Papers and verification

Open any paper from your Library — as a drawer beside the list, or as a full page — to see everything CoChat knows about it.

At the top: the title, authors, year and venue, a DOI link, an Open-access PDF link where one exists, and an Ask button that starts a chat about the paper. A strip of four cells summarises Claims (verified / total), Cites (flags in the reference list), Questions (how many it fits) and Full text (indexed, PDF, or none). Click a cell to jump to that section. While CoChat is still pulling the paper’s citations in the background, the Cites cell says Checking… instead of a final answer.

Is it real? — verification

Every paper with a DOI is checked against the publisher registries (CrossRef and DataCite). The result is a coloured badge:

Badge	Meaning
Verified (green)	The DOI resolves and the title and year match the registry record.
Verified · note (green)	A real paper, but one field differs slightly — a truncated title, or an online-first year. The badge shows which.
Checking… (blue)	Not checked yet.
Unverified (amber)	The check couldn’t complete — a timeout or a rate limit. It says nothing about the paper; it’ll be retried.
Disputed (red)	The DOI resolves, but to a clearly different paper. Don’t cite it.
Not found (red)	No registry has any record of it. This is the sign of a fabricated citation.
Your document (grey)	You uploaded it, so there’s no registry record to compare against.

Two things worth knowing: Unverified is not the same as fake — it only means the check hasn’t settled. And because the DOI is what’s looked up, the title is the real test: a DOI that points at a different title is disputed, not verified.

Click the badge for the registry consulted, when it was checked, and any field-by-field differences. What each check can and cannot see is in How we check papers.

A preprint whose registry record names its published version shows Preprint — published as and that version’s DOI under the authors (on the public report, under the DOI); the published paper shows Published version of a preprint. The two versions can differ, so check the published one before relying on the preprint’s claims or verdicts. A paper flagged as a preprint that has no published version yet is re-checked about every two weeks, so the link can appear after publication.

Overview

• Snapshot — an AI-written summary (TL;DR, findings, methods, limitations), written when the paper is verified (see Verify paper below). It’s written from the paper’s full text; a paper with only an abstract gets no snapshot.
• The abstract, topics, and whether the paper is open access or has its full text indexed.

Verification

Verify paper — at the top of the tab, and in the paper’s header — verifies the paper in one step: its claims, the Checkup, the Stats check, an independent check of the findings, whether each claim’s cited sources say what it relies on, and the snapshot. Most papers in the Library are already verified in the background, and opening a verified paper, or one being verified, is free. When the paper still needs the work, a click uses one of your paper verifications. Free: 1 per day (UTC), up to 5 per month. Starter: 50 per month. Pro: 300 per month. Starter and Pro have no daily limit. The line under the button says how many you have left this month, and on Free how many today. When today’s is used, a new one is available after midnight UTC. Asking again for the same paper never uses a second one while its text stays the same (a new version of its full text can count again), and a verification that couldn’t run is not counted. While it runs, the tab shows each step as it finishes — you can leave and come back. A paper with only its abstract available can’t be verified. Checking that the paper and the works it cites exist (the badge above, and the Citations tab) happens on its own and has no button.

Once a paper is verified, the button becomes Re-verify next to “Verified <date>”. CoChat decides for you whether there is anything to redo — a step stuck behind an older check, one that failed, or one whose result predates a new upload of the paper’s text — and says “A newer check is available” when it thinks there is. Clicking always checks first: if nothing has actually changed, nothing runs and none of your paper verifications is used — the paper just stays marked Verified. Only genuinely new work uses one, under the same free-while-nothing-changed and never-charged-twice rules as the first check.

Below it, four sections, one at a time. The row at the top switches between them and shows each one’s status: how many claims and how many you haven’t read, and whether the Checkup or Stats check found anything.

Claims this paper makes. Verifying the paper reads all of it and lists its claims, each anchored to a verbatim quote. The claims are one list, in the same order as on the paper’s public page: first the claims whose cited sources dispute them or are misattributed, then the claims whose cited sources were checked, then the paper’s own results, then the rest. Claims of lower confidence (not yet classified, or not a substantial claim) come last. The list shows five claims at a time; Load more shows the next five. Filter by kind (findings, interpretations, …) at the top.

Each row shows only the claim and one line about it: its kind, how strongly it is worded, Verified by N when N of the sources it cites confirm it (Disputed by N when a source doesn’t say it), and May combine two claims when it looks compound. A confirmed verdict from the current checker on a source the paper cites was measured right 95% of the time; the other verdicts are experimental until a blind sample measures how often each kind is right (how we check papers). A dot marks a claim you haven’t read yet; opening it marks it read, for you only. An open claim shows its quote, the labels (basis, strength, causal wording, centrality — the last two are estimates), Is this a real claim of the paper? with Yes / No, Show in PDF, and Details: every place the paper states it, the sources it cites with their verdicts, and any possible issues. The paper’s Assumptions are the last group — what its conclusions rest on without data, a citation or an argument; they’re not counted as claims.

