Where do headers, footnotes and comments go?

The short answer

txtfetch does not strip headers, footers, footnotes or comments. All of them come back as plain text. You must remove what you do not want.

A Word header and a Word footer come back once for each section, not once for each page. A PDF header and a PDF page number come back on every page.

A Word footnote body follows its paragraph. Its mark becomes the literal text [footnoteRef:1]. An endnote does the same with [endnoteRef:1].

A Word comment follows its paragraph as Comment by Reviewer Name: …. The reviewer's name is in the text. A PDF sticky note, a PowerPoint comment and an Excel cell comment also come back with the name. Only an OpenDocument comment comes back without one.

The engine is tika-server-standard 3.3.1.

Source: terraform/lambda-layers/build/versions.env (versions.env: TIKA_VERSION), accessed 2026-10-20.

What comes back, case by case

We built one small file for each format. Each part holds a marker sentence, so you can see which part came back and where. We sent each file to the production API. Every row shows its source. The inputs and the raw responses are in the repo.

What this means for your pipeline

These parts land in your chunks. A header, a footer or a comment can end up inside a chunk next to body text that has nothing to do with it. Decide what to keep before you split the text.

Footnote marks stay in the text as [footnoteRef:N]. Remove them, or match each mark to its note. The first snippet below removes them.

Comments are the sensitive part. A Word comment puts the reviewer's name in the text. The metadata object holds no document properties. Remove the comments before you index a file that others have reviewed. The second snippet does it. What personal data txtfetch returns lists the other places a name can hide.

A PDF repeats its header and page number on every page. The filter for that is on Headers, footers and page numbers in every chunk. The snippets here do not repeat it.

How to remove them yourself

Run each snippet next to your txtfetch call. Each snippet ran on the saved response named under it.

Strip the footnote and endnote marks

The marks are literal text. A short regex removes them and keeps the note bodies.

Python, run on responses.json

import json
import re

with open("responses.json", encoding="utf-8") as f:
    text = json.loads(json.load(f)["notes-docx"]["text"])["extracted_text"]

# "[footnoteRef:1]" and "[endnoteRef:1]" are marks, not words.
cleaned = re.sub(r"\[(?:foot|end)noteRef:\d+\]", "", text)
print(cleaned)

It printed

HEADER two for section two
Body one.
Body two has a footnote.
FOOTNOTE text here.

Body three has a comment.	Comment by Reviewer Name: COMMENT body here.
Body four has an endnote.
ENDNOTE text here.

HEADER one Acme Confidential

FOOTER one text
Body five is in section two.

Split the comments off a Word response

Each comment follows its paragraph after a tab. Collect the comments, then remove them from the body. Use the plain text output. The markdown output has a space where the tab was.

Python, run on responses.json

import json
import re

with open("responses.json", encoding="utf-8") as f:
    text = json.loads(json.load(f)["notes-docx"]["text"])["extracted_text"]

# A comment follows its paragraph after a tab: "<tab>Comment by <author>: <text>".
COMMENT = re.compile(r"\tComment by ([^:\n]+): ([^\n]*)")

for author, body in COMMENT.findall(text):
    print("comment:", author, "|", body)

print(COMMENT.sub("", text))

It printed

comment: Reviewer Name | COMMENT body here.
HEADER two for section two
Body one.
Body two has a footnote.[footnoteRef:1]
FOOTNOTE text here.

Body three has a comment.
Body four has an endnote.[endnoteRef:1]
ENDNOTE text here.

HEADER one Acme Confidential

FOOTER one text
Body five is in section two.

What we did not test

We tested uploaded files, and one example of each part. We did not test a page that txtfetch fetches from a URL. We did not test the legacy .doc, .xls and .ppt formats. We tested tracked changes on a separate page. We did not test a first-page header, an even-page header, or a text box in a header. We did not test a comment with several paragraphs, or a reply to a comment. We make no claim about them.

Read Tracked changes, the Office page and Attachments and archives for the rest of each format.

Frequently asked questions

Does txtfetch remove headers and footers from a Word file?

No. In our probe the header text and the footer text of a Word file came back as plain lines. A Word header came back once for each section, not once for each page. Remove them on your side if you do not want them.

Where does a footnote come back?

In a Word file, the footnote body follows the paragraph that holds its mark. The mark itself becomes the literal text [footnoteRef:1]. An endnote does the same with [endnoteRef:1]. In an OpenDocument file the footnote number and body join the paragraph.

Do Word comments come back?

Yes. Each comment follows its paragraph after a tab, as Comment by, the reviewer's name, a colon and the comment. The reviewer's name is in the text. Remove the comments before you index a file, if the names or the notes are private.

Do comments come back from ODT, PDF, PowerPoint and Excel files?

Yes, in our probe. A PDF sticky note came back with the author name on its own line, before the note. A PowerPoint comment came back with the name and the initials. An Excel cell comment came back with the name, inside its cell. An OpenDocument comment came back without the author name, although the file holds one.

Does a PDF header come back on every page?

Yes. A header or a page number drawn on each page comes back once for each page, as a plain line. See /fixes/headers-and-footers-in-every-chunk for a filter.

Does txtfetch return the document properties, such as the author?

No. The metadata object holds no document properties. A comment is different. It is part of the text, so a reviewer's name can come back inside it.