Connecting Zotero and Claude

Dear Reader,
​
After talking to many of my colleagues, I realised that for most academics, AI is near synonymous with using ChatGPT in the browser. But in the last 6-12 months (since Claude Code and Codex were released), things have dramatically changed, to the point where I feel that using ChatGPT wouldn't even cover 10% of what I do with AI today (See the second block in this email, if you're intrigued).

This all started a few months ago, when the co-founder of OpenAI proposed a new method to organise information that allows AI to handle large collections of data (ideal for academics, btw).

Before: If you asked a question, AI would search sources, extract relevant parts, and reply. This is called "RAG" for Retrieval-Augmented Generation. But with a growing number of documents, RAG systems perform worse and worse:

You might also have noticed a "lack of detail" when asking AI critical questions about your academic papers. That's also a much weaker form of semantic collapse and is related to context rot - the problem where, in long conversations, AI forgets what was said "in the middle". Context rot leads to dramatic forgetfulness (up to 94% in some tests, according to Zhang 2026).

2026 innovation/solution: Instead of having AI search sources, it creates a network of short connected notes on the concepts and links them together. This way, AI can do the same as a human: find a related concept and explore its notes for the query. The gain in accuracy and level of detail is enormous (I tested it!)

This idea has a name: "LLM Wiki".

Effectively, it's a Wikipedia of ideas linked together. Exactly what we have been doing manually here on EffortlessAcademic (see Information Management).

Turns out it is quite easy to build an LLM Wiki for your research using Obsidian. (Leading to an explosion in Obsidian's popularity). The hard part is not to outsource to AI what makes you a researcher!

This is the topic of this week's article:

After testing it, I found that the LLM Wiki poses a few problems that you need to be aware of to use it effectively, and it might even pose a danger to academic work altogether. Since you liked the more editorial format of last week's email, I also included a mini lit review on the problems with AI use for academics in the article, so you can make up your own mind if you want to follow this trend.

A system for your research with AI + Obsidian

If you use Obsidian for research, you've run into the usual friction. Where papers should go, how to structure a project, capture meeting notes, or the perpetually breaking Zotero connection. You've also seen the upside: searching your whole vault at once, writing papers faster, staying on top of your teaching/mentoring obligations.

With the new technical developments, you can now eliminate most of this friction.

And it is because, for a few months now, you can run LLMs inside your vault instead of in a browser tab, and they can do things like organising and writing notes, connecting to other software (e.g. Zotero), or querying websites, forming very flexible and powerful workflows.

For example, you want AI to import a paper from Zotero, outline its argument, check it against notes you took last week, and pull follow-up reading from Consensus. You build a skill for it once, which takes about five minutes, and the agent runs the whole chain. It fetches and arranges; you check the result and make sense of it. AI is like a prep chef for your cooking!

The trifecta of notes, papers, and skills is what makes this different. AI stops being a chat window and starts operating on your actual material.

My guess is that academic productivity will increasingly depend on how well you've digitised your research and made it available to AI. The reason is that AI lets you test more questions and drop dead ends sooner, rather than grinding through them. To do that, it needs access to all of your academic materials, papers, data, notes, calendars, etc.

And that's what you can start building today!

With a small group, we set up this system in the last learning mastermind cohort. It was one of the highest-rated sessions, which is why I think it's worth running again for a wider group.

Given that it is a bit technical, I'd rather do it as a hands-on workshop than a webinar, so you leave with a fully functional system you can use right away.

Interested?

If you fill out the form to help me identify relevant topics for you, I'll send you a discount for the workshop when it starts. (It also acts as a waiting list, in case the seats need to be limited.)

Thanks for being part of EA!

Wishing you a wonderful weekend,

Ilya Shabanov, The Effortless Academic​

The Effortless Academic is a community, newsletter and learning place for the modern academic, where you can learn to leverage academic tools, digital note-taking strategies and AI for a stress-free, successful career.
​
If you're ready to dive in, start with the foundation of Digital Note Taking to remember and synthesise ideas, the AI Lit Review Course to learn about recent (AI) tools for finding, reading and skimming relevant academic papers, or the AI writing and publishing courses to effectively publish your ideas.