9.4 Very high
Tau-PET burden and cognitive decline in preclinical Alzheimer's disease
JAMA Neurology · Ossenkoppele, Hansson +9
Ranked from PubMed
The AI literature feed
The AI literature feed
that shows its working.
Paper Scout reads what's new on PubMed, ranks the handful of papers that actually matter to your research, and tells you — in plain English — why each one is in front of you. When it misses something, you can ask it what went wrong, and fix it.
One month free Set up in a couple of minutes No keyword alerts to babysit
9.2 Very high
Semaglutide and cardiovascular outcomes in obesity without diabetes
N Engl J Med · Lincoff, Brown-Frandsen, Nissen +21
Why it's relevant
Directly matches your interest in GLP-1 cardiometabolic outcome trials, and Nissen is on your followed-authors list.
What one run actually does
Thousands of new records in. A shortlist out.
Every stage has a published bound, and every time a bound is hit the run says so in its own journal. Nothing is quietly dropped to make the numbers look tidier.
STAGE 01
8
searches, written for you
A model turns your interests into PubMed queries — some tight on your niche, some deliberately wide, so adjacent work can still reach you.
STAGE 02
400
candidates gathered
Results are de-duplicated across all eight searches and capped. Hit the cap and the run records it, so a thin day is never mistaken for a quiet week.
STAGE 03
60
abstracts read in full
A fast pass over titles picks what deserves proper attention. Everything it drops is recorded by name, which is what makes the next section possible.
STAGE 04
0–10
scored, with a reason
The strongest model reads each abstract against your interests and scores it honestly — with a one-line reason, and a flag when something sits outside your core area.
How it works
Tell it what you work on. Once.
No boolean strings to maintain, no folder of seed papers, no checkbox taxonomy of specialties. You describe your research in your own words — or let it read your publications and write the first draft for you.
1
It reads your work
Give it your name or your lab page and it drafts your research interests from your own publications — as a paragraph of plain English you can read, disagree with and edit on the spot. Prefer to write it yourself? Do that instead; it's the same box.
2
It reads the literature
Each run searches PubMed across the window you care about, screens the titles, then reads the surviving abstracts in full and scores each one against your paragraph — not against a keyword list.
3
You read six papers
A short ranked feed, each entry carrying the reason it's there. Save it to your library, queue it as audio for the commute, share it with the lab, or dismiss it and tell the system why.
A reason, not a similarity score
Every paper arrives with its own argument for being there.
Recommenders hand you a ranked list and a thumbs-up button. You are asked to trust a number that nobody, including the people who built it, can explain for any individual paper.
Paper Scout gives you a sentence instead: what in your interests this paper touches, and why the model thought it worth your time. If the reason is wrong, you can see that it's wrong — which is the entire difference between a tool you can use and an oracle you have to believe.
It also flags the papers it thinks sit just outside your core area, rather than hiding them. The searches are written deliberately wide for exactly this reason: a feed that only ever returns your own niche back to you is a feed that will never tell you anything you didn't already know.
7.1 Adjacent
Spatial transcriptomics of the invasive margin in resected NSCLC
Nature Communications · Vasquez, Lindqvist +14
Why it's relevant
Not your core area, but the margin-annotation method is directly transferable to the immune-infiltrate work in your interests, and the cohort overlaps yours.
The part nobody else can do
Ask it why it missed something. It will tell you.
Every researcher's real objection to every recommender is the same: how do I know what it isn't showing me? Usually there is no answer, because there is nothing inside a similarity score to point at.
Paper Scout keeps a journal of every run — the searches it wrote, everything it retrieved, everything it screened out and everything it scored. Paste in a paper it should have found and it replays that run against it, then names the exact stage that failed.
Then it proposes a specific change to your interests, tests it, and lets you accept or reject it. Every edit lands in a version history you can roll back — whether it came from you, from a diagnosis, or from the profile it wrote on day one.
