Why it matters
When you paste data into an AI tool, it leaves your control. Depending on the service, your plan and your settings, what you send may be stored, read by staff reviewing for abuse, or used to improve future models. Business and enterprise plans usually promise more than free ones. Either way, check your provider's data controls. And even with the strictest settings, you've copied client data to another company.
For a lot of questions the AI doesn't need the private parts. “Which donors gave more than last year?” works just as well if the donors are called [Name 14] and [Name 15]. The trick is to swap out the private details consistently, so the AI can still tell people apart, and to be able to swap them back afterwards.
What counts as private
The obvious columns: names, email addresses, phone numbers, street addresses, dates of birth, Social Security numbers, account and policy numbers. The less obvious ones:
- Notes and comment columns. “Son Mark handles calls: 801-555-0166” has a name and a phone number in the middle of a sentence.
- Small groups. City + age + gift amount can point to one person in a small town, even with no name.
- IDs from other systems. A donor ID or client number links back to your database.
You don't have to remove everything. You have to remove what the AI doesn't need to answer your question.
Your options
1. Delete the columns by hand
Open the file, delete the name, email and phone columns, save a copy. It's free and fine for a small file. It has two problems: it misses details hiding in notes columns, and once the names are gone, the AI's answer (“the third donor on the list…”) is hard to tie back to real people.
2. Run an AI model on your own computer
Tools like LM Studio and Ollama run a model locally, so nothing leaves your machine. It works, but you need a capable computer and some setup, and local models are still behind the best online ones for analysis. Worth it if you do this all day; heavy for the occasional spreadsheet question.
3. Replace private details with placeholders
Swap every private value for a consistent label, ask the AI, then swap the labels back. This is what the Privacy Filter does, in your browser, without uploading the file anywhere.
Step by step with the Privacy Filter
- Open your file. Go to the Privacy Filter and choose “Open a spreadsheet”. CSV and Excel both work. It's read on your computer; nothing is uploaded.
- Check each column. The tool guesses what each column holds from its name and its values. Columns that look private are set to “Replace with placeholders”, and notes columns to “Scan text for private details”. Change anything it got wrong. The preview shows the first rows as they'll come out.
- Clean and download. You get a copy where Kenji Nakamura is [Name 14] on every row, plus, with Pro, a key file that remembers who [Name 14] is. Keep the key file somewhere private; it undoes the cleaning.
- Ask the AI. Upload the cleaned copy to ChatGPT or Claude, and tell it what the placeholders are (the prompt below helps).
- Put the real values back. Copy the AI's answer, choose “Put real values back” in the Privacy Filter, load your key file, and [Name 14] turns back into Kenji Nakamura, still on your computer.
A prompt that works
Words in [brackets] replace private details. Each one stands for one real value: [Name 14] is always the same person, [Phone 3] always the same number. Keep the placeholders exactly as written in your answer, so I can swap the real values back in.
The last sentence matters. If the AI rewrites [Name 14] as “Name 14” or “the 14th donor”, it can't be restored automatically.
If you clean files every month
By default the numbering starts fresh in each file, so Kenji might be [Name 14] in January and [Name 37] in February. If you compare files over time, turn on “Use the same placeholders in every file” and choose a passphrase. The same person then gets the same placeholder in every file, on any computer, as long as you use the same passphrase. The passphrase isn't saved anywhere, so write it down somewhere safe.
What no tool can do for you
Pattern matching finds what private details look like: emails, phone numbers, addresses, card numbers. It can't know that a nickname in a notes column is a person, or that “her chemo appointment” is health information. Read the cleaned file before you send it. If your data falls under HIPAA, GDPR or a client contract, a tool like this reduces what you expose; it doesn't make a file compliant on its own. Check with whoever owns those rules.