Anonymize a medical letter before an AI summarises or rewrites it
Paste a clinic letter, a discharge summary or a referral. The patient's name, date of birth, NHS number, relatives, phone numbers and the clinicians' names become codes; diagnoses, medication and doses stay readable, because that is what the AI has to work with. Restore its answer here.
Restoring uses the key of this tab: code your text here first, keep this tab open, then come back with the answer.
Adjust
See the key (it stays in this tab)
The key links each code to the real value. It exists only in this tab's memory: close or reload the tab and it is gone.
connect-src 'none' in its security policy) and loads no analytics. Turn Wi-Fi off once the page has loaded: the tool works just the same. The engine is the one inside the Hidden Tokens app.How it works
1 · Code
Paste a text or drop a file. Names, companies, addresses, emails, phone numbers, bank details, ID and case numbers become codes such as NAME-001 or COMPANY-001. Unknown proper nouns are coded too: when in doubt, the engine codes.
2 · Send the codes
Copy the coded text, or download the coded file, and give it to ChatGPT, Claude, Gemini or Copilot. The AI reasons on the codes: it still knows who did what, not who they are.
3 · Restore
Paste the AI's answer in the Restore tab, in the same browser tab. The codes become the real names again, on your screen only.
Example: a short clinic letter
Dear Dr Ferris,
Re: Mr Douglas Harrower, DOB 22/01/1957, NHS number 999 123 4567
I saw Mr Harrower in clinic today. He has stage 2 hypertension and type 2 diabetes. I have started amlodipine 5 mg once daily. Please repeat his bloods in six weeks.
His daughter, Mrs Carol Harrower, can be reached on 07700 900552.
Yours sincerely,
Dr Miriam Achterberg, Consultant PhysicianDear Dr NAME-001,
Re: Mr FIRSTNAME-001 NAME-002, DOB DOB-001, NHS number NHS-001
I saw Mr NAME-002 in clinic today. He has stage 2 hypertension and type 2 diabetes. I have started amlodipine 5 mg once daily. Please repeat his bloods in six weeks.
His daughter, Mrs FIRSTNAME-002 NAME-002, can be reached on PHONE-001.
Yours sincerely,
Dr FIRSTNAME-003 NAME-003, Consultant PhysicianThe patient and his daughter share the code for their surname, so the AI understands they are related without knowing the name. The diagnoses and the prescription stay readable: they are health data, and they are what the AI is asked about.
Health data needs more than coded names
Under the GDPR, health data is a special category: the bar for processing it is higher, and professional secrecy applies on top. Coding the identifiers is a strong first step; it is not the whole answer.
- Coded: the patient, relatives, clinicians, dates of birth, NHS and other ID numbers, addresses and phone numbers.
- Readable: diagnoses, symptoms, medication, doses, test results and dates of consultation. Without them the AI cannot help.
- The risk that remains: a rare condition, an age and a town together can point to one patient. The tool warns when it sees such combinations; generalise them (“in his sixties”, “a town in the region”) before sending.
Check your organisation's rules before using any AI with patient information, coded or not. The free tool handles one letter at a time; the Hidden Tokens app keeps one key per patient file on your computer and keeps a signed record of what the AI received.
The free tool or the Hidden Tokens app?
This free tool
- One text or one file at a time
- Up to 3,000 words or 1,000 spreadsheet rows
- The key lives in this tab and disappears when you close it
- Word, Excel, CSV and text
The Hidden Tokens app
- Whole folders: the same person keeps the same code in every document
- One key per case file, kept on your computer
- PDF and scanned documents, read on your computer
- Inside Claude Desktop, and in ChatGPT, Gemini and Copilot with the extension
- Industry packs, firm rules and a signed proof of what the AI saw
A word on the word “anonymize”
Strictly speaking, this tool pseudonymizes: each value is replaced by a code that can be reversed with the key. Under the GDPR, pseudonymised data is still personal data for whoever holds the key, which here is you, in this tab. What the AI receives no longer names anyone directly. Clues can remain, such as a job, a town and an age together: the tool warns you when it sees them. Read the result before sending it.
Questions
Is the diagnosis coded?
No. The clinical content stays readable because the AI needs it. Only what identifies the patient and the people around them is coded.
Are NHS numbers recognised?
Yes, an NHS number in its usual format is coded, as in the example. Check the coded text for local identifiers such as hospital numbers, and add them under Adjust → Also code if needed.
Can I use this for a whole patient record?
The free tool takes up to 3,000 words at a time, with a new key each time. For complete records, use the Hidden Tokens app, which keeps one key per file on your computer.
Does coding make it acceptable to use ChatGPT with patient data?
That depends on your organisation's rules and the AI service's terms, not on this tool alone. Coding reduces what the AI receives; it does not replace that decision.
Other ways to use the free tool
By file type
By AI tool
By task
For whole folders, use the app
Hidden Tokens codes whole case files on your computer with one key, works inside Claude Desktop and in your browser, reads PDFs and scans, and puts the real names back into Word and Excel files.