✓
Publication window — published inside the window this run covered
in_window
✓
Retrieval — returned by search 5 of 8, ranked 92nd of 400
retrieved
✗
Title screen — dropped here. The title reads as a methods paper, and your
interests don't mention assay development.
dropped_at_triage
—
Deep rank — never reached this stage
not_scored
Suggested change to your interests
Add: I also follow assay and method development where it applies to circulating tumour DNA.
Tested against your last four weeks: 6 of your 7 saved papers rank the same or higher. No dismissed paper rises.
Calibrate
It learns from you out in the open.
Most personalisation happens behind your back. You click, and somewhere a vector shifts, and the feed changes for reasons you will never be told.
Paper Scout does the opposite. Mark a handful of papers good or bad — it will pre-fill the list from what you've already saved, dismissed and listened to — and it proposes changes to your interests in words. Each suggestion is then replayed against papers whose verdict you already gave, so you can see what it would have done differently before you accept it.
The whole model of you stays in one editable paragraph. That's a deliberate constraint: the moment ranking depends on state you can't read, every explanation the product gives you starts quietly lying by omission.
Good
Single-cell atlas of alveolar macrophages in fibrotic lung
you listened to all of it
Good
Spatial transcriptomics of the tumour margin
you saved this
Bad
Murine model of bleomycin-induced pulmonary fibrosis
you dismissed this
Was I right?
CRISPR screen in organoid models of NSCLC
dropped at title screen
Proposed, and tested
I don't need animal-model studies unless they carry a human validation cohort.
4 of the 5 papers you marked bad fall below the fold. ⚠ One paper you marked good also falls — review it before applying.
Briefings
The week's papers, as a briefing you can listen to.
Ask for one whenever you want it and you get a few minutes of audio — not a robot reading abstracts one after another. Paper Scout groups what it found into two to four themes and tells you what actually moved: the through-line, what's genuinely new, and why this set of papers matters to the work you're doing right now.
It plays in the background with proper lock-screen controls, so it works on a commute, at the bench or between clinics. Any single paper can also be turned into its own spoken summary and dropped into the queue, and every briefing has a transcript if you'd rather read it.
Your reading list is the play queue. Save a paper during the day, and it's waiting for you in your ears on the way home.
01
Resistance mechanisms are converging — three papers, from three
groups, landing on the same escape pathway
3 papers
02
The screening cohorts finally have follow-up — and the effect
is smaller than the 2025 interim suggested
2 papers
03
One from next door worth your time — a methods advance in an
adjacent field that your assay could borrow
1 paper
Everything else in the app
Built to be trusted with your reading list.
Follow the people, not just the topic
Name the authors you never want to miss. Every run searches for them by name, and their papers are always read in full and marked in your feed — even when the title screen would have skipped them.
A library that's also a queue
Save anything worth coming back to. Search your own library by title or abstract, reorder it, and play it back as audio in the order you chose.
Version history for your interests
Every change to your profile is recorded with who made it — you, a diagnosis, a calibration, or the first automatic draft — and any one of them can be rolled back.
Listen to a single paper
Turn any paper into a spoken summary written for someone with your background — from the full text where it's open access, and from the abstract otherwise. Not a machine reading the PDF aloud.
Interests you write, in prose
Your profile is a paragraph, not a checkbox tree. Change a sentence, run it again, and watch what moves. It's the whole model, and it's yours to edit.
Share what you find
Send a paper and its reason to a colleague or a group chat in two taps. The reason travels with it, which is usually the part they wanted anyway.
On your phone
Made for the ten minutes you actually have.
A native app on iPhone and Android — not a newsletter you'll archive unread, and not a web dashboard you'll open twice.
Straight answers
What it doesn't do — yet, or ever.
You are being asked to trust a machine with your reading. Here is where it stops, in the same plain English as everything else.
It doesn't search outside PubMed
PubMed is a lot — tens of millions of citations, including the preprints it indexes. But if your field lives on arXiv, or on bioRxiv outside that index, Paper Scout cannot see it, and a diagnosis will say exactly that rather than pretend otherwise. Broader sources are on the roadmap.
