: Data Controls FAQ (conversation use for model improvement, consumer controls)
- OpenAI Help: How your data is used to improve model performance (consumer vs business training overview)
- OpenAI Help: Data usage for consumer services FAQ (consumer content handling orientation)
- OpenAI Help: Managing data, sharing, and privacy in ChatGPT Business (Business workspace privacy notes)
- OpenAI: Enterprise privacy (business data handling commitments overview)
- OpenAI: Privacy policy (legal baseline; not a substitute for counsel)
- OpenAI Help: Apps in ChatGPT (apps, permissions, important actions, third-party data sharing notes)
- OpenAI Help: Admin controls for plugins and apps (workspace governance hooks)
- Analytics Made Simple: Learn ChatGPT from scratch (series home)
- Analytics Made Simple: ChatGPT product map (recommended next series)
- Analytics Made Simple: Learn (broader learning paths)
