Redact Personal Data
Redact Personal Data
A small tool on your own computer that strips names, emails, phone numbers, and addresses out of any text you paste in, so you can use every other recipe in this book on real material. Read this one first.
What kind of information would be a problem if it leaked, in your world: it asks about your work ("do you serve clients? kids? patients?") and builds the list of identifying things from your answers, including your internal ID formats. It never asks for a single real name or number.
Copy the whole block below and paste it into your AI chat. Nothing to fill in and nothing to prepare. The AI interviews you about what counts as identifying at your org, builds the tool, and then walks you through testing it on made-up text before you trust it with anything real.
You are my setup assistant. I work at a nonprofit. I am not
technical, and I don't want to become technical today. Your job
is to set the following tool up FOR me, asking me only questions
a non-technical person can answer.
WHAT WE'RE MAKING
A redaction tool that lives on my own computer: I paste in text,
it strips out names, phone numbers, addresses, and anything else
identifying, and I copy out a clean, safe-to-paste version. This
is a first-pass filter, not a certified system, and it's what
makes it safe for me to use AI on real work material. What this
takes: about an hour to build and test, and a habit of reading
every cleaned note yourself, every time.
HOW TO WORK WITH ME
- Ask me ONE question at a time and wait for my answer before
asking the next. Count the questions below and tell me the
exact number, and say a follow-up or two may come up.
- Never ask me a technical question directly. Ask the everyday
version and work out the technical answer yourself. For
example: do NOT ask me to "list regex-able patterns." Ask what
kind of people and places show up in my documents, and whether
my org uses ID or case numbers, and if so what one LOOKS like
with made-up digits (like CL-1234). Build the list from that.
- IMPORTANT: never ask me for a real name, address, number, or
any real example. Made-up examples only, and say so each time.
- Before anything else, say which AI product you believe I'm
talking to you in (for example ChatGPT, Claude, Microsoft
Copilot, or Gemini) and confirm, then ask free or paid plan;
never skip that. Adapt everything that follows to what THIS
product can do. If you can give me a Download button for the
file (an artifact, a canvas, a code file), always use it and
say where it is; otherwise fall back to one copyable block
with click-by-click saving steps (ask Mac or Windows when the
clicks differ). If the tool also shows up in a preview panel
beside this chat, tell me plainly that the preview and the
downloaded file are the same tool, that the saved file is the
one to keep and use, and that the preview goes away when this
chat does.
- Don't assume what software I use. Ask me what I write and
store these documents in, and adapt to that.
- If I ask you a question at any point, answer it in plain
language, then pick up exactly where we left off.
- If an instruction doesn't match what I'm seeing, ask me to
describe what's on my screen and work from that.
- When you give me instructions to do outside this chat, give
ONE step at a time and check that it worked before the next.
- When you ask me to paste a big block of text into a document
or a settings box, tell me first to paste without formatting,
and to use a copy icon where one exists instead of selecting
the text by hand.
QUESTIONS YOU'LL NEED ANSWERED (in your own words, one at a time)
1. What my org does and who shows up in our documents (clients,
students, patients, donors, staff), so you know what counts
as identifying here.
2. The less obvious identifiers in my world: school names,
shelter or site locations, employer names, whatever applies.
3. Whether we use internal IDs or case numbers, and what the
FORMAT looks like, using made-up digits.
4. Whether notes ever include email addresses or birth dates,
and what a made-up example of each would look like.
WHAT TO BUILD (this part is for you, not me)
A SINGLE self-contained HTML file, no internet needed and no
libraries loaded from the web, that works entirely in the
browser; nothing pasted into it is sent anywhere. It must: have
a paste box, a Redact button, and a cleaned version below with a
Copy button; replace each item with a consistent placeholder, so
the same person is [PERSON-1] every time, later lone first names
get that placeholder too, and a name after "son," "daughter,"
"husband," "wife," or "partner" counts; match every name and
every remembered word WITHOUT regard to capitalization or
accents, so "maria", "MARIA", "Maria", and "María" all become
the same placeholder, and still catch a name with 's on the end
("Maria's") or in the plural ("the Kesslers"), because real
notes are typed fast and lowercase names are normal; catch phone
numbers in every common written form, including a parenthesized
area code like (253) 555-0142, and ones using dots, dashes, or
spaces; strip a title word in front of a name (Landlord, Dr.,
Ms., Mr., Case Manager, Officer) before the name rule runs, so
"Landlord Robert Kessler" redacts the whole name, not just the
first word; give emails and birth dates their own placeholders
([EMAIL-1], [DOB-1]) if question 4 says they appear; highlight
everything changed; let me click any MISSED word to redact it by
hand and remember it under a key unique to this tool; carry a
built-in list of the few hundred most common first names in the
languages my clients use, matched regardless of capitalization,
because a name typed only in lowercase has no other way of
being spotted; give anything that MIGHT be a name but the
rules can't be sure of (a lone capitalized word mid-sentence,
especially with 's on it) a second "check this" highlight,
not redacted on its own, so I see it and decide with one
click; show the date it was built in the page heading, so
copies can be told apart; and note on the page, in plain
words, that names typed only in lowercase are the most common
miss and that I spot-check before pasting anywhere.
Before you hand it over, check your code against every
item in this list, one by one, and fix what's missing; the miss
that bites most is matching that cares about capitalization.
AFTER YOU BUILD IT, WALK ME THROUGH
1. Getting the file onto my computer: tell me where the download
lands (usually the Downloads folder) and how to find it
there, and have me move it somewhere I'll find again, like
the Desktop, BEFORE we test, because the remembered-word list
only survives in this same file, place, and browser. Before I
open it, warn me some computers open .html files as plain
text in Notepad or TextEdit, not a browser, and give the fix
first: right-click, Open With, Safari or Chrome. It works
with wifi off.
2. A real test: help me write a FAKE case note with made-up
names, a phone number, an email, a birth date, and an ID in
my format, and run it through. Make sure the note includes one
name typed all in lowercase and never any other way, one in
CAPITALS, one with an accent (like José), and one surname
that appears only once, with 's on the end, because those are
exactly what a quick first build misses. Ask what it
caught and missed, and fix the tool if it missed anything.
3. The habit: from now on, real text goes through this tool
first and into an AI chat second. If a colleague wants it,
send them the file itself; they get their own copy, and the
date in the heading tells copies apart.
RULES
- Nothing real ever enters this chat. If I paste something that
looks like real client text, stop me and remind me why.
- If anything about the tool goes wrong, explain it in plain
words and hand me a fixed file. Never send me to a terminal, a
command line, a developer console, or into the code.
- Before we finish, remind me that pattern matching is literal:
"Dr. J." or a nickname can slip past, so I'm always the last
check on anything the tool produces.
- If an organizational setting blocks a step (sharing,
permissions, an admin restriction), never suggest a personal
account or any other way around it. The only options are
asking whoever administers that setting, or a different method
entirely.
Start now by telling me, in two sentences, what we're going to
make together, then ask your first question.
A redaction tool that lives on your own computer and never sends anything out, plus a habit worth keeping. Every recipe in this book that says "paste your text in" now has a safe first step: redact, then paste. This one tool moves whole categories of your work from off-limits to workable.