Automatic morning delivery is still rolling out
Today you pull to refresh, a run starts, and a couple of minutes later the results are yours. Scheduled daily briefings and push alerts are being switched on account by account rather than promised to everyone at once — we'd rather turn them on late than promise a morning that doesn't arrive.
You can't subscribe on Android yet
Subscriptions are sold through Apple's App Store, so today they can only be started in the iPhone app. The Android app is real and the account is the same one, but there is no way to pay from it yet — rather than show an Android user a purchase button that cannot work, we say this instead.
It doesn't learn behind your back
Nothing you click silently reweights the ranking. Calibrate proposes changes in writing and waits for you to accept them. That's slower than a hidden model, and it's the only way the explanations stay true.
It doesn't invent papers, or fill space
Every entry is a real PubMed record with a real identifier. If a run fails you get an error and a retry button; if a refresh isn't due yet you're told when it is, and the feed you already have stays exactly as it was. What you never get is a stale list, a cached one, or a keyword score dressed up as a judgement.
There are limits, and here they are
One run an hour — five minutes if the last one failed, and immediately whenever you change your interests, because that's exactly when the old answer stopped being right. Ten miss diagnoses and six calibrations a day. Those exist to stop a runaway loop, and none of them are metered against your plan; the audio summary is the one thing that is, at ten a week on Reader and twenty on Unlimited. When you hit any limit the app tells you which one and when it lifts; it never quietly hands back the old feed as if it were new.
It's a first pass, not a systematic review
Paper Scout is built to tell you what's new and worth your attention. For an exhaustive, auditable search of the whole literature on a question, you still want a librarian and a protocol.
It won't read the paper for you
Summaries and briefings are written from abstracts and metadata. They're for deciding what to open — the judgement about what a paper actually shows is still yours to make.
Pricing
Start with a month free. Then £4.99.
The feed, the ranking and the reason on every paper are the product, so they are in both plans and neither is rationed. The only thing counted is the audio: how many papers you have read aloud to you.
Reader
One month free£4.99 / month
Free for the first month, then £4.99 unless you cancel. Apple allows one free month per Apple Account, ever.
- The whole ranked feed, refreshed as often as hourly
- A plain-English reason on every paper
- 10 papers read aloud a week
- The weekly briefing, as audio and a transcript
- Ask why a paper was missed, and fix it
- Library, queue, sharing and version history
For most people this is the whole product. Start here.
Unlimited
£9.99 / month
For people who listen more than they read. No introductory month — the free month is Reader's.
- The whole ranked feed, refreshed as often as hourly
- A plain-English reason on every paper
- 20 papers read aloud a week, subject to fair use
- The weekly briefing, as audio and a transcript
- Ask why a paper was missed, and fix it
- Library, queue, sharing and version history
The audio count is the only line that changes. Everything else is identical.
Subscriptions are bought in the iPhone app, through the App Store. Android and the web can't subscribe yet.
A week is a rolling seven days, not a calendar week: each summary comes back to you seven days after you used it, rather than everything resetting on a Monday. Cached audio replays free, and a summary that fails to generate is never counted.
Both plans renew monthly until you cancel, which you do in your Apple Account settings at least 24 hours before the period ends. If you stop paying, your feed freezes rather than disappears — every paper already in it stays readable, and nothing you saved is deleted.
Questions
Before you sign in.
How is this different from a PubMed saved-search alert?
A saved search matches strings. It returns everything containing your terms, in no particular order, and it silently misses anything phrased differently — which for a fast-moving area is most of it.
Paper Scout writes eight searches from your interests rather than one, deliberately casting wider than you would, then reads the abstracts and ranks them by judgement. You get a short ordered list with reasons instead of a long unordered one without. And when it gets it wrong, there's a stage name to point at and a fix to apply.
What does it actually search?
