Remove HTML Comments

About this tool

Comments are for the people writing the page, not the people reading it. Ticket numbers, commented-out sections, staging notes and build stamps all ship to production inside the HTML, where they add bytes and occasionally leak something you would rather not publish. The usual fix — a <!--.*?--> search-and-replace, or reaching for a minifier — trades one problem for another: the regex deletes things that only look like comments, and the minifier reformats the whole document so the diff is unreviewable.

This tool does exactly one thing: the output is the input minus the comment bytes. No whitespace collapsing, no tag or attribute normalization, no re-indentation, no quote-style changes. Run it on a template and the diff shows the comments you removed and nothing else.

It also knows that not every comment is a note. Comments that carry meaning — Internet Explorer conditionals, server-side includes, licence banners — are recognized as kinds and kept by default, and everything else can be selected by regular expression.

Worked example

Input:

<!--! (c) 2026 Example Ltd -->
<h1>Hello</h1>
<!-- internal note: swap the hero copy -->
<p>Ship it.</p>

Output, with every control left at its default:

<!--! (c) 2026 Example Ltd -->
<h1>Hello</h1>

<p>Ship it.</p>

The licence banner survived because its text starts with !, which is the long-standing convention for "do not strip this". The note is gone — and the line it sat on is now empty, because the default is to change no whitespace at all. Set Blank lines to Drop lines a removed comment left blank and the third line disappears too:

<!--! (c) 2026 Example Ltd -->
<h1>Hello</h1>
<p>Ship it.</p>

Switch Output to the dry run and you get an inventory instead of markup, so you can check a rule before trusting it:

line,kind,action,comment
1,bang,kept,<!--! (c) 2026 Example Ltd -->
3,plain,removed,<!-- internal note: swap the hero copy -->

The report output answers the other question — how much this actually saved:

metric,value
comments_found,2
comments_removed,1
comments_kept,1
removed_plain,1
removed_conditional,0
removed_ssi,0
removed_bang,0
css_comments_removed,0
bytes_before,104
bytes_after,62
bytes_saved,42
percent_smaller,40.38

Controls

• HTML — the markup to clean, pasted in. Up to 5,000,000 bytes.
• Keep IE conditional comments — on by default. <!--[if lt IE 9]> … <![endif]--> and the downlevel-revealed split forms are instructions to old browsers about which files to load; deleting them changes behaviour, so it is an explicit choice.
• Keep server-side includes — on by default. A comment whose text starts with #, such as <!--#include virtual=… -->, is a directive the web server acts on. Removing it silently removes part of the page.
• Keep banner / licence comments — on by default. A comment whose text starts with ! is the industry marker for a copyright header or licence notice that must survive minification.
• Pattern and Pattern means — a regular expression matched against each comment's inner text (what is between the delimiters). Under Keep matching comments it is a protect-list: RELEASE keeps every comment mentioning a release, whatever kind it is. Under Remove ONLY matching comments the tool inverts — only matches are deleted and everything else stays, which is how ^\s*/?wp: clears WordPress block markers out of exported content without touching real notes.
• Also remove /* … */ comments inside <style> — off by default, so a run is purely an HTML-comment operation. The CSS pass is string-aware: a /* inside content: "/* not a comment */" is left alone.
• Blank lines — keep (default) changes no whitespace; trim drops lines a removal emptied while leaving lines that were already blank; collapse also folds runs of blank lines into one.
• Output — the cleaned HTML, the per-metric report, or the dry-run listing of every comment with its line, kind and fate.

What it will not mistake for a comment

This is where a regex quietly corrupts a file and this scanner does not:

• Raw-text elements. Inside <script>, <style>, <textarea> and <title> the HTML tokenizer is in a raw-text state, so var s = "<!-- hi -->" is a string, not a comment. Those regions are copied through untouched (apart from the opt-in CSS pass). Note that <pre> is not in that list — a comment inside a <pre> really is a comment.
• Quoted attribute values. <a title="<!-- hi -->"> contains no comment, and a > inside an attribute cannot end a tag early.
• Nesting. Comments do not nest in HTML. <!-- a <!-- b --> c --> ends at the first -->, so c --> is text — the same conclusion a browser reaches.

Limits and edge cases

The document is capped at 5,000,000 bytes; one byte over is refused rather than truncated. An unterminated comment — a <!-- with no closing --> — is an error that names the line, because the alternative is deleting the entire rest of the document. An empty document is an error too. The pattern uses the Rust regex engine, which has no backreferences and no lookaround; that restriction is exactly what guarantees linear-time matching, so nothing you can type here will hang the page. Selecting Remove ONLY matching comments with an empty pattern is rejected instead of silently doing nothing. Line numbers in the dry run are 1-based and refer to the input. Blank-line trimming tracks lines emptied by an HTML comment removal only, so a <style> line left blank by the optional CSS pass stays where it is. Everything runs locally in your browser; the markup is never uploaded.

FAQ