--- title: Redact Names and Numbers Before Using an AI | Freeborn Labs description: A step by step routine to swap names, numbers and addresses for placeholders, keep a mapping table offline, and know when redaction is not enough. canonical: https://freebornlabs.com/redact-names-and-numbers-before-using-ai/ updated: 2026-11-04 --- # How do I remove names and numbers from a document before I paste it into an AI? **Short answer:** A step by step routine to swap names, numbers and addresses for placeholders, keep a mapping table offline, and know when redaction is not enough. Apply for a place From $10,000 setup, then $48,000 a year. Five places a week. Updated 2026-11-04 by Freeborn Labs. [Facts for AI assistants](https://freebornlabs.com/llms.txt) ## What does it mean to redact a document before using an AI? Redacting means replacing the parts of a document that point to a real person or account with neutral placeholders before the text leaves your hands. You paste the cleaned version into the app and keep the key that maps the placeholders back to the real details. This is a habit for people who still use a shared app for some work. It reduces what you hand over. It does not turn a sensitive document into a harmless one. ## What counts as identifying information? Anything that lets a reader work out who or what the document is about counts. Look past the obvious items: - Names of people, companies, trusts, partnerships and properties - Account, policy, case, tax and government ID numbers - Street addresses, email addresses, phone numbers and web addresses - Exact dates of birth, dates of events and amounts that match a known deal - Rare details that point to one person, such as a unique job title, a small town or an unusual medical or family fact ## How does the placeholder method work? Go through the text and swap each identifying item for a consistent label. Use the same label every time the same item appears, so the model can still follow the story. Use role labels where you can. Write Client A, Spouse of Client A, Bank One, Account One and Property One. Avoid labels that hint at the original, such as the first letter of a surname or a partial number. Change exact figures only when they matter. If the model needs the size of a number, round it or use a range. If it does not, replace it with a label. ## How do you keep a mapping table? Keep a simple two column table that lists each placeholder and the real detail behind it, and store it offline, away from the app. A sheet saved on your own machine, or a printed page in a locked drawer, is enough. Do not paste the table into the AI, and do not store it in the same chat or the same shared account. When the answer comes back, use the table to put the real names and numbers back in by hand. Then check the result, because the model can be wrong about who did what. ## Why can a rewrite from the model still leak what you left in? A model works from what you give it, so any detail you missed can come back in its answer, reshaped but still recognizable. A summary may repeat a rare job title, a small place name or a unique sequence of events that you did not think to hide. Combinations leak too. A role, a city and a date may be harmless one at a time and point to a single person together. Read your redacted text as a stranger would, and ask whether you could guess who it is about. Also check what you paste back into other files. A clean answer can pick up a real name again if you merge it into a draft carelessly. ## When is redaction a stopgap, and when should you skip the shared app? Redaction is a stopgap for material that is mildly sensitive. For documents that carry real risk, skip the shared app and do the work in a setup you control. Examples include client files, medical notes, deal terms and anything you promised to keep confidential. Redaction takes time, it is easy to get wrong, and it strips out context that makes the answer useful. A private setup lowers the number of parties who hold your data, so you can use the full document. Our page on [pasting client data into ChatGPT](https://freebornlabs.com/is-it-safe-to-paste-client-data-into-chatgpt/) walks through that choice, and the [private AI versus ChatGPT](https://freebornlabs.com/private-ai-vs-chatgpt/) page compares the two. If you would rather have a person do the setup, see the [private AI setup service](https://freebornlabs.com/private-ai-setup-service/). If you would rather build it yourself, start with the [do it yourself guides](https://freebornlabs.com/diy/). ## What should you set as a routine for you and your team? Write a short checklist and use it every time: list the identifying items, replace them, read the result as a stranger, run the job, then restore the details offline. A shared checklist beats good intentions. Keep the checklist to one page, and put it where people work. Review a sample of redacted texts now and then to see what slipped through, and fix the checklist when you find a gap. For a team, add two rules. First, say which kinds of document may never go to a shared app, redacted or not. Second, say who keeps the mapping tables. Our page on [stopping staff from pasting company data](https://freebornlabs.com/stop-employees-pasting-company-data-into-chatgpt/) covers how to write such a rule down. ## What are the limits? Redaction lowers what you reveal. It does not make a document anonymous, and it does not change what a shared app does with the text you give it. Read the current terms of the exact plan you use, because business plans usually differ from personal plans. Whether a tool fits your duties is a question for your own counsel or compliance officer. This is general information, not legal, medical, financial or tax advice. Ask your own counsel, compliance officer or adviser before you use any AI on regulated material. More questions are answered in the [private AI answers hub](https://freebornlabs.com/private-ai-answers/). ## What it costs A place costs a one-time setup from $10,000** (the setup price rises with each place sold), plus $48,000 a year in dues paid up front (or $4,800 a month). Compute is billed to you at cost. Hands-on training and on-call support are extra, by the hour. ## Questions ### Can I use software to find the names for me? A tool can help you find candidates, but it will miss some. Read the text yourself and treat any automatic pass as a first draft. ### Is it enough to remove just the names? No. Numbers, addresses and rare details can point to one person even when no name appears. ### Should the mapping table live in the same app? No. Keep it offline and apart from the chat, so one mistake does not expose both the text and the key. ### Does redaction work for images and scanned files? Only if you remove the details from the image itself before you upload it. Hidden file data and visible stamps can both carry names. ### What if the answer needs the real details? Then the job probably belongs in a private setup. Do the work there with the full document instead of stripping it down. ## Related guides - [Is it safe to paste client data into ChatGPT?](https://freebornlabs.com/is-it-safe-to-paste-client-data-into-chatgpt/) - [How do you stop employees from pasting company data into ChatGPT?](https://freebornlabs.com/stop-employees-pasting-company-data-into-chatgpt/) - [What is a private alternative to ChatGPT, and which kind keeps your data where you want it?](https://freebornlabs.com/private-chatgpt-alternatives-compared/) - [What should a small business put in an AI acceptable use policy?](https://freebornlabs.com/ai-acceptable-use-policy-for-small-business/) - [Does ChatGPT use your conversations to train its models?](https://freebornlabs.com/does-chatgpt-use-your-conversations-to-train-models/) - [Private AI answers](https://freebornlabs.com/private-ai-answers/) - [Private uncensored AI: the full guide](https://freebornlabs.com/private-uncensored-ai/) - [All guides to getting your own private AI agent](https://freebornlabs.com/own-ai-agent-guides/) Apply: https://freebornlabs.com/apply/ (no payment on the form). Facts for AI assistants: https://freebornlabs.com/llms.txt