AI privacy: what not to paste

The dangerous paste rarely looks dramatic. It is a “quick context dump” that includes a client’s personal email, a one-time password you meant to delete, or a sentence about someone’s health because it explained why a deadline slipped. The model answers helpfully. The text has already left your machine. This guide is a practical checklist for people who keep notes locally and still use AI — not legal advice, not a compliance certificate, and not a promise that any vendor is “safe enough.”

Three builds ship from this site. SEO Note 2.1 is the current one: an Electron app (about 121 MB installer) for Windows 10/11 64-bit and Linux x64, with Markdown, the tree editor, AI talking to a model you connect — a local one or any OpenAI-compatible endpoint, a web clipper and git sync of the file. For Android 6.0 and newer there is a release-signed APK (~10 MB) of the same interface; on the phone notes live as a folder of Markdown files on the device, and passwords, sync, the clipper and AI answer there with an explicit refusal. The classic build 1.3.34 (~2.5 MB, Win32) stays on the download page for Windows 7, 8 and 32-bit PCs, where 2.1 does not install; its notes file is .seo. All three work offline with no account: pick your build, or read how 2.1 and 1.3.34 differ.

SEO Note is a free Windows tree outliner: one local file, no mandatory cloud account. Its AI calls a model you connect yourself — Ollama or LM Studio on this PC, OpenAI, DeepSeek, OpenRouter or any OpenAI-compatible address — and can be switched off entirely. Privacy here means you choose the slice: the model sees only what you give it — the current note, the tab tree, @-mentioned notes and attached files, or the few notes its local search picked for your question. With a local model nothing leaves the machine. That is the workflow this site keeps repeating — and the reason a do-not-paste list belongs next to your prompt library.

Where the work happens

Stays in SEO Note

• Full vault and archives
• Secrets and credentials
• Health and sensitive PII
• Raw dumps you have not redacted

May go to AI (curated)

• Redacted project facts
• Decisions without secrets
• Questions you own
• Prompt templates without keys

A privacy-minded tree

Put sensitive material under clear parents so you do not “accidentally” select too much. Labels beat cleverness.

Work.sn
  Projects
    Client A — public facts only
  Secrets (NEVER paste)
    Credentials
    Tokens / keys
  People notes (careful)
    — redact before any AI use
  Health (local only)
  AI-safe briefs
    Client A — 2026-08 brief
  AI scratch
    dated, still review before re-paste

Keep an AI-safe briefs branch: copies you already scrubbed. Paste from there, not from the live Secrets or Health nodes.

Checklist: do not paste

• Passwords, API keys, tokens, recovery codes — even “temporarily,” even into a “private” chat
• Government IDs, tax IDs, full bank details
• Other people’s PII you do not have a clear reason to send (phones, home addresses, passport data)
• Client secrets under NDA: unpublished strategy, credentials they shared, internal URLs with tokens
• Health details about you or anyone else — symptoms, diagnoses, prescriptions
• Children’s data, school records, precise locations of minors
• Full vault / whole PARA tree / years of daily notes “just in case”
• Screenshots that hide secrets badly (taskbar passwords, notification previews)

What a careful paste looks like

One project node. Facts and decisions rewritten without secrets. Placeholders like [CLIENT] or [BUDGET] when the real value is not required for the task. A prompt that says “do not invent missing numbers.”

Reusable redaction prompt (for you, before send)

Before I paste into any AI chat, scrub this note:
- Replace names with roles ([CLIENT], [MANAGER])
- Remove emails, phones, addresses, IDs
- Remove passwords, keys, tokens, URLs with secrets
- Remove health details
- Keep only facts needed for: [TASK]

Output the scrubbed version only.

You can run that scrub in a local editor or — carefully — in a tool you already trust, then paste the scrubbed text into the model that does the real job. Prefer scrubbing yourself when the note is sensitive. Store the scrubbed brief under AI-safe briefs.

Default rule: if you would hesitate to put the paragraph on a shared office whiteboard, do not paste it into a chat. SEO Note keeps the whiteboard offline; you still decide what leaves. This is operational hygiene, not legal advice — check your contracts, employer policy, and local rules when work data is involved.

