Free tool Webpage to Markdown: see what an AI reads from your page
Enter a URL and get the page as clean Markdown: the article without menus, cookie banners and footers, with headings, lists, tables, links and images kept. That is roughly the text an AI crawler or an LLM works with, and the format to paste into ChatGPT, Claude or a RAG pipeline.
- Main content extraction
- Fetch as GPTBot or ClaudeBot
- Token count before and after
Convert a URL or HTML to Markdown
Until you click Convert, the result is a sample: a made-up blog post on www.example.com, so you can see what gets kept and what gets dropped.
check.tapcub.com, which fetches the page once from Cloudflare’s network with the user-agent you chose and a From: tools@tapcub.com header, then returns at most 512 KB of HTML to your browser. Like many AI crawlers, it does not run the page’s JavaScript. The service records only the host name, status and timing for rate limiting, not the path, the query or the page content. Limits: 10 requests a minute and 200 a day per IP address. In Paste HTML mode nothing is sent at all.The page to convert. In Paste HTML mode it is optional and only used to complete relative links; nothing is fetched.
Some sites block AI crawlers. Pick GPTBot or ClaudeBot to see what they get.
Paste the page source (View source, or Copy outerHTML in developer tools). Scripts in it are never run.
Words
190
In the Markdown, CJK characters counted one by one.
Tokens (est.)
398
Of the Markdown. About 4 characters per token in English.
Tokens in the HTML
912
What the raw HTML would cost the same way.
Headings
4
H1–H6 kept in the Markdown.
What happened
- NoteMain content taken from the page’s <article> element.
- OKMarkdown is about 398 tokens against 912 for the raw HTML: 56% less.
---
title: "How to choose a tent for wet weather"
source: "https://www.example.com/blog/wet-weather-tent"
author: "Maya Chen"
date: "2026-10-14"
description: "Hydrostatic head, seams and ventilation: what keeps a tent dry in rain, explained with numbers."
lang: "en"
fetched: "2026-11-21"
---
# How to choose a tent for wet weather
By Maya Chen · 14 October 2026 · 6 min read
A tent that stays dry in a summer shower can still soak you on day three of a wet trip. Three things decide which one you get: the **waterproof rating** of the fabric, how the **seams** are sealed, and how well the tent **ventilates**.
## 1. Read the hydrostatic head correctly
The hydrostatic head (HH) is the height of a water column, in millimetres, that the fabric holds before it leaks. Treat it as a minimum, not a promise: coatings wear and pressure rises where you kneel.
| Part | Light showers | All-day rain |
| ----------- | ------------- | ------------ |
| Flysheet | 1,500 mm | 3,000 mm |
| Groundsheet | 3,000 mm | 5,000 mm |
## 2. Check the seams
- Taped seams on the flysheet and groundsheet
- Silicone-coated fabric needs liquid sealer instead of tape
- Fewer seams overhead means fewer places to leak

## 3. Plan for condensation
Most "leaks" on still, cold nights are condensation. Open both vents and pitch with the door away from the wind. See our [pitching guide](https://www.example.com/guides/pitching) for the details.
Version v1.0Updated Update history
What the numbers and notes mean
The sample shows the typical pattern: the article survives, the page furniture does not, and the token count roughly halves.
Words
Counted in the Markdown, so they are words a model will read. In the sample, 190 words of article remain once the cookie banner, menus, breadcrumbs, share buttons, newsletter box and footer are gone.
Tokens (est.)
A rough estimate: about four characters per token for English and one per Chinese character. Each model’s tokenizer differs a little, so treat it as an order of magnitude. Use it to see whether a page fits a prompt or a context window.
Tokens in the HTML
The same estimate on the raw HTML. The sample is a small, tidy page (912 tokens); real pages with inline scripts, styles and tracking tags often run to tens of thousands, so the saving is usually much larger than the 56% shown here.
How the content was found
In your browser, Mozilla Readability picks the main content. If it cannot, the tool falls back to the page’s <article> or <main> element, then to the body with menus, banners and footers removed. The notes say which one was used.
Few words on a page you know is long usually means the content is built with JavaScript. Many AI crawlers fetch HTML only and do not run scripts, so they see that short version too.
Three steps
One request per conversion. Pasting HTML sends nothing.
-
1
Enter the URL
Any public http or https page. To see the page as an AI crawler gets it, choose GPTBot or ClaudeBot under Fetch as. For a page behind a login, open it in your browser, copy the HTML and use Paste HTML.
-
2
Pick what to keep
Main content for articles and docs; full page when the menus matter. Turn links and images off to save tokens; turn front matter on to keep the source URL and date with the text.
-
3
Copy or download
Copy the Markdown into a chat, or download
page.mdfor a knowledge base, anllms.txtcompanion file or a RAG index.
