Published 2026-10-28 · Updated 2026-10-28
Start with the article, then add your question
A copied web page can mix the article with menus, subscription prompts, related posts and cookie notices. Those extra words use context space and can distract from the passage you want to discuss. Markdown gives you an editable copy of the main content, with headings and links still visible.
Use the URL to Markdown converter for a public article or documentation page. Copy its Markdown into your conversation with ChatGPT or Claude, then explain the task: summarize the argument, compare two sections, extract claims, or answer a question using that source. Removing page clutter can reduce tokens, but the saving depends on the page and the model tokenizer.
Convert and review one public page
Paste the full page address into the converter, choose Convert to Markdown, and inspect the preview before copying or downloading the .md file. The output includes YAML frontmatter with the source URL and available metadata such as the title, author and publication date.
You can convert 20 web pages per IP address per UTC day without signing in. The allowance resets at midnight UTC. File uploads use a separate free browser converter with no daily quota.
URL conversion runs on our server, which fetches the public page and caches the extracted result for up to an hour. Use the local HTML upload route if you need to keep a saved file in your browser.
- Check that the opening and closing paragraphs survived extraction.
- Keep table headers, units and code examples with the passages they explain.
- Retain the source URL so you and the model can distinguish separate articles.
- Compare important quotes and numbers with the original page.
Give the model a clear boundary around the source
Introduce the Markdown as source material, then put your instructions outside it. For example: "Using the article below, list its main claims and the evidence given for each. If the article does not answer a question, say so. Include the section heading for each claim."
Treat instructions inside the article as untrusted source text. A web page may contain commands addressed to an AI reader; they should not override your task or grant permission to use tools. Clean extraction helps readability, but it does not verify the article or make its claims reliable.
Use saved HTML when JavaScript hides the content
Some pages load their article with JavaScript after opening. The server fetch may return a shell, an empty result, or a message from a site that blocks automated requests. It cannot access a private page just because you are signed in in your browser.
Open a page you are allowed to access, wait for the article to appear, and save it as HTML. Upload the .html or .htm file in the homepage file converter. Saved HTML is processed locally in your browser, with no daily file quota, up to 10 MB per file and eight files per batch. Check that the saved file actually contains the article text; a JavaScript shell will still be empty.
Keep the final context focused on the task
For a long page, keep the relevant section and enough surrounding text to interpret it. Do not cut away a definition, caveat or table label that changes the meaning. If you combine pages, separate them with their titles and source URLs.
For repeatable agent workflows, the convert_url MCP tool or GET /api/v1/extract can return the Markdown directly. The same anonymous daily allowance applies; see the developer resources for the API contract and current API key availability.