How-to: one sitting cleanup

1. Create Secrets (NEVER paste), Health (local only), and AI-safe briefs parents.
2. Move credentials and tokens under Secrets. Search your file for “password”, “token”, “Bearer”.
3. Pick one project you habitually dump into AI. Build a scrubbed brief under AI-safe briefs.
4. Save the weekly or project prompt next to it (see prompt templates).
5. For the next AI session, paste only from AI-safe briefs. Close the chat when done; write accepted outcomes back without reintroducing secrets.
6. Copy the notes file to a backup disk you control — privacy includes not losing the only copy.

	Risky habit	Safer habit
Scope	Whole notebook	One scrubbed brief
Secrets	Inline in project notes	Secrets branch, never selected
People	Full names + contacts	Roles / initials if needed
After chat	Trust the thread	Update local truth node

Workplace and client reality

Freelancers and staff often keep the same Windows file for personal and client work. That is fine until a paste crosses the wrong boundary. A practical split: client projects under Projects with an AI-safe brief each; personal Health and family notes under parents you never select when briefing work models. If your contract or employer bans third-party AI for work data, obey that — a local outliner does not override policy. Use the tree to draft offline and keep the chat for material that is already cleared.

Redaction examples that catch people: calendar titles with full names; email footers with phones; support tickets pasted “for tone”; bank SMS codes sitting in Inbox; meeting notes that quote someone else’s medical leave. Replace with roles and outcomes: “[CLIENT] wants launch before Easter; blocker is copy approval,” not the thread that named every stakeholder’s private number.

Vendor toggles (“do not train on my data”) may help your risk posture. They do not turn a chat into a local vault and they are not a substitute for reading the current terms. When unsure, paste less and keep the authority in SEO Note — same spine as source of truth.

Pitfalls

• Assuming “I turned off training” equals “data never leaves processing logs”
• Pasting screenshots of dashboards that include customer rows
• Re-pasting an old AI scratch node that still contains a secret you later moved
• Using AI to “remember my vault” — that fights the local source-of-truth model in source of truth + AI
• SEO Note does not encrypt the file by itself like a password manager; OS permissions, disk encryption, and backups are still your job
• This article does not replace counsel, DPA review, or HIPAA / GDPR assessments

For cost pressure that also pushes you toward smaller pastes, see token cost workflow. Daily journals need the same redaction before weekly review: daily notes → AI review.

Next steps

1. Add Secrets and AI-safe briefs branches today
2. Scrub one real brief you were about to paste
3. Delete or relocate any password lines sitting under Projects
4. Keep the do-not-paste list as a node you reopen monthly
5. Download SEO Note free (no account) for a local vault

Keep going

• Tree templates and PDF guide — role and method skeletons you can copy straight into your own tree.
• If this guide helped, you can say hello or grab ready-made trees on the templates page.

Adjacent tools (VPS for local models, courses, VPN) are not stuffed into every paragraph. If an affiliate link appears, it will be labeled and is never required to use SEO Note.

Mini FAQ

What should I never paste from my notes into an AI chat? Passwords, API keys, tokens and recovery codes, government or tax IDs, full bank details, other people's personal data, health details, children's data, and your whole vault or years of daily notes.

Does SEO Note encrypt my notes file like a password manager? It does not encrypt the file by itself like a password manager. Operating system permissions, disk encryption and your own backups remain your job.

If I turn off model training, is my pasted data private? Toggles such as 'do not train on my data' may help your risk posture, but they do not turn a chat into a local vault and are no substitute for reading the current terms.

How do I share project context without leaking client secrets? Keep an AI-safe briefs branch of copies you already scrubbed and paste from there, never from the live Secrets or Health nodes. Replace names with roles such as [CLIENT] and remove emails, phones, IDs and keys.

What if my client contract bans third-party AI for work data? A local outliner does not override that policy, so obey it. Use the tree to draft offline and keep the chat for material that is already cleared.

Does this guide replace legal counsel on 152-FZ and GDPR? No. This is working hygiene, not legal advice: for production data check your contracts, employer policies, and local regulations. A DPA review or a 152-FZ/GDPR assessment is separate professional work. If your contract bans third-party AI for work data, follow it — a local outliner does not override that policy.

Related: Source of truth + AI · Token cost · Daily notes review · PARA tree · Prompt templates · Offline notes · FAQ