Automate Browser Using Tampermonkey and ChatGPT

Introduction: Why bother tweaking the browser?

I’ve automated back-end stuff, infra, even my WhatsApp messages (coming in a post soon👁️👁️…)
But the front-end of my day (Chrome / Edge) stayed painfully manual.

Yet that’s where I:

• open the same dashboards 20× a day,

• poke the same buttons (download / export / filter),

• stare at UIs that are almost perfect — but not quite BosTheCoder-perfect.

Building browser automation would be the perfect solution to these types of problems.

The real power shows up when either of these is true:

1. You visit one page all the time and want it to fit your workflow.

2. You need to repeat the same action on many pages (think coupon finders, “save-to-Pocket”, Readwise-highlighter, etc.).

---

Potential use-cases

There’s SO MANY different ways I know I could benefit from some sort of Browser Automation. Off the top some that come to mind are:

• Instant download & sideload

  • Lets say a freind recommends you a book you should read by sending an amazon link.

  • You open it up and take a look and it actually looks 🔥so you want to save it so you can read later

  • Imaging in one click you could:

    • look up ISBN → queue in StoryGraph / Goodreads

    • fetch from Library Genesis → add to Calibre

    • shoot the EPUB to Kindle → sync all devices.
      That’s like 7 clicks → 1.

• Auto-file receipts

  • scrape order confirmation pages and dump PDFs straight into my “Receipts → YYYY/MM” folder on OneDrive.

• Hide spoilers

  • every sports page that mentions Arsenal’s score before I’ve watched the game? Insta-blur.

Once you think in “tiny browser robots”, you’ll spot ideas everywhere.

---

“Do I really have to build a Chrome extension?”

Nope.
The lazy-genius route is a userscript manager.

Some of the most used ones today can be seen in the table below:

A userscript is just a .user.js file that the manager injects at page-load. Pure JavaScript.
The beauty of which is that this means ChatGPT can write it for you in one shot pretty easily.

And if you use a good reasoning model like o3 it seems to get to the solution pretty quickly as long as you can fit the source for the web page into your prompt (more on this later 🥲)

---

Example: NeetCode blind-practice mode 🎯

As most of my friends and family know, I stay grinding coding interview type questions on NeetCode.io😎 (check out my last article if you don’t know).
It’s a GREAT website don’t get me wrong but there are a few things that I think would enhance my experience, mainly:

• The list shows the difficulty (Easy / Medium / Hard).

• Categories shout “Graphs”, “DP”, “LinkedList”.

Those are hints…I wanted zero signals. Just a shuffled list where every problem is a surprise.

In my head an easy solution to this would be to:

• Hide all category headers.

• Hide the Difficulty column.

• Merge + shuffle every table into one mega-table.

(Basically, turn the site into Leetroulette.)

Using ChatGPT to crank out the script

The prompt itself was relatively simple, it basically was just:

“Generate a Tampermonkey userscript that hides <p class='header'>…</p>, removes the Difficulty column, merges all <app-table> elements, and shuffles the rows. Must work on a single-page Angular site. See page source below: <page source>”

I did however run into a few issues…

Final script

You can grab the final script here and follow the instructions to use in TamperMonkey.
To use it navigate to NeetCode’s Problem list. :)

Take-aways

1. Userscripts give 80 % of a browser-extension’s value for 5 % of the effort.

2. ChatGPT (o3) + a crisp prompt = rapid DOM surgery.

3. Always remember SPAs mutate the page after load — listen for changes or you’ll rage-debug a “working” script.