What happens to what you paste depends on the AI vendor, your plan and one or two settings.
Use a business plan or turn model training off, never paste passwords, bank details or ID numbers, and remove client names and addresses when the task does not need them. The tool below does the removing for you, in your browser.
Step by step
1Check each person's training setting
If people in the firm use personal AI accounts for work, check each person's training setting today. It takes a minute per person.
2Use a business plan for daily admin
If AI is part of the firm's daily admin, a business plan is the cleaner option: training is off by default and an administrator controls the settings. Plans and their terms change; read the current terms.
3Never paste these, on any plan
Passwords, login codes and API keys. Bank account and card numbers, including a contractor's details on an invoice. Government ID numbers: Social Security numbers, passport numbers, tax IDs of individuals. Anything a client asked you to keep confidential, or that your contract says stays within the project team.
4Remove client details the task does not need
Client and family names, site and home addresses, email addresses and phone numbers, and exact amounts when the task is about wording, not numbers. A reminder email reads the same with [CLIENT] and [AMOUNT] in it; you put the real values back before you send.
5Check your client contracts
Some include confidentiality clauses that limit where project information can be stored or who can see it. Read yours before you put project material into any third-party tool.
6Ask a lawyer about data protection law
Rules on personal data differ between US states and between countries in Latin America. If your firm works across borders, ask a lawyer which rules apply to your client records.
Free tool
Redact, then restore
Paste text below. The tool replaces email addresses, phone numbers, amounts, long ID-like numbers, web links and any names you list with placeholders such as [EMAIL 1] and [NAME 2]. Copy the cleaned text into your AI chat. When the AI answers, paste its reply into the second box and the tool puts the real details back.
2. Cleaned text to paste into the AI
4. Reply with the real details back in
The tool runs entirely in your browser. It has no network code; nothing you paste leaves this page, and nothing is saved when you close it. It catches common patterns, not everything: read the cleaned text before you paste it anywhere.
An architecture firm holds more private information than it tends to notice: clients' home addresses and floor plans, budgets, family circumstances behind a brief, contractor bank details, staff pay. Admin work with AI touches much of it. This page covers the basics a small firm needs: what the three main chat tools say they do with your text, what never to paste, and a habit that keeps client details out of AI chats when they are not needed. It is not legal advice. Data protection law differs by country and state.
What the main AI chat tools say they do with your chats
Summarized from each vendor's own help or policy pages, checked November 5, 2026. These pages change; follow the links in the sources before you decide.
Tool
Personal plans
Business plans
ChatGPT (OpenAI)
Conversations can be used to train models unless you turn off "Improve the model for everyone" in Settings, Data Controls. With it off, new chats stay in your history but are not used for training.
OpenAI says it does not use data from ChatGPT Business, Enterprise or Edu workspaces to train its models by default.
Claude (Anthropic)
On Free, Pro and Max, each user chooses in Privacy Settings whether chats are used for training. If you allow it, Anthropic keeps the data for up to five years; if not, the 30-day retention period applies.
The training change does not apply to services under Anthropic's Commercial Terms: Claude for Work (Team and Enterprise), Government, Education and the API.
Gemini (Google)
With Keep Activity on, Google uses activity to improve its services, including training generative AI models, and human reviewers read a subset of chats; reviewed chats are kept for up to three years. With it off, future chats are not used for training unless you send feedback, and are kept for 72 hours. Temporary chats are not used for training.
With a qualifying Google Workspace edition, Google says chats and uploaded files are not reviewed by humans or used to train models without permission.
Google's own privacy notice adds a line worth repeating to the whole firm: "Please don't enter confidential information that you wouldn't want a reviewer to see."
Settings decide training. They do not make a chat tool a safe place for passwords or bank details. Some vendors keep chats for a period for safety review whatever the setting.
Automations
Automations raise the same questions. Before connecting an AI step to your mailbox or drive, check which account it runs under and what that vendor's terms say. The readiness check includes this as question 10.
A one-page AI rule for the firm
Small firms rarely need a long policy. Five lines, agreed and shared, cover most of it. Copy and adapt.
When: Agree it once with everyone who uses AI for firm work, and review it when your plan or tools change.
Subject: How we use AI tools for admin at [FIRM NAME]
1. Work chats use [the firm's business plan / personal accounts with model training turned off].
2. Never paste passwords, bank or card details, or ID numbers into any AI tool.
3. Remove client names, addresses and contact details when the task does not need them.
4. Read every AI draft before it goes to a client. Check every amount, date and name against the source.
5. If a client contract limits where project information can go, that contract wins. Ask [OWNER] if unsure.
Reviewed by [NAME] on [DATE].
Frequently asked questions
Does ChatGPT train on what I paste?
On personal plans it can, unless you turn off "Improve the model for everyone" in Settings, Data Controls. OpenAI says it does not use data from ChatGPT Business, Enterprise or Edu workspaces to train its models by default.
Does Claude train on my chats?
On consumer plans (Free, Pro, Max), Anthropic asks each user to choose in Privacy Settings. Anthropic says it does not train on data from its commercial products, such as the Team and Enterprise plans and the API, unless the customer opts in.
Does Gemini train on my chats?
With a personal Google account and Keep Activity on, Google says it may use chats to improve its services, including training models, with help from human reviewers. With a qualifying Google Workspace edition, Google says chats and files are not reviewed by humans or used to train models without permission.
What should a firm never paste into an AI chat?
Passwords and login codes, bank and card details, government ID numbers, anything a client asked you to keep confidential, and anything your contract says stays within the project team. Remove names and addresses when the task does not need them.
Does the redaction tool on this page send my text anywhere?
No. It runs in your browser with no network code. Nothing you paste into it leaves the page.
Archon & Partners