Claims about this paper. Add your own with Add claim — “CA1-specific NMDAR knockout impairs spatial memory” — then open it and click Verify to check it against the paper. Verdicts are confirmed, partly supported (the source backs part of the claim, not all of it), disputed, misattributed or unverifiable. A check the verifier isn’t confident about reads as unverifiable, never as disputed. Until that kind of verdict has a measured accuracy, the verdict is labelled Experimental — accuracy not yet measured. Anyone else’s claim and a literature review’s claims show their verdicts too, but aren’t offered a check here. Claim verification has a daily allowance on Free and Starter plans.

Checkup checks the paper against itself: does each reported number match what can be recomputed from the paper’s own data, and does the text agree with itself? A finding only appears when the quote and the arithmetic both check out. The list shows one line per issue with its severity; open one for the whole explanation, its quotes and Show in PDF. An empty result is a good result. You can add a Note, Dismiss or Confirm each finding.

Stats check recomputes the test results the paper reports in APA style and lists the p-values that don’t match, in the same one-line-per-issue list.

Citations

A paper’s three citation lists, one at a time. The row at the top switches between them and shows each list’s count. The paper header shows how often the paper was cited (Cited by).

Every work in a list, like every paper CoChat cites, shows up to ten author names as in a reference list, then et al. when it has more, then the year, venue and DOI.

Cites — what the paper cites, with retracted references first and the number of retracted references above the list. If the list is empty, the reason is spelled out — the paper has no DOI, the publisher hasn’t deposited its reference list, or the lists haven’t been loaded yet. While the lists still load, an empty Cites or Cited by list says Checking… and fills in on its own — there is nothing to click.

Each reference in Cites has a dot from CoChat’s own check of the reference as the paper prints it. Verified (green): its DOI or PMID names the printed work. Verified · note: a small difference, such as a shortened title. Citation error (amber): the citation as printed is wrong — its DOI and PMID name two different works, its DOI is not registered (a DOI printed with typographic dashes counts as the same DOI with hyphens), its year matches no date of the record (its print date, its online-first date, or for an arXiv paper any version from the first on), or its identifier points to a different work and no index we searched (OpenAlex, Crossref, Semantic Scholar, PubMed) has its title. A reader can’t verify such a source, so fix it before you rely on the claim it supports. When a web search finds that title, it is still a citation error, and the dot says only that the identifier points to a different work. Hover the dot for the reason and the indexes searched. A reference CoChat couldn’t check has no dot.

Cited by — who cites this paper, and how: the most-cited citing papers. When the list holds fewer papers than the paper’s citation count, the row at the top says how many it lists, for example 7 of 978. Where a citing paper has been read, labels show how it uses this paper: Background, Method or Result, and Influential. Hover a label to see what it means.

Related — works related to this paper.

Question fit

One card per research question the paper belongs to: the question, a relevance bar, the triage status, and where the paper came from. The relevance rating is automatic, but you can override it with Low / Medium / High (choose Auto to hand it back). From here you can attach the paper to another question or create a new one.

PDF

Your uploaded PDF, if there is one. Otherwise CoChat shows the open-access copy, with a link to the original. If the publisher blocks access, you’ll be told; if there’s no copy at all, upload one to a project and it appears here.

Metadata

The bibliographic record — authors, venue, year, DOI, arXiv id, Cited by and References counts, type, and dates — plus Referenced in: every chat and literature review table of yours that cites this paper, with links back.

Check a paper without an account

Papers lists the public papers, most cited first, and the papers checked last. Each field has its own page, such as Medicine, and each page of a list has its own link. The search box finds papers that have been checked.

Paper verification runs the same checks without signing in:

• By DOI or arXiv id. The report is public, at /papers/<doi>.
• Your own PDF, such as a draft or a preprint. Choose Or check your own PDF, privately, confirm you have the right to upload it, and select Verify my PDF. The report is private:
  • It opens from a link kept on the device you uploaded from.
  • Copy private link gives you a link that opens it anywhere. Anyone holding that link can read the report.
  • It is never listed publicly.
  • How long CoChat keeps an upload follows the notice shown under the file picker when you uploaded it. The current notice says CoChat keeps your PDF and its analysis for one year to improve the product, then deletes them. They never appear in anyone’s Library or on a public page, and the report stays readable from your link until then. An upload sent from the earlier form, whose notice promised deletion after 30 days (the form in use until late September 2026), keeps that promise: its PDF was deleted once read, and its text and report after 30 days. Either way, the report shows the date it will be deleted.
  • If the PDF matches a published paper, the report links to that paper’s public report.
  • Numbered citations such as “[3]” can only be checked when the PDF matches a published paper, since the lookup uses that paper’s reference list.

Both kinds of check count toward the same few free checks a day. The maximum PDF size is 25 MB.

A paper’s public report shows every claim, with or without an account. A report on a PDF you upload shows its first five claims without an account, starting with those whose cited sources were checked. Create a free account to see every claim and every source verdict.