PubMed, live, through NCBI's E-utilities — tens of millions of citations, including the preprints PubMed indexes. Journal-level metadata comes from OpenAlex. Nothing is mirrored into a private corpus, so a paper indexed this morning can appear in a run this afternoon.
It does not currently search arXiv, or bioRxiv and medRxiv beyond what PubMed carries. If a diagnosis finds that a paper you cared about simply isn't in PubMed, it says exactly that.
Do I have to write my research interests myself?
No. Give it your name or your lab page and it finds your own publications and drafts the paragraph for you. Most people then spend two minutes correcting it, which is the point — you can see what it believes about you, so you can fix what it got wrong.
You can also just write the paragraph yourself if you'd rather, or if your current project has nothing to do with what you've published so far.
Does it learn what I like?
Only when you let it, and only in writing. Calibrate looks at what you saved, dismissed and listened through, proposes changes to your interests as sentences, and replays each one against papers you've already judged so you can see the effect before you accept it.
Nothing reweights the ranking silently. That is a deliberate design decision: if the ranking depended on hidden state, the explanation on every paper and every miss diagnosis would be incomplete, and we'd rather be slower than misleading.
How do I know it isn't missing things?
Test it. Take three papers from the last few months you'd have been embarrassed to miss, and paste them in. For each one Paper Scout replays the run that should have found it and tells you whether it was outside the window, never retrieved, dropped at the title screen, scored too low, or lost to a cap — and then suggests the edit that would have caught it.
No recommender built on similarity scores can answer that question, because there is no stage inside a cosine distance to point at.
Is my data used to train anyone's model?
Your research interests and the abstracts being ranked are sent to our AI providers for one reason: to produce your results. We don't sell your data, we don't share it with advertisers, and there are no advertising or tracking SDKs in the app.
Exactly who those providers are, what each one receives, how long anything is kept and how to delete it are all set out in the privacy policy — including the parts we're still improving.
Who is it for?
It's built for people whose career has a cost attached to missing a paper: clinician-researchers, principal investigators, postdocs writing grants, and anyone tracking a fast-moving area where "I hadn't seen that" is an expensive sentence.
It's tuned on biomedical research, because that's what PubMed covers. If you work outside the life sciences it is not the right tool yet.
How often can I refresh?
Once an hour, and five minutes after a run that failed. That isn't a paywall lever — it's that PubMed indexes roughly daily, so a second run twenty minutes after the first re-reads the same papers and re-pays for the same judgement. The app tells you when the next one is due rather than quietly returning the feed you already had.
Changing your research interests is the exception: that starts a new run straight away, because watching what your edit does is the whole point. None of this is metered against your plan — refreshing the feed is free however often you do it. Miss diagnoses are separately capped at ten a day and calibrations at six, which stops a runaway loop rather than selling you anything.
What does it cost?
Reader is £4.99 a month, and the first month is free. It includes the whole ranked feed, a reason on every paper, the weekly briefing, and ten papers read aloud to you a week. Unlimited is £9.99 for twenty a week, subject to fair use.
The audio is the only thing counted. The feed, the ranking, the reasons, the miss diagnosis and the briefing are in both plans and are not rationed — which is also why the two plans differ by exactly one line. A "week" is a rolling seven days rather than a calendar one, so each summary comes back to you seven days after you spent it.
Subscriptions are bought in the iPhone app through the App Store, so Android and the web can't subscribe yet. Cancel any time in your Apple Account settings, at least 24 hours before the period ends.
What happens if I stop paying?
Your feed freezes; it doesn't vanish. Every paper already in it stays readable, your library and reading list stay exactly as they were, and audio you've already generated still plays. What stops is the part that costs money to run: new runs, new summaries and new briefings.
Nothing is deleted for non-payment, and you don't need a subscription to get your data out.
Find out what you've been missing.
Two minutes to set up. Then bring it the three papers you wish you'd seen sooner, and see what it says.
One month free · Sign in with Apple or Google