From Zotero Export to Thesis Draft: A Zotero Obsidian Workflow for PhD Researchers, Walked Paper by Paper

If you are mid-thesis, you already know the failure mode. The paper is somewhere in your Zotero library. The note about it is somewhere in your Obsidian vault. The paragraph it is supposed to support is somewhere in your draft. Around chapter three you notice you have re-found the same paper three times, on three different days, in three different places.

I stopped calling that a discipline problem. Re-finding is not laziness; it is a missing pipeline. A folder holds notes. A pipeline moves a paper. So this post is a walkthrough: one paper, followed from a Zotero export row all the way to a thesis draft section, with every intermediate stop named and the front-matter fields spelled out. At the end I apply the same idea to an unread queue, because the queue is where pipelines die.

You should be able to steal the whole scheme from this post without buying anything. What I sell at the end is the wiring diagram already drawn — conventions decided, templates locked, examples worked through — for the price of a coffee.

Stop 1: The Zotero export row

Everything starts as a row. In the plugin-free path you export a Zotero collection as CSV, and the row carries authors, title, year, venue, DOI, and the Zotero URL. It is raw material, not a note.

One practical warning before you export anything: citation keys and export formats vary between Zotero versions, so check that your export matches the documented header row in the Citation Pipeline Guide. An eight-column file with a differently named column will fail loudly, and loudly is where you want it.

Stop 2: The literature note — eleven locked fields

The Literature-Note template carries eleven front-matter fields: type, citekey, title, authors, year, venue, doi, zotero_url, status, triage, and tags. Six body sections sit below the front matter, so a literature note is a structured reading record — claims, methods, limits — not a pasted abstract.

"Locked" is the operative word. The field set is fixed; you do not decide per paper which fields you feel like filling that day. That is the entire point of a field contract: when every literature note has the same eleven fields, the vault can answer questions like "which papers are still in triage=defer?" without you remembering what you named the property. Field names are conventions, and conventions are exactly what erode first when you write them fresh per paper at 1 a.m.

The citekey follows a scheme rather than a lookup table: surname, year, first distinctive word of the title, diacritics stripped, leading articles dropped. A 2022 paper by Jörg Müller titled "The Structure of…" becomes muller2022structure — ASCII, stable, sortable.

One honest limit: the kit generates citekeys and front matter; it does not know your department's rules. Verify your institution's citation style before citing anything in submitted work.

Stop 3: Three atomic notes

From the literature note, the paper fans out into three atomic notes. The Atomic-Note template has seven front-matter fields and one iron rule: one claim per note, each with evidence and counter-evidence sections.

Why not one big note per paper? Because a claim you can link is a claim you can reuse in two chapters. A claim buried in a 900-word literature note is a claim you will re-read the literature note to recover. And the counter-evidence section earns its keep later: it is the section that stops your future self from citing a finding as settled when you originally wrote down exactly why it might not be.

Worked Example A in the kit shows the full file contents for this stage — a fake RCT with the citekey okonkwo2023spaced, its literature note, its three atomic notes, and the chapter MOC they feed. The paper is fake and the DOI sits on the reserved 10.5555 prefix, on purpose: you can copy the structure without worrying that you are citing something that does not exist.

Stop 4: The chapter map-of-content

The Thesis-Project-MOC template is a chapter argument spine plus a section list. Its Chapter-2-MOC file wiki-links the atomic notes under the argument they support, so the chapter's logic is visible in one screen: here is the claim of the chapter, here are the linked notes that back it, here is where each one is being used. When a new atomic note arrives from a new paper, it goes into the MOC or it does not exist yet as chapter material — a decision that takes ten seconds at capture time instead of an afternoon at write-up time.

Stop 5: The draft section, via a supervisor meeting

Worked Example B is the same fake project after a supervisor meeting dated 2026-10-14. The Supervisor-Meeting-Note template records three numbered decisions and a dated Task/Owner/Due action table; the meeting note links back into the project, the Chapter-2 MOC gets updated with drafting status, and out comes a draft section — ch2-3-retrieval-practice.md — citing claims that are already linked to the literature note that came from the export row in Stop 1.

That is the whole loop closed. The paper was never re-found. It was moved, five times, each move a small documented step.

Two ways into the literature note

You can reach Stop 2 by plugin or by file, and the kit documents exactly two import paths.

• The plugin path uses zotlit or the obsidian-zotero-integration plugin: the literature note is generated from Zotero directly, citekey included.
• The plugin-free path is a Zotero CSV export plus a bundled utility, tools/zoterocsv2notes.py. It is stdlib-only — nothing to pip install — runs on Python 3.8 or newer, makes no network calls, and ships with a --selftest that passes. Manual .bib field mapping is documented as the fallback for people who already keep a hand-maintained bibliography file.

