Safe Paste
Hide names, emails, phone numbers and account numbers before you paste into an AI chat, then put them back in the answer. The AI still understands the task; it just never sees who it is about. Nothing you paste leaves this browser.
New to this? What it does, in plain words
It swaps the private details in your text for labels like [NAME_1], so you can ask an AI for help without handing over who and what it is about.
Use it when the text is about a real customer, employee, patient or deal. A support lead can draft a reply to a complaint. An HR manager can get help with a sensitive email. A finance team can summarize a contract without the account numbers in it.
What you need
Nothing. No API key, no sign-up, no install. It works with ChatGPT, Claude, Gemini, Copilot or any other AI, on a phone too. Check your company’s AI policy as well: this lowers the risk, it does not make every text safe to share.
How to use it
- Paste your text. Private details are found and listed as you paste.
- Untick anything that does not need hiding, and add anything that was missed, such as a client or project name.
- Copy the safe version, or open it straight in ChatGPT or Claude.
- Paste the AI’s answer into step 4 and press Put details back.
Your text
What will be hidden
Names, emails, phone numbers, card and bank numbers, ID and account numbers, dates of birth and addresses are found as you paste.
Safe to paste
Put the details back
This page makes no calls with your text. The list of what was hidden stays in this tab and is gone when you close it.
Get AI’s help without giving away who it is about
Most of what makes a text sensitive is a handful of details: names, contact details and numbers. The AI rarely needs them to do the job.
- In this browser. Paste your textAn email, notes or a document about real people.
- In this browser. Details become labelsSarah Khan becomes [NAME_1] everywhere, so the AI can still tell people apart.Instant
- In the AI you use. Ask the AIIt drafts, summarizes or rewrites, using the labels.
- In this browser. Labels become detailsPaste the answer back and every label turns into the real name or number again.
What it finds
- People. Names after a greeting or a sign-off, after Mr, Ms or Dr, after “Name:”, and names that act in a sentence (“Sarah Khan said”). Two capitalized words in a row are offered as possible names, for you to keep or untick.
- Contact details. Email addresses, phone numbers in local and international formats, street addresses and postcodes.
- Money and accounts. Card numbers (checked with the same checksum banks use), IBANs (checked the same way), and account, policy, invoice, order, employee and patient numbers that follow a label.
- Identity. US Social Security numbers, UK National Insurance numbers, Indian PAN and Aadhaar numbers, dates of birth and IP addresses.
- Anything else you name. A client, a project code name, a product that is not public yet. Tick “Always hide” to keep the list for next time.
What it cannot do
Rules miss things a person would spot: a name written in lowercase, a job title that identifies one person, or a detail that only means something inside your company. Read the safe version before you paste it. Hiding names also does not make every document safe to share: if your company’s policy says a kind of document must not go into an AI tool at all, that still applies. The AI at work risk checklist covers the rest.
Questions
Does my text go anywhere?
No. Finding, hiding and restoring all happen in this browser, and this page sends nothing you paste to thinkhard.si or anyone else. The list of what was hidden is kept for this tab only, so you can restore the answer after a reload, and it is gone when you close the tab. Only if you press Open in ChatGPT or Open in Claude does the safe version go to that app, because you sent it there.
What if the AI changes a placeholder?
Restoring is forgiving: [NAME_1], NAME_1 and [Name 1] all count. Leaving “Ask the AI to keep the placeholders” ticked adds one line that asks the AI to keep them exactly, which makes this rare. If the answer contains a placeholder that was never yours, it is pointed out so you can check it.
Why does the same person always get the same label?
So the AI can still follow the text. If two people became the same [NAME], an answer about who said what to whom would be nonsense. Each different value gets its own number, and every mention of it gets the same one.
Can I use it with Copilot, Gemini or on my phone?
Yes. Copy the safe text and paste it into any AI, in any browser or app. Nothing needs installing, which also means it works on work computers that block browser extensions.