How to Read Papers With ADHD: a Three-Tool Stack That Actually Works

You have 800 papers in your reference manager. You downloaded each one with intent. You have read maybe one in twelve. The rest sit there as a slowly compounding source of guilt while you refresh email instead.

If you have ADHD, or attention that behaves like ADHD under workload, this pattern is universal. We hear it from PhD students, postdocs, and PIs alike. The problem is rarely willpower. The problem is friction: the PDF format is dense and static, methods sections eat working memory, figure legends are where the dopamine drops out, and the act of starting to read is more expensive than any single inbox refresh.

This post is a working stack for reading papers with ADHD. It is three tools, one for each layer of the job. Each tool is best in its layer, which beats any single all-in-one app at the task that matters most: actually finishing the paper.

Why scientific reading specifically breaks down with ADHD

Generic ADHD productivity advice misses the texture of what fails when an ADHD brain meets a research paper. The failures are specific and they compound across a reading session until the paper is closed without finishing.

The PDF opens and the inbox wins. The decision cost of "start reading" is high; the dopamine of an inbox refresh is instant. The paper loses every time. An entire afternoon disappears to this loop without a sentence read.

The abstract is read three times and methods are never reached. Re-reading the same paragraph is not poor comprehension. It is the working-memory equivalent of trying to hold water in a colander. By the third sentence, the first has fallen out, and re-reading feels like progress when it is actually a stall.

Figure legends are where the dopamine collapses. A panel figure with eight sub-plots and a 300-word legend is the natural breaking point. You skim, tell yourself you will come back, and you never do. The paper goes into the unread pile with a "saved" label that does no work.

The PDF format itself is hostile. PDFs do not reflow on mobile. They are not searchable across your library by default. They cannot read themselves aloud. The technology of the academic paper is optimised for print, and the ADHD brain is the worst possible customer for print-shaped attention.

The stack below addresses each of these failure modes directly. It will not turn you into someone who reads a paper a day in flow. It will turn you into someone who finishes papers, which is the metric that matters.

The three-layer stack for reading papers with ADHD

The architecture has three layers, and each tool is the best in its layer. The cost of switching tools at a layer boundary is much lower than the cost of running one mediocre tool for everything.

Layer 1: Zotero as the library

Zotero is the floor of the stack and the only non-negotiable tool. It is free, open source, and runs locally with optional cloud sync. The browser connector is one click on any paper page, and it captures the citation metadata, the PDF, and the abstract together.

Two organisational axes do the work, and the distinction matters. Collections are folders: hierarchical and exclusive, so a paper sits in one parent collection at a time. Tags are flat and additive: a paper can carry "spatial-transcriptomics", "method", "to-read-deep", and "shared-with-marc" simultaneously. Most researchers use only collections, which forces a choice between mutually valid groupings and produces the "where did I file this" failure.

A workable Zotero setup is collections for projects (one collection per manuscript or grant) and tags for everything else (topic, status, collaborator, follow-up question). Tags are searchable in two keystrokes. A tag you will apply in three seconds beats a folder structure you will not maintain.

Two plugins are worth installing by default. ZotFile auto-renames PDFs to a consistent format and moves them to a folder you control, which matters because Zotero's default storage is opaque. Better BibTeX generates stable citation keys, which matters the moment you write in LaTeX or share a .bib file with a co-author.

Layer 2: Readwise Reader as the reading surface

Once a paper is in Zotero, the question is where you actually read it. The answer is not Zotero's built-in PDF viewer, which is fine but offers nothing for ADHD specifically. The answer is Readwise Reader, which costs about £95 per year and is the single highest-value subscription a researcher with ADHD can hold.

Three Reader features change reading materially.

Text-to-speech that does not sound like a 2008 satnav. Reader's TTS is good enough to listen to papers on a walk to the lab or at the gym. This is the biggest unlock. Reading is no longer something you have to sit at a desk to do, and the ADHD brain accepts auditory input in contexts where it rejects visual input on a screen. TTS converts "I cannot make myself sit and read" into "I am reading while doing something else".

Ghostreader, the in-built AI assistant. Highlight a methods paragraph and ask "what reagent concentrations did they use" and Ghostreader returns an answer grounded in the paper, with source sentences quoted. Ask for a section summary. Ask "what is the comparison condition in Figure 3" without scrolling. The answers are bounded by the paper text, so the failure mode is "I don't know" rather than hallucination.

Four-colour highlights with spaced-repetition review. Reader supports yellow (claim), red (problem), blue (method), and orange (follow up). Highlights flow back to a daily Readwise review showing ten of them with morning coffee. Six months later you still encounter what you read, in tiny doses, without re-opening the paper.

Reader handles PDFs, web articles, EPUBs, YouTube transcripts, and emails in the same surface. For papers specifically, import is one click from a Zotero attachment or a direct URL.

Layer 3: NotebookLM for audio triage

NotebookLM is free, made by Google, and the audio overview format is the cheating move for ADHD paper triage. Upload a paper, click generate, and within about five minutes you get a 10 to 25 minute conversation between two AI hosts discussing the paper.

Conversational format works on ADHD attention where prose does not, because dialogue carries built-in turn-taking and salience cues that single-narrator audio lacks. The hosts disagree, ask each other questions, and signpost when something is important. Papers can be absorbed on a bus that would never have been finished as PDFs.

Audio overviews do not replace reading the paper. They are a triage tool, not a comprehension shortcut. If a paper still sounds relevant after the audio, it goes into Reader for a deep read. If it does not, tag it "low-priority" in Zotero and move on without guilt.

NotebookLM also supports grounded questions across multiple uploaded sources, which is useful when comparing three competing methods papers without holding all three in working memory at once.

The workflow, in steps you can copy

The flow is one-directional and lossless. Zotero is the source of truth for what you own. Reader is where you read. NotebookLM is where you triage. Nothing lives in two places, and nothing requires you to remember where you put it.

Why the all-in-ones do not work

The all-in-one tools promise to do all three layers in one app. We have tested them. The promise does not hold for ADHD reading.

The pattern is identical across the four. One-app solutions optimise for "everything in one place", which sounds like an ADHD win and is actually a loss, because the all-in-one is mediocre at the reading layer, which is the layer where the ADHD brain breaks. You do not need fewer apps. You need the right tool at the layer where you fail.

What the stack does not fix

It does not give you the executive function to start. That problem is real and outside the scope of any reading workflow.

It does not turn boring papers into interesting ones. A poorly written methods section is still poorly written in audio.

It does not eliminate the unread pile. The pile gets smaller and less shameful, but it does not vanish. Trying to drive it to zero is itself an ADHD failure mode. Aim for circulation, not completion.

It does not absolve you of taking notes. The Zotero one-line note is non-negotiable. Reader highlights only work if you actually highlight. A passive setup with no engagement output produces nothing.

The stack is enabling, not curing. What it does is remove the friction that compounds with ADHD until "read this paper" becomes a six-step decision tree the brain refuses to enter. Cut the friction, and reading becomes default.

The Bottom Line

The reading problem and the discoverability problem are two halves of the same picture. This post addressed the reading half: how to consume papers when attention fights you.

The other half is what we work on at Academic SEO: making sure the paper you publish, which took two years and seventeen revisions, is found and read by the people you wrote it for. Most papers are not under-read because they are bad. They are under-read because they are not discoverable. The most common pattern we see is papers with views but no citations, which signals a clarity-and-positioning problem upstream from any reading workflow. The more direct version is discoverability as a funding requirement, which is where most funders are now moving.

Reading better and being read better are two sides of the same coin. Both deserve a system around them.