Four ways to replace
Numbered tokens, realistic fake values, solid blocks, or delete. Tokens and fake values are reversible; blocks and delete are one-way on purpose.
Clean your text before the AI sees it
Paste anything you were about to send to ChatGPT, Claude or Gemini. Names, emails, phone numbers, SSNs, card numbers, API keys and addresses become placeholders — then the model’s answer gets the real values back. Nothing is uploaded, because there is nowhere to upload it to.
- 30 detectors
- reversible round trip
- no account, no upload
- works offline
Your text
Safe to paste
The cleaned version appears here as you type — highlighted so you can see exactly what
changed.
no personal data yet Everything below happens in this browser tab. Nothing is uploaded, and nothing you type is
saved.
Mapping — 0 values (only visible in this tab)
| Placeholder | Real value | Uses |
|---|
Check a few rows before you paste — this is the fastest way to spot a detector that grabbed the wrong thing. Nothing here is stored or sent anywhere.
What to look for — 30 detectors, custom terms, exceptions
People & organisations
Government & health IDs
Financial
Technical & secrets
Location
For the things no word list can know: internal codenames, unusual surnames, a product
that isn’t public yet. Custom terms are never saved.
Values on this list stay in the text and out of the mapping — useful when the model needs
your own company name to answer usefully.
Step 3 — paste the AI’s answer here to put the real values back
The model replies using the placeholders. Paste its answer below and every
[NAME_1] becomes the real name again — using the mapping held in this tab, which
is the only copy that exists.
How it works
- Paste the email, ticket, note, log or spreadsheet rows you were about to send to an AI.
- Every name, ID, key and address the detectors recognise is highlighted as you type.
- Copy the placeholder version and paste that into ChatGPT, Claude, Gemini or anything else.
- Paste the model’s answer back into step 3 — the real values return, only in your tab.
What makes this different
The round trip is the point
Every other redaction tool hands you a cleaned string and stops. Here the placeholder mapping stays in your tab, so you can paste the model’s answer back and get the real names, numbers and addresses returned in place — including when the model bolds, lowercases or reformats the placeholders. That only works locally: a hosted service would have to store your mapping on its servers.
Detection that validates, not guesses
Card numbers must pass the Luhn check and carry a real issuer prefix. IBANs are checked with mod-97, US routing numbers with the ABA weighting, Aadhaar numbers with the Verhoeff digit, NHS numbers with mod-11. SSNs respect the ranges the SSA never issues. The result: your order IDs and reference numbers survive the scrub.
Your words, your rules
Custom terms catch what no word list can know — codenames, unusual surnames, internal systems. A never-redact list keeps the values the model needs.
Nothing is remembered
No text, no custom terms and no mappings are ever stored. The site remembers only which detectors you switched on, in this browser.
Honest about the misses
Pattern matching cannot catch a name it has never seen, an ID with no label, or the sentence that identifies someone by circumstance — “the only left-handed pitcher on the roster”. Every page here says so, and the highlight view exists so you can check the machine’s work in a couple of seconds rather than trusting it.
Why this step exists at all
The reason to scrub is not that AI companies are careless — it is that a chat window is a one-way door. On consumer plans, conversations are used for training unless you opt out, they are retained for months or years, and litigation can freeze that data in place: in the New York Times case, OpenAI was ordered to preserve consumer output logs including deleted chats, an obligation that only ended on 26 September 2025. Whatever the policy says this quarter, the text you paste stops being yours to delete.
Vendor telemetry from Cyberhaven put a number on how often that matters: it reported that 4.7% of employees had pasted company data into ChatGPT and that around 11% of pasted content was confidential. Directional, not gospel — but nobody who has watched a support team work will find it surprising.
Scrubbing first turns the question from “do I trust this vendor with my client’s name” into “is this abstracted text useful enough to get a good answer”. It usually is. The longer version is here.
Frequently asked questions
What does ScrubLocally actually do?
It reads text you paste, highlights every value that looks like personal or secret data — names, emails, phone numbers, government IDs, card numbers, API keys, addresses, IP addresses — and gives you a version where each one has become a stable placeholder such as [NAME_1]. You paste that safe version into an AI chatbot. When the model answers, you paste its reply back and the real values return, using a mapping that never left your browser tab.
How is this different from the browser extensions that do this?
An extension has to be able to read every page you visit; that is a much bigger permission than most people realise they are granting. This is a web page: it can only see what you paste into it, it installs nothing, and it works in any browser including a locked-down work laptop. It is also reversible — most extensions scrub one way and leave you retyping the real names into the answer.
Is my text really not uploaded?
Yes, and you can prove it in ten seconds. Open DevTools (F12) → Network, paste a document and watch: no request is made. Or disconnect from the internet entirely — the page keeps scrubbing, because the detectors are JavaScript running in your tab. The site is a static site with no upload endpoint and no server-side processing anywhere.
Do the AI companies really train on what I paste?
On consumer plans, by default, several do. OpenAI states that ChatGPT improves by training on conversations unless you opt out, while Business, Enterprise, Edu and API usage is excluded by default. Anthropic changed its consumer terms in August 2025 so Free, Pro and Max chats are used for training unless you opt out, with retention of up to five years; Team and Enterprise sit under separate commercial terms. Policies change — check the settings in the account you actually use.
Is a scrubbed document anonymous?
No. Replacing values with placeholders is pseudonymisation: the mapping still exists, so under GDPR Recital 26 and the EDPB’s 2025 guidelines the data remains personal data. And a document can identify someone through context alone — “the only paediatric cardiologist at our clinic” names nobody. Use the tool to remove the identifiers, then read what is left.
Is it free, and what is the catch?
Free, no account, no email, no daily cap, no watermark. Your device does all the work, so running the site costs almost nothing. If it ever carries ads, they will sit outside the tool and cannot see your text — the no-upload architecture is not a policy that could change, it is how the site is built.