An AI-generated meeting note or README needs different treatment depending on whether the recipient will read it or edit it. Changing the extension is not enough: headings, tables, reference links and images must reach the recipient in a useful form.
We exported the same short Korean-and-English sample through Toolit. HTML and DOCX kept table structures; TXT kept a link label but lost its destination. Images also showed why being visible and being embedded in a file are different things.
Decide what must survive the export
Start with HTML for browser reading, PDF for submission or printing, DOCX for Word editing, and TXT for plain body text.
The exported sample retained ‘툴릿 안내 / Toolit about’ but not the address behind that label.
HTML retained a remote reference. DOCX stored descriptive text and the URL instead of embedding the image.
What we checked with one shared sample
- 1
On September 26, 2026, we pasted a bilingual sample into the local Toolit build in Chrome on Windows. It contained four headings, a three-row/two-column table, a list, emphasis, a two-line code block, one web link, one web image, and one relative image.
- 2
We used Clean style, Standard width and a table of contents. The extra document title was off because the source already had an H1. We saved HTML with external images off and on, then DOCX and TXT with that setting still on.
- 3
We inspected actual HTML tags and addresses, DOCX heading/table/hyperlink/image entries, and the saved TXT strings. We did not treat the browser preview as proof of the final Word layout.
- 4
For PDF, we checked the opening of the print document and the print styles. Final PDF files and page-by-page Word rendering were not verified in this record. The measured table covers HTML, DOCX and TXT; PDF selection and checks are discussed separately.
What survived in the exported sample
Preservation does not mean pixel-identical appearance. Keeping heading text, retaining a heading structure, and using the same font and position are different tests. The DOCX column reports internal structures, not verified Word rendering.
| Feature | HTML | DOCX | TXT |
|---|---|---|---|
| Four headings | Four heading tags | Four heading styles | Text only |
| 3-row, 2-column table | One table structure | One table structure | Cell text only |
| Two-line code | Code block | Code text and formatting data | Code text |
| Web link | Label and destination | Hyperlink relationship | Label without destination |
| Allowed web image | Remote reference | Alt text/URL; 0 embedded images | No image or alt text |
| Relative image | Notice | Notice | Notice |
Plain text can lose information even when the words remain
The sample displayed the link label ‘툴릿 안내 / Toolit about’ with https://toolit.kr/about/ behind it. Only the label survived in TXT. For references that must be opened later, write important URLs into the body as well, or choose HTML or DOCX.
Cell text such as ‘초안 / Draft’ and ‘완료 / Done’ survived, but in this sample the cells became lines without tab separators. Do not treat TXT as an Excel-ready table or CSV. Use a format that preserves the table when column relationships matter.
This output extracts visible text; it does not reconstruct missing information. It is useful for removing formatting, but keep the original MD when links, table relationships and images must remain available for later use.
Why a shared HTML file may lose its pictures
With external images off, the saved HTML contained a blocking notice. With them on, it retained the web image URL. The image bytes were not embedded in the HTML, so connectivity and the image server's access conditions can affect later display.
The basic document styles are inside the HTML. That does not mean every image or font file is bundled too. Fonts available on another device can also change line wrapping.
A relative reference such as ‘./figure.png’ cannot be resolved from an MD file alone. It became a local-image notice in this sample regardless of the external-image setting. If pictures are essential, check their actual presence in the final document.
DOCX is for editing, not an exact copy of a web page
The DOCX contained four heading styles, one table, code text and a web hyperlink relationship. The web image became alternative text and an address, with zero embedded image entries. That is a choice in Toolit's current conversion, not a limitation of the DOCX format.
Insert essential pictures in Word and check table widths and code wrapping. Do not assume an HTML theme or table of contents is copied into DOCX unchanged. App-specific fonts, spacing and actual pagination were outside this structural check.
When should you choose PDF, and what should you check?
PDF is a practical starting point when the recipient needs fixed pages to read, submit or print rather than edit. Toolit uses the browser's printing workflow, which divides a continuous web document into paper-sized pages.
As MDN's printing documentation explains, screen and print styles can differ. Preview width is not the same as PDF paper width. Check paper size, orientation, scale and margins; check the background-graphics setting if colors matter.
We did not measure final PDF page count, links or image preservation in this record. Before submission, reopen the saved PDF and check the table's rightmost column, long code lines, searchable Korean text, and missing pictures.
Choose for the recipient's next action
Start with what the recipient will do, rather than the smallest file or the most familiar extension.
| Purpose | Start with | Check before sending |
|---|---|---|
| Read and follow links in a browser | HTML | Remote images; allowed attachment types |
| Submit or print | Accepted format; page breaks; clipping | |
| Edit text and tables in Word | DOCX | Reinsert images; inspect tables and code |
| Paste body text elsewhere | TXT | Whether links and table relationships matter |
| Maintain the original document | Keep the MD too | Keep linked images and attachments |
Official documentation used for this analysis
Choose by the information that must survive, not just the extension
In this sample, HTML and DOCX retained heading and table structures, while TXT retained readable words. Link destinations and web images particularly deserved a separate check.
Choose for reading, submission, editing or plain-text copying, then verify a table, a link and an image in the saved file. Keeping the MD source lets you make a different export without starting from a reduced version.