Key point. PromptScrubber does not call ChatGPT or any other model. It only rewrites text locally. You decide what to paste into the assistant afterwards.
The concrete problem
Useful prompts are often full of real context: a customer message, a project update, a contract clause or a bug report. Copying that text into a chatbot can reveal far more than the question you actually wanted answered.
Replacing every name with a blank or with X makes the prompt harder to follow. The assistant loses track of who did what, and you end up explaining the same relationships again.
A good preparation step keeps relationships while removing identities. That means stable placeholders, a short check of the result, and a clear decision about what should never leave your device.
Practical steps
- Write the prompt as you normally would, including the background the assistant needs.
- Paste it into PromptScrubber and review the highlighted entities before doing anything else.
- Keep role-like placeholders such as [PERSON_1] and [COMPANY_1] so the conversation stays coherent.
- Add anything the engine missed manually, especially internal names or project codes that are not structured identifiers.
- Copy the anonymized prompt into ChatGPT. When the answer returns, paste it into the restore box if you need the real values back.
How PromptScrubber helps
Detection is local and based on patterns, checksums and small dictionaries. Emails, phones, IBANs, card numbers, IPs and recognisable tokens are detected more reliably than names or organisations described in prose.
The mapping that allows restore stays in the current tab. If you export it, treat the JSON file as sensitive because it contains the original values.
Limits and responsibilities
- An assistant can still infer identities from unusual details. Remove context that is not needed for the question.
- The tool cannot guarantee that every identifier was found; review the preview before you paste.
- Using placeholders does not by itself make a prompt anonymous in a legal sense.
- PDF scans are not processed; DOCX extraction is text-only and does not preserve formatting.
- Restoring an answer puts the original values back into the text, so handle the result with the same care as the input.
Frequently asked questions
Does PromptScrubber send my prompt to ChatGPT?
No. It never contacts an AI service. It prepares the text in your browser; you choose whether and where to paste the result.
Why not just replace names with X?
Constant placeholders break relationships. If three sentences refer to the same person, [PERSON_1] keeps that link visible to the assistant without revealing the name.
Can I restore the answer I get back?
Yes, while the tab is open. Paste the answer into the restore field to put the original values back from the local mapping. Export the mapping only if you intend to keep and protect it.
Should I include API keys in a prompt?
No. Remove them before any external conversation. Recognisable keys and tokens may be detected, but you should also check the text manually.
Prepare your text
Open the tool: paste, review and anonymize text without uploading it.