Importing PDF and Word Files

Most work in Return starts with a file you already have: a contract as a PDF, a draft in Word. Return opens these files as a read-only preview. To edit the text, annotate it, or ask the AI about it, you convert the file first. This article explains what that step does and what to check afterwards.

Why Return converts

Return works on text. The AI reads text, annotations attach to text, and every change the AI proposes is a change to text that you accept or reject.

A PDF or a Word file is not plain text. It is a description of a printed page: fonts, positions, margins. So Return makes a text copy of the document and works on that copy. The copy is a Markdown file, which is a plain text file with simple marks for headings, bold, and lists. You don’t have to learn Markdown. Return shows it formatted, and the file opens in any text editor if you ever stop using Return.

How to convert

1. Click the file in the Explorer.
2. Return shows a preview of the original.
3. Click Convert to Markdown. The button is in the TopBar, next to the file name, and in the panel on the right.
4. The converted document opens in the editor, ready for AI and annotations.

Conversion usually takes a few seconds.

A file that you open with Open document or drag into Return is converted straight away, without the preview step. See “Opening or dragging a file in” below.

Your original stays untouched

Converting never changes or moves the original file. Return reads it and writes a new file next to it, in the same folder, with the same name and the .md ending. Lease Agreement.pdf becomes Lease Agreement.md. If that name is taken, Return uses Lease Agreement-2.md.

The two files stay linked as long as they are in the same folder:

• In the converted document, View Source opens the original.
• In the original, Open Markdown opens the converted document.

If you move one of them to another folder, both files still work, but these buttons no longer connect them.

Opening or dragging a file in

A PDF or Word file that you open with Open document in the TopBar, drag onto the editor, or drop onto a folder in the Explorer skips the preview. Return converts it right away and opens the result as a new, unsaved document. The original is not changed here either.

Nothing is written to disk until you save. When you save, Return asks where to put the document.

Pictures appear after you save. Until then the document shows a line of text such as ![](assets/img-1a2b3c4d5e6f.png) where each picture will be. Saving places the pictures in an assets folder next to the saved document, and they show up in the editor.

Supported formats

Files larger than 50 MB are not converted.

What carries over, and what does not

Conversion keeps the words and the basic structure. It does not keep the look of the page. Fonts, colors, margins, and page breaks are gone, by design.

Word (.docx)

Kept: text, headings (the Heading 1 to Heading 6 styles, including headings that Word numbers automatically), automatic numbering, bulleted lists with their nesting, tables, bold, italic, line breaks, footnotes and endnotes, links with their web or email address, and pictures (PNG, JPEG, GIF, WebP).

Numbering. Word numbers clauses for you and recalculates them as you edit. Return writes those numbers into the text as Word displayed them: 1., 4.2., (a). They read the same in the editor and in every export, and the AI can refer to them. They do not renumber themselves when you add or remove a clause in Return, exactly like numbers you typed by hand. If the original used unusual numbering (lists continued across sections, restarts made by hand), compare the numbers with the original with View Source.

Tables. Every table becomes a table in the document. Cells merged across columns or rows become empty cells, so the columns stay aligned. A table placed inside another table’s cell is kept as text in that cell, and Return warns you. Word documents often use tables for layout, for example two-column signature blocks: these come through as tables too.

Footnotes appear as a numbered marker in the text, with the footnote text at the end of the document, in the form Obsidian and GitHub use for footnotes. Return shows them as text.

Pictures. Each picture becomes a file in an assets folder next to the converted document, and the document shows it where it was in the original. A picture that sat in the middle of a paragraph appears right after that paragraph. The files are named after their content, so converting the same document twice never creates duplicates or overwrites anything. Pictures inside a table cell are kept as a link in the cell.

