Export text for AI tools

Why redact before you upload to AI

Pasting a document into a chatbot is a fast way to summarise, classify, or rewrite it. It is also a fast way to share identifying details with a third party you do not control. Lex Cloak lets you choose, per document, what you are producing: a redacted PDF that blacks out the original, or anonymized AI / LLM text you can paste into a chatbot with every detected sensitive part replaced by a stable label.

Lex Cloak removes identifying details before you share a document with AI. It does not analyze the document with AI itself, and nothing about your document leaves your computer when you export. The clipboard and the file you save are on your machine.

Producing AI / LLM text

Set the Output control, in the toolbar or in Settings under Redaction, to AI / LLM text. The setting is remembered per document, so one file can produce a redacted PDF while another produces text. In AI / LLM text mode the per-person role labels appear in the sidebar: leave a person on Auto, or label them by role, for example a patient or an examiner, so an AI can follow who-did-what across the document without ever seeing the real names.

Click Export Text and pick AI / LLM text. Lex Cloak copies the document's text to your clipboard with the detected sensitive parts replaced. It keeps the document's structure so most AI tools read it well: page breaks become visible rules, real headings stay headings, and absolute dates are rewritten as relative day numbers (Day 1, Day 15) so the timeline stays readable without the calendar dates. The output starts with a short note saying the text was auto-redacted and may be incomplete. Read it once yourself before you paste it anywhere, because auto-detection is not exhaustive and anything missed would travel with what you paste.

Export style panel

The Export style panel at the top of the menu controls how the AI / LLM text reads, and is remembered per document:

• Names. Role-typed replaces a person's placeholder with their role where one is known, for example a patient or an examiner. Numbered keeps the plain numbered placeholder.
• Ages. Exact keeps the numbered placeholder. Bracketed replaces an age with a life-stage band, minor, adult, or senior, so the cohort survives without the exact number.
• Dates. Relative rewrites dates as day numbers from the earliest date. Exact leaves them as numbered placeholders. When dates are relative, you can also share one date anchor across a batch so day numbers line up across documents you scanned together.

Use defaults restores the recommended set at any time. These choices change only the AI / LLM text, not the redacted PDF you download.

What is and is not in the exported text

• Included. The document's narrative text, in reading order, with detected sensitive matches replaced. Page numbers and section headings come through where they are part of the text layer.
• Replaced. Every match Lex Cloak proposed and you did not turn off, plus everything on the Always Redact list and any selections you made. Each replacement looks like a labelled placeholder so the surrounding sentence still reads naturally.
• Not included. Images, signatures, complex tables, page geometry, and anything inside an image without legible text. Visual context that the page conveys through layout rather than words is lost.
• Worth a manual pass. Auto-detection misses what it has not seen before. A name in an unusual format, an identifier in a language the engine has not learned, or a handwritten note may slip through. Read the text yourself before you paste it.

For the source-of-truth saved file, use preview, checklist and download. Export Text is a companion to the saved PDF, not a substitute for it.