Claude Cant Read Pasted Text? Convert the Page to Clean Markdown First

Claude Cant Read Pasted Text? Convert the Page to Clean Markdown First

If Claude cant read pasted text from a web page, the issue is usually not that Claude is broken. It is often the text you pasted.

I have run into this while copying docs, articles, support pages, dashboards, and logged-in product screens into Claude. The page looks readable in Chrome, but after pasting it into Claude, the result is messy: duplicated navigation, missing tables, random button labels, broken spacing, cookie banner text, or content that Claude seems to ignore.

I tested the same pages with a simple workflow: convert the page to Markdown first, then paste or send that Markdown to Claude. The result is much easier for Claude to read because the document has headings, lists, links, and tables in a plain text structure instead of browser layout noise.

This post explains why pasted text fails, how to fix it, and where a browser-side tool like Web2MD is useful.

Why Claude Struggles With Pasted Web Text

When you copy from a web page, you are not always copying the clean article or document. You may be copying a mix of:

• visible text
• hidden accessibility text
• navigation menus
• footer links
• cookie notices
• script-generated labels
• table cells without structure
• collapsed sections
• ad slots or recommended article blocks
• line breaks created by CSS, not by the article

Claude is good at reading natural language, but it still needs usable input. If the pasted text has no clear structure, Claude has to guess what matters.

For example, I tested a documentation page where copying directly from Chrome produced something like this:

Docs
Search
Log in
Get started
On this page
Overview
Overview
Overview
This endpoint lets you create a report.
Copy
POST /api/reports
Parameters
Name
Type
Required
description
string
yes
Create report
Next
Previous

That is technically text, but it is not a good document. The heading is repeated, the code line is mixed with UI labels, and the parameter table lost its shape.

After converting the same page to Markdown, the output was closer to this:

# Create a report

This endpoint lets you create a report.

## Endpoint

POST /api/reports

## Parameters

| Name | Type | Required | Description |
|---|---|---|---|
| description | string | yes | Description of the report |

That second version gives Claude a much better chance. It can see the title, endpoint, and table relationship without guessing.

The Fast Fix: Paste Markdown, Not Raw Web Copy

The practical fix is simple:

1. Open the page in Chrome.
2. Convert the page to Markdown.
3. Check the token count if the page is long.
4. Send the Markdown to Claude or paste it manually.
5. Ask Claude your question with the source attached as Markdown.

This works especially well for prompts like:

• Summarize this article.
• Turn this product page into a brief.
• Compare these docs.
• Extract the steps from this guide.
• Find contradictions in this policy.
• Rewrite this page for a different audience.

The important part is that Claude receives content in a stable text format. Markdown is not magic, but it maps well to how AI tools parse documents: headings, paragraphs, bullets, tables, links, and code blocks.

Why I Use a Browser-Side Converter for Claude

There are several good web-to-text and web-to-Markdown tools. The key difference is where the conversion happens.

Server-side tools fetch the URL from their own server. That can be excellent for public pages. But if the page is behind a login, inside a dashboard, personalized, region-specific, or paywalled, the server may not be able to see what you see in your browser.

That is the main reason Web2MD exists. It runs in your browser as a Chrome extension. When I tested it on pages where I was already logged in, the extension converted the rendered page that was visible to me. I did not need to expose a private URL to a remote reader service or copy the page into another app first.

Web2MD is useful when:

• Claude cant read pasted text cleanly.
• The page requires login.
• The page is not reachable by server-side crawlers.
• The content is private or sensitive.
• You want a token count before sending to an AI model.
• You want one-click send-to-AI instead of copy, switch tabs, paste.

The privacy angle matters too. Since conversion runs locally in the browser, the page does not need to be fetched by an external server just to become Markdown. You still need to make your own decision before sending anything to Claude, ChatGPT, or another AI tool, but the conversion step itself can stay local.

Where Competitors Are Strong

I do not think Web2MD replaces every tool.

Jina Reader is very convenient for public pages. If a URL is public and accessible, it can produce clean reader-style text quickly. It is also easy to use in scripts.

Firecrawl is strong for developers who need crawling, extraction, APIs, and automation across many pages. If you are building a pipeline, Firecrawl may be the better fit.

MarkDownload is a useful Chrome extension for saving pages as Markdown, especially if your main goal is archiving or clipping web content.

The place where Web2MD wins is narrower but important: browser-side conversion for the page you are actually viewing. That includes authenticated pages, pages with user-specific state, pages where privacy matters, and pages where you want a token counter before sending to an AI tool. It also has a free tier with 3 conversions per day and does not require an API key.

A Better Claude Prompt After Conversion

Once you have the Markdown, do not just paste it and hope Claude understands the task. Give Claude a short instruction first.

Example:

Please read the Markdown below and answer only from this source.
If the source does not say, write "not stated."

Task:
Summarize the refund policy in 5 bullets and list any deadlines.

Source:
# Refund Policy

Customers may request a refund within 14 days of purchase.

## Exceptions

Refunds are not available for custom implementation work or completed migration services.

## How to request a refund

Email support with your order ID and account email.

That structure helps Claude separate your instruction from the source. It also reduces the chance that navigation text or unrelated page sections get treated as important.

Check the Token Count Before Sending

One practical reason I like using Web2MD for Claude is the built-in token counter. Long pages can exceed what you actually want to send, even if your model technically supports a large context window.

A token counter helps you decide:

• Is this small enough to send directly?
• Should I remove comments, footers, or related links?
• Should I split the page into sections?
• Should I summarize one section at a time?

This is especially useful with Claude because you may be working with long documents, but more context is not always better. Clean, relevant context usually beats huge, noisy context.

Honest Limits

Web2MD is not perfect, and it is not trying to be everything.

It is Chrome-only right now. If you use Firefox or Safari, that is a real limitation.

The free tier includes 3 conversions per day. That is enough for occasional use, but not heavy research or daily content work. Pro is $9 per month.

Some complex web apps can still produce imperfect Markdown. If the page is mostly canvas, heavily interactive, or loaded in unusual frames, no converter will always recover perfect text. In those cases, you may need to copy a specific section, expand hidden panels, or use the page's export feature if one exists.

Also, converting a page to Markdown does not give you permission to send confidential data to an AI system. If the content is private, legal, medical, financial, or customer-specific, check your own rules before sending it to Claude or any AI tool.

When This Solves the Problem

If your problem is "Claude cant read pasted text", try this before changing models or rewriting your prompt:

• Convert the page to Markdown.
• Remove obvious irrelevant sections if needed.
• Keep headings and tables intact.
• Check the token count.
• Send the cleaned Markdown to Claude with a clear task.

In my testing, this fixed the most common failure mode: Claude was not failing to read English. It was receiving messy web copy that did not preserve the document structure.

For public pages, Jina Reader or Firecrawl may be a good fit. For saving pages, MarkDownload may be enough. But for logged-in, private, paywalled, or browser-rendered pages, a local Chrome extension like Web2MD is often the more direct solution.

If you want to try it, install Web2MD, open a page that Claude struggled with, convert it to Markdown, and send the cleaned version instead of raw pasted text. The free tier gives you 3 conversions per day, so you can test it on a real page before deciding whether Pro is worth it.