Not kept:

• Pictures in formats Return cannot display (EMF and WMF drawings from older Word versions, TIFF, BMP, SVG). Return leaves a marker where each one was and warns you, naming the format.
• Charts, SmartArt and shapes. A marker and a warning, as above.
• Pictures linked from another file or a web address. Return never downloads anything during conversion.
• Headers, footers, and the text of comments. Return warns you how many comments the document had.
• Tracked changes. The result reads as if all changes were accepted: inserted text is kept, deleted text is left out. Return warns you when the document had tracked changes.
• Strikethrough. Text that was struck through by hand, without tracked changes, comes through as ordinary text. The strike is not shown, so check such passages against the original.
• Symbols inserted from a symbol font, such as Wingdings check boxes. They are left out, and Return does not warn you about them.
• Links to places inside the same document (bookmarks, table of contents entries) keep their text only.

PDF

Kept: text, paragraphs, headings (recognized by their larger font size), numbering exactly as printed on the page, and pictures: signatures, stamps, photographs and other images placed on the page land in the assets folder next to the converted document and appear where they were on the page.

Pictures. Return skips what is decoration rather than content: a logo repeated in the same place on every page, a background behind the text, tiny icons and rules. It tells you how many such images it left out. Return captures up to 64 pictures per PDF, and stops earlier if they add up to more than 100 MB or take longer than 20 seconds to prepare. Beyond that it leaves a marker and warns you.

Not kept:

• Charts and drawings made of lines rather than placed as pictures.
• Pictures on rotated pages. Return leaves a marker and warns you. A rotated page that is a scan is still kept as a page image.
• Table layout. The text is kept line by line.
• Bold and italic.
• Columns. In a page with two or more columns, lines from different columns can end up mixed. Single-column documents, like most contracts, are not affected.

Two things to expect in longer PDFs: page headers, footers, and page numbers show up as separate lines in the text, and a paragraph that runs across a page break is split in two.

Other formats

• OpenDocument (.odt): text, lists, tables and pictures are kept, with pictures handled the same way as for Word. Embedded charts and formulas are not.
• Rich Text (.rtf): text and basic formatting are kept. Tables and images are not, and Return warns you how many images the file had.
• Web pages (.html): headings, lists, links, and emphasis are kept.

Scanned PDFs

A scanned PDF is a set of photographs of pages. There is no text inside it to copy, and Return does not read text from images. Converting a scan gives you the pages as pictures: each scanned page appears in the document as an image named “Scanned page N” that you can read in the editor, and Return warns you which pages were scans and that their text was not recognized. The picture limits described above apply to scanned pages too, so a long scan is captured only in part: the pages beyond the limit are left out with a marker and a warning. In Cloud mode the AI can look at those page images when you ask it to, as long as the document is inside your workspace and the picture is under 3 MB (a dense colour scan can be larger).

A quick test: open the PDF and try to select a sentence with the mouse. If you can’t, it is a scan. To search or edit the text of a scan, run it through text recognition (OCR) in your scanner software or PDF application first, then convert the result.

Check the result

Treat the converted document as a working copy and the original as the record. Before you rely on the converted text:

1. Read the warnings Return shows after conversion.
2. Use View Source to compare against the original, especially clause numbering, tables, and anything that was an image.
3. Fix what matters directly in the editor.

Converting again

Re-convert, next to View Source, runs the conversion again and replaces the content of the converted document. Edits you made in it are lost, so Return asks for confirmation first. Use it when the original has changed and you want to start over.

Privacy

Conversion runs on your computer. The file is not uploaded anywhere to be converted.

• Local mode: nothing leaves your computer.
• Cloud mode: conversion itself sends nothing. If you click Analyze in the document details while Cloud mode is on, Return sends the first 3,000 characters of the document to its AI service to detect the document’s type and language and to write a short summary. These are saved in the document’s details. See AI Providers for what leaves your machine in each mode.

Troubleshooting

• “Conversion failed” on a PDF - the file may be damaged or password protected. Remove the password in your PDF application and try again.
• “Conversion timed out” - conversion stops after 60 seconds. Very large or complex files can hit this limit. Splitting the file into parts usually helps.
• The text has odd line breaks or stray page numbers - this is expected for PDFs, see above. Delete them in the editor.
• A .doc file won’t open - save it as .docx first.