Frequently Asked Questions

SummarizePaper helps you read, follow and use research with AI. For any paper from arXiv, a journal or a preprint server, you get a detailed summary, the key points, a lay summary and a blog-style article, which you can have rewritten in 16 languages and for four reading levels. An AI assistant answers your questions about one or several papers with the sentences of the papers that support its answers. Collections keep your papers together and compare them; a daily digest and alerts follow your field; and tools for researchers help you check a bibliography, review a question, prepare a journal club or a conference.

Every arXiv paper, and papers from many other sources:

• Journal articles: any paper with a DOI. Paste the DOI or the link of the article, or search the open-access literature from the home page (we use OpenAlex, which lists more than 250 million works).
• bioRxiv and medRxiv preprints.
• HAL, ChemRxiv, the OSF preprint servers (PsyArXiv, SocArXiv, EarthArXiv, EdArXiv...), and the articles of PubMed and Europe PMC: paste their link, their PMID or their PMCID.

When a journal article also has an arXiv version and its own license does not let us read it, we take you to the arXiv page. You cannot upload your own PDF to have it summarized.

Search for a paper with the search bar on the home page (on arXiv, or in the journals and preprints), or paste its arXiv ID, its DOI or its link. If the paper has already been summarized, the summary shows up immediately. If it is the first time someone looks for it, our AI has to read and summarize it first, which can take a few minutes; the progress is shown on the page. You can open the PDF of the paper beside its summary to compare them.

A summary is a derivative work, so our AI reads the full text of a paper only when its license allows it: CC BY, CC BY-SA, CC BY-NC, CC BY-NC-SA or CC0 (public domain). For the other papers (arXiv's default license, “no derivatives” licenses, articles behind a paywall), the summary and the assistant use the title, the authors and the abstract, and the page says so. Many bioRxiv preprints, for instance, are under CC BY-NC-ND.

We read the full text of the paper and send it to recent OpenAI language models with a series of prompts that produce each part of the summary. For arXiv, we read the HTML version when it exists and the PDF otherwise; for journals and other preprint servers, the structured text of the publisher or of Europe PMC when there is one, an open PDF otherwise, or the arXiv version. Free users get our standard model; Premium subscribers get our most capable model, for summaries as well as for answers.

Yes. On the page of a paper, ask for a version in one of 16 languages (English, French, Spanish, German, Chinese, Japanese, Arabic, Hindi...) and for one of four readers: a 12-year-old, a high school student, an undergraduate or an expert. Each version is written once from the summary and the abstract, then shared by every reader. Versions already written are free to read; you can ask for 3 new ones per day without Premium, as many as you want with Premium.

Yes, when the license of the paper allows it (see above); otherwise it answers from the abstract and tells you so. When a paper is too long, the assistant leaves out its references and appendices first, then keeps the passages that matter most for your question, and each paper says how much of it is read. Its answers appear as they are written: you can stop, copy or ask again.

Each claim of an answer comes with the sentence of the paper that supports it, and we check that this sentence is word for word in the paper. A claim whose sentence is not found, or a number no paper states, is marked. When several papers are discussed, tags such as [Smith et al. 2021] say which one each claim comes from and link to it. The assistant can still be wrong: read the passages before you rely on an answer.

Yes, on the chat page: search arXiv, the journals and the preprint servers in one list, and add up to 10 papers. You can reorder them, which orders the answers. With Premium, a comparison table reads each paper whole and gives each value with the passage of the paper it comes from.

Your recent conversations are kept: you can come back to them, rename them and delete them, and export an answer with the BibTeX of its papers. A conversation can be shared as a public, read-only page of what it was when shared, which never says who asked. The link works for 90 days after it was last shared, and you can stop sharing it at any time; deleting the conversation stops it too.

A collection keeps papers together for a project: add them from any paper page, or import a BibTeX or RIS file or a Zotero collection. Each paper can have notes, tags and a reading status, and shows its year, journal, editorial status, citations and newer version. A preprint and its published version count as one work. You can chat with a whole collection, order its papers and export them (BibTeX, RIS, Markdown or a spreadsheet) with your notes and tags. A free account has 3 collections of 20 papers, Premium 100 collections of 200 papers.

