What not to paste into ChatGPT or Claude
AI saves agents hours on emails, listing copy and contract summaries, but your clients' data is your responsibility. Here's what to keep out, the two settings to change today, and a prompt that redacts names and numbers for you.
Published Oct 19, 2026 · Settings from OpenAI and Anthropic's own help pages, checked Oct 19, 2026
Never paste these
- Social Security, driver's license or passport numbers, and scans of IDs.
- Bank, loan, escrow and wire details: account and routing numbers, wire instructions, loan numbers. Wire fraud targets real estate deals; keep this data out of every tool that doesn't need it.
- Credit reports, pre-approval letters, pay stubs and tax returns from buyers.
- Full birth dates and health or family details a client shared in confidence (divorce, probate, disability).
- Login details for your MLS, CRM, transaction platform or email, ever.
- Seller motivation or a client's bottom line (lowest price, deadline pressure) tied to their name. Keep negotiating details anonymous.
Usually fine with placeholders
- Property facts for listing copy: beds, baths, upgrades, neighborhood features (the address is public once listed, but you don't need it in the prompt).
- Client emails you want help replying to, after you swap names and contact info for [CLIENT 1] and [CONTACT].
- Contract or disclosure language you want explained in plain English, with names, addresses and numbers removed. AI can misread contracts, so it's for understanding, not legal advice; ask your broker or an attorney before you rely on it.
ChatGPT vs Claude privacy settings
| ChatGPT | Claude | |
|---|---|---|
| Turn off training | Settings > Data controls > "Improve the model for everyone" off (applies across devices when signed in) | Settings > Privacy > "Help improve our AI models" off |
| Private one-off chat | Temporary chat: not in history, not used for training, kept up to 30 days for safety | Incognito chat: not used for training, even if the training setting is on |
| Retention | Saved chats stay until you delete them | 30 days if training is off; up to 5 years (de-identified) if on; deleted chats removed from back-end within 30 days |
| Feedback buttons | Thumbs up/down can send the whole chat for training, even if opted out | Thumbs up/down stores the whole chat for up to 5 years |
| Team plans | Business and Enterprise workspaces not used for training by default | Claude for Work (Team, Enterprise) and API under commercial terms, not the consumer training setting |
| Our score | 98/100 (rank #1) | 94/100 (rank #2) |
Sources: OpenAI: Data controls in ChatGPT ↗, Anthropic: model improvement settings ↗, Anthropic: how long data is stored ↗ and Anthropic: is my data used for training ↗, checked Oct 19, 2026. Vendors change these settings; confirm in your own account.
Five-minute privacy checklist
- Read your brokerage's AI or technology policy. Some brokerages approve specific tools or require a team plan. Follow it first.
- Turn off training in both ChatGPT and Claude (signed in, the setting follows your account).
- Use temporary (ChatGPT) or incognito (Claude) chats for anything with client details, even redacted.
- Redact before you paste with the prompt below, or by hand.
- Delete old chats you no longer need, and skip the thumbs up/down on client conversations.
Copy-paste redaction prompt
Paste this into a new temporary or incognito chat first, then paste your email or notes. It's a helper, not a guarantee, so scan the redacted version yourself before you continue:
General information only, not legal advice. California agents also have duties under state privacy and real estate law; ask your broker or an attorney about your specific situation.
FAQ
Is it safe to paste client information into ChatGPT or Claude?
Treat any consumer AI chat like an email to an outside vendor. Paste what you need for the task (the property facts, a redacted email, your own notes) and leave out anything that identifies a client or could be used for fraud: Social Security numbers, bank, loan and wire details, full birth dates, ID scans, and pre-approval or credit documents. Check your brokerage policy first; it may limit which tools you can use.
Does ChatGPT train on my chats?
On personal plans (Free, Go, Plus, Pro) it can, unless you turn off "Improve the model for everyone" in Settings > Data controls. Temporary chats are not used to train models and may be kept up to 30 days for safety. OpenAI says it does not use content from ChatGPT Business or Enterprise workspaces to train models by default.
Does Claude train on my chats?
On Claude Free, Pro and Max you choose. If "Help improve our AI models" is on, Anthropic may keep that data in de-identified form for up to 5 years; if it is off, chats stay on the standard 30-day retention. Change it in Settings > Privacy. Incognito chats are not used for training. Claude for Work (Team and Enterprise) and the API are under separate commercial terms.
Does turning off training delete my old chats?
No. Turning off training only affects future training. Delete chats you no longer need. Anthropic says a deleted Claude chat is removed from its back-end within 30 days.
Is a thumbs up or down on an answer private?
Not fully. Both OpenAI and Anthropic say that when you send feedback, the whole conversation can be used to improve their models, even if you opted out of training. Skip the feedback buttons on chats that contain client details.
See ranked tools for contracts & docs → · Personal productivity tools → · More guides →