How to turn a webpage into Markdown

Which guide fits the job

Read these four pages in the order of the decision, not in the order they were published. The first question is whether you already have Markdown you trust. If the preview is still empty, or it is full of navigation and share buttons, you are still in conversion. That is Convert a webpage to Markdown. It tells you which of three surfaces can see the page: the URL to Markdown tool, the HTML to Markdown tool, or the Chrome extension on the open tab.

The second question is where the text goes after the preview looks like the article. A file on disk, a note you will file later, or an offline copy of the figures is the save guide. A chat box is the send guide. People often do both, and the calmer order is to save a copy you can cite, then send a copy of that same text. The send guide explains why some chats open with the prompt filled in and others wait for a paste.

The third question is format, and it is easy to skip until a prompt gets expensive or confused. HTML vs Markdown for LLMs is the page for that question. It shows what full-page HTML still carries, what article HTML still carries, and what Markdown keeps. The numbers on that page are an illustration of one imaginary essay. Use them to understand the shape of the tradeoff. Use your own preview when you need the truth about a specific URL.

If you already know the destination is a research folder, an Obsidian vault, or a docs diff, the use cases pick up after these guides. The guides stop at a correct Markdown result. The use cases describe what you do with the file tomorrow.

What each guide actually teaches

Convert a webpage to Markdown is the map of the product. A public blog, an RFC mirror, or an open docs page can go through the URL tool, which calls the conversion API and returns JSON to the page. HTML you already copied, including a fragment from view-source, goes through the HTML tool, which converts in the browser and strips script, style, and similar non-content tags before conversion. A tab you reached by logging in goes through the extension: the article is taken from the live page, converted to Markdown, wrapped with an optional template, and formatted when formatting is on. The guide also lists what to check before you trust the preview: heading levels, tables, and fenced code.

Send a webpage to ChatGPT, Claude, or Gemini separates two senders that look similar and behave differently. On the web tools, ChatGPT, Claude, and Perplexity can open with the prompt in a query parameter, Gemini is a copy-and-paste, and any prompt that would make a URL longer than 8,000 characters is copied while the assistant home page opens. In the extension the built-in platforms with a query template are ChatGPT, Claude, Perplexity, and Grok. A prompt that is too long is copied in full and a shortened URL is opened, with a toast that says the content was truncated. The guide also lists the built-in prompts, because several of them substitute only the page URL, not the article.

HTML vs Markdown for LLMs is a comparison, not a converter manual. It keeps one running example so you can see why chrome dominates a full-page paste, why a code-heavy page shrinks less, and why a vision request for an image is a separate problem from an image URL in Markdown. It states plainly when HTML is the right input: you need attributes, ARIA, or a browser that can click. It also points at the developer API for people who want the same public-page conversion from a script.

Save a webpage as Markdown is about filenames and archives. The web tools slugify a download name. The extension’s single-file download underscores the title and strips characters that are unsafe in file names. Download .zip uses the page title as given, packs the Markdown, and rewrites successful images to images/<file>. The guide walks through a failed image, a data URL, and a page with no remote images, using the toasts the extension actually shows. Templates, including the built-in Basic Frontmatter template, are covered there because the file you save is the templated text, not a second hidden export.

A ten-minute first capture

Use a public essay you are allowed to store. Skip anything behind a login for this first pass, so you can compare the URL tool and the extension on the same document.

1. Open URL to Markdown and paste the link. Wait for the preview. Read the first heading and the first code sample if the page has one.
2. Copy the Markdown into a scratch note. That is the fastest save, and it keeps remote image URLs.
3. If you want a file name, download the single .md. The tool lowercases the title, turns other characters into hyphens, and cuts the name at 60 characters.
4. Install the extension and open the same URL in a tab. Convert again. The two converters are not the same code path. A small difference in a fence or a list marker is expected. A missing section is a reason to look at the preview, not to average the two files.
5. In the side panel, open the download menu and choose Download .md if you only need text. Choose Download .zip if a figure has to travel with the note. Scroll the page to the bottom first when images load as you scroll. The pack step can only fetch images that are already in the Markdown.
6. Optional: Send to AI on the web preview, with ChatGPT selected, only if the article is short enough to fit a query string. If the button switches to show that the prompt was copied, paste it into the chat that opened. Long pages should be saved, then uploaded or pasted, which the send guide covers.

After that pass you know which surface you will use on Monday. The URL tool is the one you can hand to a teammate who does not want an install. The extension is the one that can see the wiki you reached after single sign-on. The HTML tool is the one for a snippet that never had a clean URL.

What a careful preview looks like

This is a shortened example of the shape you want, not a capture from a live site. Headings are headings, the source URL is visible, and the code is fenced. A sidebar of related links is absent.

# Cache invalidation notes

Source: https://example.com/posts/cache-invalidation

## What changed

Writers delete the key on update. A short TTL only hides a stale value until that delete exists.

## Snippet

```text
DEL user:42:profile
```

If your preview still contains “Share”, “Cookie settings”, or a list of unrelated posts, go back to the conversion guide before you send or file the text. Cleaning that residue later, inside a chat transcript, is how a research note starts citing the menu.

Mistakes that waste the next hour

• Pointing the URL tool at a localhost docs site or a private IP. The API refuses those targets. Use the extension on the tab.
• Assuming the built-in “Summarize” prompt in the extension contains the article. The stock template quotes the URL. Edit the prompt so it includes {{markdown}} when the model must read the body. The send guide lists every stock prompt.
• Expecting Download .zip from the website tools. The archive is an extension action. The website download is one Markdown file.
• Moving a ZIP’s Markdown into a vault and leaving the images folder behind. Successful links are relative (images/photo.png). They resolve next to the note.
• Treating the extension output and the API output as byte-stable. Pipelines that hash a file should pin one path. The developers use case explains the split.
• Looking for a Notion export in the side panel. That button is not currently available. Paste the Markdown yourself if Notion is the destination.

Where to go from here

Feature pages describe a single control in more detail than a workflow does. Markdown conversion, templates, table of contents, Send to AI, and ZIP export are the references. Pricing is the plan page. Changelog is where behavior changes are listed, which matters if a guide and a screen ever disagree: trust the screen, then check the changelog.

If you are choosing between WebInk and another clipper, the comparison pages are MarkDownload and Obsidian Web Clipper. Those pages are honest about jobs WebInk does not do, including creating a note inside an Obsidian vault.

FAQ