Padlet to Notion
Over the past few months I've been moving a collection of bookmarks from Padlet into a Notion database. It seemed straightforward. Instead it became an accidental experiment in working with three AI assistants (Claude, ChatGPT, and Notion's built-in AI), each useful for a different part of the job.
This isn't a tutorial. It's a reflection on what it's like to use AI tools for a real, messy, multi-step project that doesn't fit into a single conversation.
The Problem That Wasn't One Problem
The plan was simple: export bookmarks from Padlet as a CSV, import into Notion, done. But anyone who's tried this knows what happens. When you import URLs via CSV, Notion treats them as plain text. You lose the preview cards with thumbnail images that appear when you paste a link manually.
Those thumbnails (Open Graph images) matter. The database is for design students browsing resources, and a wall of text links isn't inviting. Visual previews help people find what they need.
So the real project became getting hundreds of bookmarks into Notion with their preview images intact. Then, once the bookmarks were in, the tagging system I'd inherited from Padlet in the CSV turned out to be a mess: geography mixed with topics mixed with format types, vague categories, and the project expanded again.
ChatGPT: diagnosis, then a wall
I started with ChatGPT and the OG image problem. It explained why Notion doesn't fetch metadata from CSV imports and listed workarounds: manual pasting, automation tools like Make.com, or a Python script to pre-fetch the image URLs.
When I asked it to process my CSV directly and add the OG image URLs, we hit a limit. ChatGPT's environment couldn't access the web to fetch the images. It offered to write a script I could run locally instead. Good diagnosis, but the solution lived outside the conversation.
Claude: Iterating Through the Edge Cases
I brought the same problem to Claude, which could run code and attempt to fetch the OG images directly.
The first script worked, mostly, but some sites returned "no image found" even with images clearly visible on the page. Others timed out. Some threw 403 Forbidden errors that blocked the requests entirely.
What followed was a debugging collaboration. I'd report what failed and the script got refined:
Version 2 fixed a dimension-parsing bug and added fallbacks for finding images: Open Graph tags first, then Twitter cards, then structured data, then a scan for large images on the page.
Version 3 handled specific site types. Framer sites, for example, use non-standard image loading, so the script learned to look for data-framer-image-url attributes and CSS background images.
Version 4 addressed the 403s with rotating user agents, CloudScraper for sites with bot protection, and mobile user agents as a fallback.
Finally on the few remaining bookmarks where no image could be retrieved, I manually added them.
Notion AI: Restructuring the Mess
With the images sorted, the tagging system was the next problem. Legacy Padlet tags jumbled topics ("UX research," "Typography"), formats ("Tutorial," "Tool"), and geography ("London," "USA") together, alongside vague categories.
Notion's built-in AI worked well here because it was operating inside the actual database. I used it to design a cleaner taxonomy:
'Type' field, single-choice, e.g. Article, Video, Organisation, Individual, Asset Library, Knowledge Base
'Collections" field, multiple-choice, e.g. Agencies, People to follow, Design inspiration, Typography resources
"Region" field for geography e.g. for creative agencies
'Tags' field - many options, and the least useful to users, but potentially useful to me when putting together new Collections
I set up a migration workflow to batch-process old records, mapping legacy tags to the new structure and ran it through Notion's AI. Its value was contextual, helping organise what was already there.
What I Learned
Different tools for different phases. ChatGPT suited the initial diagnosis. Claude suited the iterative, code-heavy execution where I needed to run things and refine based on results. Notion AI suited the organisational work inside Notion.
AI doesn't remove iteration, it speeds it up. The image script went through four versions. Each failed in new ways the last hadn't anticipated. The tools didn't predict every edge case upfront, but they responded quickly when I reported back, and the fast feedback loop was the point.
Real projects are messier than demos. Every demo shows a clean prompt and a perfect response. Real work involves sites that block requests, data that doesn't match expectations, and problems that grow as you solve them. The useful skill is keeping the conversation going when things break.
The human stays in the loop. I was testing scripts, working out which sites failed and why, and making judgment calls about which tags mapped to which topics. The tools handled what I couldn't do efficiently myself, but direction and quality control stayed with me.
The Boring Truth
The project is still ongoing. The database exists, the images mostly work, the tagging is cleaner. There are probably edge cases I haven't found, and many bookmarks will need manual attention.
That's the unglamorous reality of using AI tools for real work. They don't deliver perfect solutions in a single prompt. They're collaborators in an ongoing process, sometimes useful, sometimes frustrating, each with different capabilities and contexts.
And I'll probably find a new system soon that promises to beat Notion, and start over again. It's a fun process though.
This article was written with the help of Claude, synthesising chat logs from conversations with Claude, ChatGPT, and Notion AI.