How to use NotebookLM from web pages without copy-paste cleanup

If you have tried to use NotebookLM from web pages, you have probably hit the same boring problem I did: the page is useful, but the text you paste into NotebookLM is a mess.

Navigation links come along for the ride. Cookie banners sneak in. Related posts, footers, newsletter boxes, and random sidebar text get mixed into the source. If the page is behind a login, many server-side readers cannot see it at all.

I tested this workflow while collecting sources for a research notebook: public blog posts, docs pages, a logged-in course page, and a paywalled article I already had access to in Chrome. The best result was not to paste from the page directly. It was to convert the page to Markdown first, check the token count, then send the cleaned text into NotebookLM.

That is the main use case for Web2MD: it turns the current web page in your browser into clean Markdown for AI tools like NotebookLM, ChatGPT, Claude, and Cursor.

Why NotebookLM works better with Markdown

NotebookLM is good at working across sources, but the quality of the source still matters. A messy paste gives it extra text to reason over. Sometimes that does not matter. Sometimes it changes the answer.

For example, I pasted a pricing page into a test notebook. The raw copy included header links, footer links, and a long customer quote carousel. NotebookLM summarized the page, but it treated some of the marketing navigation text as if it were part of the main page.

The Markdown version was cleaner. Headings stayed as headings. Lists stayed as lists. Links were preserved where they mattered. The model had less junk to sort through.

Here is the kind of output I want before adding a source to NotebookLM:

# How ACME handles audit logs

ACME stores audit logs for workspace events, including:

- user login
- file export
- billing role changes
- API token creation

Audit logs are available on Business and Enterprise plans.

## Retention

Business workspaces keep audit logs for 90 days. Enterprise workspaces can extend retention to 365 days.

That is much easier for NotebookLM to use than a paste that starts with "Home Pricing Blog Sign in Book a demo" and ends with thirty footer links.

The workflow I use

My workflow is simple:

1. Open the source page in Chrome.
2. Click Web2MD.
3. Review the Markdown preview.
4. Check the token count.
5. Send or paste the Markdown into NotebookLM.

The token counter matters more than I expected. When I am building a notebook from ten or twenty sources, I do not want one giant source to consume most of the useful context. If a page is too long, I can trim the Markdown before adding it.

For a NotebookLM research workflow, I usually keep the title, main headings, the sections I care about, and the original source URL. I remove comments, related reading blocks, and anything that looks like repeated navigation.

A cleaned source might look like this:

# Source: Vendor security documentation
URL: https://example.com/docs/security

## SSO support

The product supports SAML 2.0 and OpenID Connect for single sign-on.

## SCIM provisioning

SCIM user provisioning is available on the Enterprise plan. Admins can sync users and groups from identity providers.

## Notes for my NotebookLM project

Use this source when comparing SSO and provisioning requirements across vendors.

That last note is not from the page. I add it myself when I want NotebookLM to understand why the source is in the notebook.

Why browser-side conversion matters

A lot of web-to-Markdown tools work by fetching a URL from a server. That can be great for public pages. It is also where they run into limits.

If a page needs your browser session, the server often cannot access it. That includes logged-in dashboards, member-only documentation, internal tools, course pages, and some paywalled pages you can legally read in your own browser.

Web2MD runs in Chrome, so it sees the page you are actually viewing. If you are logged in and the page is rendered in your browser, Web2MD can convert that rendered page to Markdown. That is the part that made it useful in my testing.

This also has a privacy benefit. You are not sending a private URL to a third-party scraping API just to find out whether it can access the page. The conversion happens locally in your browser. If you choose to send the Markdown to an AI tool, that is your next step, not a hidden part of the conversion.

How it compares with other tools

Jina Reader is strong for quick public URL reading. I use it when I want a fast Markdown-like view of a public article or docs page. The limitation is access. If the content depends on my logged-in browser session, a server-side reader may not reach it.

Firecrawl is more of a developer and crawling tool. It is useful when you need structured extraction, crawling, and API access across many pages. For a NotebookLM source workflow, though, I often do not need a crawler. I need the one page I am looking at, converted now, without setting up an API key.

MarkDownload is a respected browser extension for saving pages as Markdown. It is a good option if your main goal is archiving pages into Markdown files. Web2MD is narrower and more AI-focused: token count, clean conversion, and one-click send-to-AI are built around the workflow of feeding tools like NotebookLM, ChatGPT, Claude, and Cursor.

The short version:

• Jina Reader is excellent for public URL reading.
• Firecrawl is strong for API-based crawling and extraction.
• MarkDownload is good for saving pages as Markdown.
• Web2MD is best when you are already in Chrome, the page may require login, and the destination is an AI tool.

Using Web2MD with NotebookLM

NotebookLM does not need anything fancy. It just needs readable source material.

After Web2MD converts the page, you can copy the Markdown and add it to NotebookLM as pasted text. If the page is long, use the token count as a warning sign. A very long source can still be useful, but I usually prefer several focused sources over one huge dump.

A few habits helped in testing:

• Keep the original URL near the top of the Markdown.
• Remove repeated navigation if it slipped through.
• Keep headings intact because NotebookLM uses structure well.
• Split very long pages into smaller topical sources.
• Add a short note explaining why the source matters.

That last point is underrated. NotebookLM can summarize a source, but it does not know your research goal unless you make it clear. A one-line note like "Use this for the vendor comparison section" can make later answers more focused.

You can read more about the extension on the Web2MD homepage, including how it converts web pages to Markdown for AI tools.

Limits to know before you try it

Web2MD is Chrome-only today. If you use Safari or Firefox as your main browser, that is a real limitation.

The free tier includes 3 conversions per day. That is enough for light research, testing, or occasional NotebookLM work. If you are building notebooks from many pages every day, the Pro plan is $9 per month.

It also does not magically bypass access controls. If you cannot open the page in your browser, Web2MD cannot convert it. It works with logged-in and paywalled pages when you already have access in Chrome.

And like any page-to-Markdown converter, it can struggle with unusual layouts. A heavily interactive app, a page made mostly of canvas elements, or a site that loads content in odd ways may need manual cleanup.

When this workflow is worth it

If you only add one clean public article to NotebookLM once a week, direct paste may be fine.

But if you use NotebookLM for research, product comparisons, market scans, course notes, technical docs, or client work, the cleanup step pays for itself quickly. Clean Markdown gives NotebookLM a better source. The token counter helps you avoid overstuffed notebooks. Browser-side conversion solves the logged-in page problem that many URL readers cannot handle.

That is the practical reason I keep using Web2MD: it fits the moment when I am already reading a page and thinking, "This should be in my notebook."

If you want to try the workflow, install Web2MD, open a page you want to use in NotebookLM, convert it to Markdown, and paste the cleaned result as a source. Start with a messy page. That is where the difference is easiest to see.