What happens to each part of the page
The HTML is parsed as text; nothing on the page runs or loads. Then:
- Main content: Mozilla Readability (Apache-2.0) on a parsed copy of the page, else the
<article>/<main>fallback described above - Relative links and image addresses are completed against the page URL (or its
<base href>); lazy images usedata-src - Conversion uses the same rules as our Markdown converter: GitHub-flavoured tables, fenced code blocks with their language, nested lists
- Front matter values are written as quoted strings, so titles with colons or quotes stay valid YAML
| In the HTML | In the Markdown | Main content mode |
|---|---|---|
<h1>–<h6> | # to ###### | Kept |
<ul>, <ol> | - and 1. lists, nested by indent | Kept |
<table> | Pipe table with a header row | Kept |
<pre><code class="language-js"> | Fenced block ```js | Kept |
<a href>, <img> | [text](url),  | Kept, or dropped by the options |
<nav>, <footer>, <aside>, cookie and share boxes | — | Removed |
<script>, <style>, forms, iframes, SVG | — | Removed in both modes |
One page, four ways
The sample page (3.6 KB of HTML, about 912 tokens) converted with different settings.
| Settings | Words | Tokens (est.) | Compared with the HTML |
|---|---|---|---|
| Full page, links and images | 233 | 521 | −43% |
| Main content, links and images, front matter | 190 | 398 | −56% |
| Full page, no links or images | 230 | 358 | −61% |
| Main content, no links or images | 189 | 291 | −68% |
Dropping the menus and footer changes few words but many tokens, because every menu link carries a full URL. On a typical production page the HTML side is far heavier than this sample.
Why the Markdown can look wrong
Most surprises come from the page, not the conversion.
Content built with JavaScript
Single-page apps often send an almost empty HTML shell. The tool shows what that shell contains, which is also what many AI crawlers get. Server-side rendering or pre-rendering fixes it for both.
A firewall answered instead
Cloudflare, Akamai and similar services may serve a challenge page to data-centre requests or to AI crawler user-agents. The note says so; try Chrome as the user-agent, or paste the HTML from your browser.
Main content cut too much
On product pages and landing pages Readability can miss specs or price boxes. Switch to Full page and delete what you do not need.
Garbled characters
Pages in GBK, Big5 or Shift_JIS are decoded as UTF-8 by the fetching service. Copy the HTML from your browser and paste it instead.
See which AI crawlers come, and what they read
This tool shows what a crawler would get from one page. TapCub records AI crawlers separately from people across your whole site.
AI crawler report
GPTBot, ClaudeBot, PerplexityBot and other AI crawlers identified by name and behaviour, apart from human visits.
Learn moreCrawl compliance
Your robots.txt and llms.txt are fetched daily and compared with what actually crawled you.
Learn moreAI referrals
Visits that arrive from ChatGPT, Perplexity and other assistants show up as their own sources.
Learn moreWebpage to Markdown questions
Still have a question?
Chat with the team behind TapCub — we usually reply within a few hours on business days.
How do I convert a webpage to Markdown?
Paste the URL above and click Convert. The page is fetched once, the main content is extracted, and you get Markdown to copy or download as page.md. For pages behind a login, copy the HTML from your browser and use Paste HTML instead.
Why convert a page to Markdown for ChatGPT, Claude or a RAG pipeline?
Markdown keeps the structure a model needs (headings, lists, tables, links) without the markup it does not: tags, classes, scripts and styles. That cuts tokens, often by half or more, and gives chunking tools clean heading boundaries to split on.
Is this what GPTBot or ClaudeBot sees?
Close to it. Choose GPTBot or ClaudeBot under Fetch as and the page is requested with that crawler’s user-agent, without running JavaScript, which matches how many AI crawlers fetch. It is not their exact pipeline: the request comes from Cloudflare’s IP addresses, so sites that verify crawlers by IP may treat it differently.
Why is the Markdown almost empty?
The page probably renders its content with JavaScript, or a firewall served a challenge page. Check the notes under the counts. If the content only appears after scripts run, crawlers that do not run scripts miss it too.
What is the difference between main content and full page?
Main content keeps the article and drops navigation, breadcrumbs, cookie banners, share buttons, sidebars and the footer. Full page converts everything visible, which is useful for home pages, pricing pages and other pages without a single article.
Can I convert HTML I already have?
Yes. Switch Source to Paste HTML and paste it. Nothing is sent anywhere; the conversion runs in your browser as you type. Add the page URL if you want relative links completed.
What is the front matter for?
It keeps the title, source URL, author, publication date and description attached to the text, so a knowledge base or an LLM can cite where it came from. Turn it off if you only need the body.
How is this different from llms.txt?
llms.txt is a file on your site that lists your important pages for language models, ideally with Markdown versions of them. This tool produces those Markdown versions; the llms.txt generator builds the index.
Update history
New features, fixes and data refreshes for this tool, newest first.
-
· v1.0
LaunchedFirst release.
Missing something or found an error? Tell us — requests that come up often go into the next update. All tool updates
Check what AI crawlers are allowed to fetch
Markdown shows what a crawler would read. robots.txt and your firewall decide whether it gets that far.