New — reads any article, not just papers
Don’t just read it.
Know whether to believe it.
PaperGist reads the entire study — every figure, table, and statistic — grades the evidence on a rubric you can inspect, fact-checks its own claims against the source, and hands you something worth sharing.
Paste a DOI, drop a PDF, or link any article. No account · no card required.
Randomized controlled trial · Cardiology · 6 min read
The blood-pressure drug that worked — but not for everyone
A 4,800-patient trial found a 21% relative risk reduction — driven almost entirely by patients over 60.
Faithfulness check
“The drug cuts heart-attack risk by a fifth.”
Calibrated: cut relative risk 21% in this trial — with the benefit concentrated in patients over 60.
Share kit
Resolves full text from the places research actually lives
One run, everything
Drop in a study. Walk away with all of this.
Evidence grade
Not a black box — a transparent rubric. Design, power, controls, bias, conflicts, reproducibility: each criterion rated and explained, adapted to the field. The headline score is computed from the rubric, so it can never contradict it.
Catches its own overstatements
“Coffee prevents cancer.”
Linked to lower risk in this sample — not proven to prevent it.
A second, adversarial pass re-reads the source and rewrites any claim that says more than the data shows — automatically.
Expert mode
−1.6 kg · CI [−2.9, −0.3] · p=0.02 · n=212
Effect sizes, intervals, and p-values surfaced on every finding. Flip one switch between plain language and the hard numbers.
Any article, digested
News, essays, blogs — a visual digest with take-home points, a structured synopsis, and an honest read on the framing.
A share kit, fact-checked
Stat cards, reel scripts, an Instagram carousel with a real data slide, a LinkedIn post, and an X thread — every number checked against the source first.
Ten languages
Reads sources in any language, writes the digest in yours — right-to-left included.
01
The reading
It reads the actual paper. All of it.
Not a text scrape of the abstract. PaperGist ingests the real PDF — the figures, tables, and statistics where findings actually live. Hundred-page monsters get read in parallel parts. No open-access PDF? It races Unpaywall, Semantic Scholar, and Europe PMC, then falls back to the page’s own full text rather than giving up.
- Figures, tables & stats read from the source PDF
- 100+ page papers map-reduced in parallel
- DOI, arXiv, PubMed, publisher links — resolved automatically
page 47 of 112
Table 3 — primary outcomes
✓Every reported value captured verbatim
02
The grading
It grades the evidence — then corrects itself.
Peer-reviewed research found AI summaries are ~5× more likely than human ones to overgeneralize. So PaperGist doesn’t trust itself either: an adversarial second pass re-reads the source, flags every claim that says more than the data shows, and applies the calibrated wording automatically. You read the corrected version.
- Itemized rubric — the score is computed from it, never invented
- Association stays association; hedges stay hedged
- Corrections applied into the digest, with the audit trail visible
AI overgeneralization: Royal Society Open Science, 2025
Evidence grade
Reported effect
Faithfulness check
Overstated “Coffee prevents cancer.”
Calibrated: linked to lower risk in this sample — not proven to prevent it.
Why we built it
In an age of confident, AI-generated answers, the rarest thing is an honest “this evidence is weak.”
PaperGist is built to say it.
Who it’s for
Anyone who needs the truth of it, fast
Clinicians & researchers
Triage journal-club papers and judge evidence in minutes, not evenings.
Students & trainees
Understand primary literature — and learn what makes a study strong.
Journalists & creators
Report findings accurately, with the caveats most coverage drops.
Curious minds & leaders
Decide on real evidence — and spot the studies that don’t hold up.
Read the next one like an expert.
Drop in a study or this morning’s headline and see its data, its findings, and exactly how much to trust it.
Analyze something — freeNo account · no card · works on phones