With Premium, a comparison table puts the papers of a collection side by side, in columns from a template (general, clinical, machine learning, observation, theory) or of your own. Each value keeps the passage of the paper that supports it, quoted word for word, and you can correct a value yourself. The analysis then lists where the papers agree, disagree and what they leave open, with the passages behind each point. It can also find what the collection may be missing: references several of its papers share and the closest papers it does not hold.

Yes. A read-only link shows the collection to anyone, who can copy it into their own account. An invitation link lets colleagues edit it with you: they add, remove, annotate and order papers, and run the analyses on their own plan. Only you can rename, share, watch or delete it, and you can stop either link at any time.

With Premium, a watched collection follows its papers: you get an email when they are cited, when one gets a new version, and when one is published in a journal, retracted or corrected.

On the digest page, choose up to 5 arXiv categories, bioRxiv and medRxiv subject areas or journal subfields, which cover every journal and other preprint servers such as ChemRxiv, OSF or SSRN (or import them from your ORCID iD or arXiv author identifier), optional keywords, the kind of summaries (for researchers: contribution, method and main result; or 2 sentences in plain language), their language and the time you want to receive it. While you edit, the page shows what you would have received from the latest announcement, and you can send yourself a preview.

Every weekday, you get up to 10 new papers, with a link to the full AI summary of each and to a web page where every paper of the day is ranked. No digest is sent on Saturday and Sunday in your time zone. You can also receive it once a week instead. Every email has a link to unsubscribe in one click.

With Premium, the digest ranks papers by closeness to a description of your research or to reference papers (by arXiv ID or DOI), and can hold every paper close enough rather than a fixed number. It follows authors across all categories, adds a synthesis of the week in your fields, has the full summary of your 3 best papers ready before you open the email, and can be listened to as audio. It can also gather your scoop, citation, new version and publication alerts instead of separate emails. You can have up to 5 digests, with 15 categories and 25 papers each.

• Saved searches: every new paper that matches a search, by email, in the digest or by RSS (one watch with a free account, five with Premium).
• Citation alerts (Premium): who cites your papers or the papers you follow and what they say about them, what changed when one gets a new version, and when one is published, retracted or corrected.
• Scoop watch (Premium): an email the day a new paper comes close to your project in progress.
• Weekly letters: what happens in other fields, an article per theme on the papers that matter beyond it, on the web and by RSS for everyone, one by email with a free account and all of them with Premium.

Yes. Only you see it. It is used to compare your project with the papers: an AI rewrites it as an English abstract and splits it into facets, and they are sent to our AI provider for that comparison only. If you give the PDF of a draft, it is read once to write the description and is not stored, unless you ask us to keep its text, stored encrypted, to compare it with the full text of close papers; you can delete that text at any time. You can delete a project at any time.

Besides the summaries, the assistant, collections and the digest:

• To follow your field: citation alerts, the scoop watch, saved searches, weekly letters, the rising topics of arXiv and a map of the groups that work on a topic.
• To write and submit: a literature review from a research question, “What did I miss?” for the close papers your draft does not cite, a bibliography check, and exports to Zotero, Notion and Obsidian.
• To present and meet: journal club slides, press kits, a conference companion and research placement leads for students.

See all the tools and what each offer includes.

From a research question, it finds 15 to 30 founding and recent papers, groups them by approach and writes a synthesis that cites them (Premium+). It is an exploratory review to start reading a question, not a systematic review: check the papers it cites and complete them with your own search.

Give your .bib file (or a Zotero collection), and your manuscript if you want: its LaTeX, your Overleaf project as a .zip, its arXiv identifier, or a PDF. We identify each reference with Crossref, NASA ADS, OpenAlex, arXiv and Semantic Scholar: retracted papers, expressions of concern and corrections come from Retraction Watch and Crossref, and preprints published since (arXiv, bioRxiv, medRxiv...) get their corrected BibTeX entry; you choose which corrections to download. With the LaTeX, an AI reads each sentence with citations next to the abstracts of all the papers it cites, then reads again the sentences flagged against the full text of the papers when their license allows it, and shows the passage of the paper that supports or contradicts them. A flagged citation is something to check, not a proof of error. You can mark each problem, check again after correcting, and have your references watched until your paper is published. Your files are not stored.