Same destination, two doors. Pick the plugin if you live in Obsidian all day; pick the CSV path if you want the pipeline to survive plugin abandonment, which academic tooling does with some regularity.

Triage the queue like a reviewer, not a person with a backlog

A pipeline for processing papers is worthless if the papers never enter it, and this is the stage where most academic vault setups quietly fail: the reading queue runs on willpower, which is to say it runs on nothing.

The kit ships Reading-Manager.csv — a populated 12-row queue — with a closed triage vocabulary: read-now, skim, defer, reject. The shipped queue splits 3 / 4 / 3 / 2 across those labels. Closed means closed: there is no "maybe" and no "later-ish", because a label you can negotiate with is a label you will negotiate away. Every paper that enters the queue gets exactly one label, and the Paper-Review-Queue template sections your quarterly queue by those same four labels, so the CSV and the vault speak one vocabulary.

Triage is a decision, not a mood. "Skim" means you commit ten minutes and extract at most one claim. "Reject" is a real answer, and it is the one senior researchers actually give: this paper is not for my question, and pretending otherwise costs a reading slot.

The AI triage pack, with the policy caveat up front

For queues that are genuinely too large to label by hand, the kit ships AI-Research-Triage-Prompts.md: eight prompts covering Title/Abstract Triage, a Methodology Red-Flag Check, a Related-Work Gap Finder, a Claim-to-Atomic-Note Extractor, a Contradiction Steelman, a Supervisor-Meeting Brief Generator, a Related-Work Paragraph Drafter, and a BibTeX Field Mapper.

Two caveats belong in the same breath as the prompts, and the file states them too. First: check your institution's AI-use policy before doing any AI-assisted screening. Second: LLM triage is a first pass for your personal reading queue — it is not a substitute for documented human screening decisions, and a PRISMA-style review needs those decisions recorded, machine-assisted or not. The prompts accelerate triage; they do not outsource your judgment, and in a review context they must not.

The kit: the wiring diagram, already drawn

The free community academic vaults — bramses, PaperBell, zotlit — are genuinely good, and this kit does not out-feature them. The positioning is narrower and, I think, honest:

The free academic vaults give you a philosophy. ResearchVault gives you the wiring diagram.

What you are buying is curation and documentation — conventions decided once, five templates with locked field contracts, two documented import paths, a populated queue with a closed vocabulary, the eight-prompt triage pack, and the two worked examples above shown in full file contents.

The whole kit is 15 files. Setup is unzip, copy the five templates into your vault, and run one mkdir -p command from the Setup Guide that creates the seven-folder skeleton — Literature, Zettel, Projects/Thesis-MOC, Meetings, Templates, Prompts, Data — followed by a seven-checkbox first-run checklist. About fifteen minutes, and you are capturing papers the same day.

Disclosure: this post contains an affiliate link to my own Gumroad product; I may earn a commission if you buy.

Grab it here at the $1 intro price (live-verified 2026-10-21): ResearchVault — The PhD & Researcher Obsidian Vault — $1 (intro price)

Who This Is Not For

Three honest disqualifiers. If you have never used Obsidian and are not willing to spend fifteen minutes on setup, this is not for you — the kit is a wiring diagram, and a diagram to a vault you never open is a dollar wasted. If you want a plugin that writes the thesis, this is not that product and I do not sell that product; every claim that reaches your draft still has to survive your reading of its source. And if your lab already has a documented house convention, follow your lab's — importing a second convention mid-thesis costs more in context-switching than any template saves. (If you are a supervisor who wants a documented convention to hand to new students, that is a different conversation and the templates were built with you in mind.)

Before You Go

Everything you read above is reproducible from this post alone; the kit just saves you the assembly time. A few standing caveats: the worked examples use fake papers and a fake DOI on the reserved 10.5555 prefix; the license is personal use, with no resale or redistribution; and before anything is cited in submitted work, verify your institution's citation style, because the kit generates citekeys and front matter rather than knowing your department's rules.

If you are coming from Notion and want to see the zero-plugin end of this spectrum first, I wrote Notion Made You Quit: A 5-Minute Obsidian Starter Vault That Needs Zero Plugins. And on what starter vaults actually cost before this one existed, there is Obsidian Starter Vaults Cost $15 to $99. I Built a $1 One.

Questions about the pipeline, the utility, or whether the kit fits your subfield: nandminc2020@gmail.com. I answer my own email.