Using AI to extract minutes for Google Notebook

Hi everyone,

Janina asked me to put this together to document how I used Claude to retrieve meeting minutes so they could be loaded into Google NotebookLM.

Some History
Mark Miller initially started a Notebook as a way for the Maturity Model task force to be able to have its minutes and open GitHub issues available to ask questions to.  To load pages into the Notebook efficiently, Mark used the NotebookLM - WebSync Full Site Importer<https://chromewebstore.google.com/detail/notebooklm-websync-full-s/hjoonjdnhagnpfgifhjolheimamcafok> Chrome extension. (Mark, if there's anything you'd add to this initial upload experience or if I got some of it wrong, please send an update.)

I joined Mark as co-admin to the Notebook in December. One of the reasons this happened is because I have a Google Pro account which increases the amount of "sources" you can have in the Notebook from 50 to 300. Mark had topped out sources with his initial loading of data.

Adding All the Minutes
In January, it was mentioned that it would be great to increase the Notebook's capabilities by including the minutes from 2022-2024. I took this task on since I'd need to load them through a Pro account.

I didn't want to use the same process Mark did because I didn't want 3 years of minutes to take up ~120 sources in the Notebook. They needed to be collapsed into fewer sources—one source for each year would be ideal.

I initially looked at the Gemini functionality built into Chrome. However, as I discussed with it what I needed, Gemini informed that it could not extract the exact text for me. It could only summarize what was on the page, or summarize across multiple pages.

The Solution: Claude
I'd recently gotten access to the Claude in Chrome extension and it proved to be an excellent tool to this.

Here are the steps I took to do this:

  *
I selected the Haiku model for this because this wasn't an overly complex task. It proved to be an acceptable choice.
  *
I started by telling it what I wanted. In my browser window I had the home page for the minutes archive<https://www.w3.org/WAI/about/groups/task-forces/maturity-model/minutes/> and I had it review the page so it understood what was there. Then I laid out the task:
     *
Go to each page of minutes for the meetings in 2022, 2023, 2024.
     *
From the page, extract the exact text of the minutes. Don't summarize it, but extract all of it exactly.
     *
Present the text as a markdown file in a code block.
        *
Why markdown in a code block? I like giving LLMs well formatted text when I can and asking for the extraction to be in markdown meant that headings were noted.
        *
I told it to put this in a code block so it could be easily copied into a text document.
  *
Initially I had it extract a single meeting to prove it understood the task and could do it. It did exactly what I wanted. My validation showed it extracted the text cleanly from the page.
  *
Next, I told it to extract the 2022 minutes. It went off to work on that, but it stopped after 3 or 4 months (I don't remember which). It was too much for it to do at once time (this might be a limitation of the Haiku model or the extension itself).
     *
In discussing with Claude, we decided to break it down into one month increments.
     *
Each time it finished a month, I'd get a notification so I could copy that into the text file, and tell it that it could continue. Along the way, I'd spot check some of the meetings to verify it was doing what it was supposed to. It was flawless.
  *
I did run up against a usage limit during the session. It told me I needed to come back in four hours to continue.
     *
I neglected to check my overall Claude usage monitor to see if this was a Chrome Extension limit or my account hitting a limit. I'm inclined to believe this was a Extension limit because I've do much more intensive jobs with Claude without hitting a limit that quickly.
  *
Over the course of three sessions, the three years of minutes were extracted, copied to text files (one per year as noted), and loaded into the Notebook.
     *
Three sessions makes it sound longer than it was. I spent between 3 and 4 hours gathering the minutes, and I did it while multitasking on other things.

An Even Better Way to Do This: Claude Cowork
Within days of extracting all the minutes, I got access to Claude Cowork. This would've streamlined the process even more. Cowork could've done everything the Chrome Extension did and it could've also written the text files.

Cowork is a major step up on agentic AI as it can work with the browser, and it can access areas of your computer you give it access to. It has so many options on how it can connect to things, along with skills it can access. Plus it can write skills for itself to make it repeatable what it does.

I've only had it for a couple of weeks at this point and I'm still learning all it can do. I've taken to asking it if it can do all manner of tasks that might take me a lot of time to do.

As an example, earlier today I wanted the CSUN schedule in a spreadsheet so I could start to organize what sessions I wanted to attend. I also wanted an easy way to see who was speaking and what companies they were from. I went to Cowork, gave it my requirements, and around 30 minutes later I had an Excel sheet in the folder I'd specified for it. One tab had the schedule, and another tab had all the speaker details ready to be sorted.

Notes

  *
Claude Cowork is labeled as a "research preview." It is available on all Claude paid plans and is currently only available as part of the MacOS app.
  *
The Chrome extension had an advisory: "Browser use is a beta feature that allows Claude to interact directly with websites on your behalf, which carries inherent risks. Please review Using Claude in Chrome Safely<https://support.claude.com/en/articles/12902428-using-claude-in-chrome-safely> before use." It is available on all Claude paid plans.
  *
For more details check out the Claude Cowork Getting Started page<https://support.claude.com/en/articles/13345190-getting-started-with-cowork> and the Getting Started with Claude in Chrome<https://support.claude.com/en/articles/12012173-getting-started-with-claude-in-chrome> page.

I hope this write up was helpful. If there are any questions, please don't hesitate to ask. I can't guarantee I'll know the answer but I'm happy to help where I can.

Thanks
Jeff



Jeff Adams (he/him)

Vice President – Accessibility Operations


[UsableNet]



jeff.adams@usablenet.com<mailto:chris.werely@usablenet.com>

[Certified Professional in Accessibility Core Competencies (CPACC)]