Open the page of the paper and, in its tools, click “Slides to present it” (Premium+). Choose the language (16 languages), the audience (researchers of the field, students, researchers of another field) and the length of the talk (10, 20 or 45 minutes): in a minute or two you get the slides with what to say on each. Each point comes with the passage of the paper that supports it, checked word for word; a point the paper does not support, or a number it does not state, is marked for you to check. The figures of the paper are included when its license allows it (CC BY and similar).

The slides also show the earlier work the paper builds on and the papers that cite it (from its references and citations, not from the memory of an AI), where the paper stands (publication, retraction or correction notices, a newer arXiv version), its strengths, the limits its authors state, points to discuss marked as an AI reading, and the hard questions the audience is likely to ask with what the paper lets you answer.

Edit the text and speaker notes, move, duplicate, add or remove slides, pick another figure of the paper, ask for a slide to be rewritten shorter, simpler or more technical, and save a private version. Present it full screen with a presenter view (notes, next slide, timer), share it with the participants by a read-only link, save it as PDF (16:9, one slide per page, no notes) or download the PowerPoint (.pptx) with the passages in the speaker notes.

For a team: “Propose for the journal club” adds a paper to a collection you share with your colleagues; the collection then prepares the slides of its selected papers together, with how they compare.

Yes. Each paper page lets you download the summary as a PDF, and export the paper to BibTeX, RIS (Zotero, Mendeley, EndNote) or Markdown (Notion, Obsidian), whether it comes from arXiv, a journal or a preprint server. With Premium, you can also send it to your Zotero library with its summary.

Yes. Reading summaries, asking questions, collections, the daily digest, saved searches and the weekly letters are free, within the limits below. An optional Premium subscription gives you our most capable model, higher limits and the tools that follow your field and check your work; Premium+ adds the tools that read the most for you.

Reading papers that are already summarized is unlimited. Starting new summaries is limited to 10 per day for free users and unlimited with Premium.

Questions to the AI assistant are limited to 30 per day without an account, 100 per day with a free account and 500 per day with Premium.

Premium gives you our most capable AI model for summaries and answers, no daily summary limit, 500 questions per day to the assistant, unlimited rewrites, larger collections with their comparison tables and open questions, a smarter digest, citation and version alerts, the scoop watch, “Who works on it?”, “What did I miss?” and the bibliography check. It is billed monthly or yearly through Stripe; we never see your card details. You can cancel in one click with the "Manage my subscription" button on the Premium page: Premium then stays active until the end of the period you paid for.

Premium+ costs twice Premium and adds the tools that read the most for each use: the literature review from a research question, journal club slides, press kits, the conference companion and research placement leads. Everything else is in Premium. A Premium subscriber can move to Premium+ at any time on the Premium page and pays only the difference for the current period.

No account is needed to read summaries or ask a few questions. A free account is needed for collections, the daily digest, saved searches, Premium and more questions per day. You can sign in with Google, or receive a login link by email: no password to remember.

This website is developed and maintained by a researcher in astrophysics (Quentin Kral) who wants to allow everyone to access research results.

In bibtex you can use the following snippet

@software{SummarizePaper,
  author = {{CarbonFreeConf.com}},
  title = { SummarizePaper: AI Powered Article Summarization and Virtual Assistant },
  url = { https://summarizepaper.com },
  year = { 2023 },
  date = { 2023-03-30 },
}

In other cases, anything which includes our summarizepaper.com URL is OK. For example: "CarbonFreeConf.com; SummarizePaper: AI Powered Article Summarization and Virtual Assistant; SummarizePaper.com accessed xxxx/xx/xx"

Yes, the code is open source and freely accessible. You can join us and help to make the tool even better.