PaperTok

Privacy policy

Last updated 15 November 2026.

PaperTok is an application for discovering and reading scientific research. This page describes what data is collected, why, where it is kept and who it is shared with. It is written from what the code actually does, not from a template.

In short

Who is responsible

PaperTok is a personal project, not a company. The data controller is its author, and the contact for anything to do with this data is nicomg60@gmail.com.

You can use PaperTok without an account

The paper feed, search and the public pages work without signing in. Without an account no account profile is created. Usage analytics can still connect visits from the same browser, as described under Analytics.

What is collected if you create an account

Sign-in is with Google or GitHub. PaperTok never sees or stores your password: Firebase Authentication handles the sign-in and only the result arrives here.

DataWhere it comes fromWhat it is for
User identifier, email, name and profile picture Google or GitHub, when you sign in Recognising you between sessions and showing your name and photo in your account menu
Public name, handle, bio, photo and university, and the people you followed, from earlier versions of PaperTok You added them in an earlier version Kept, but no longer shown to anyone
Lists, saved papers and highlights Your use of the app So your library is still there next time
Plain-words summaries, each with the paper's title, identifier and PDF address, the reading level, the model that wrote it and the date The AI model, when you read a paper in plain words Reopening a paper without writing its summary again, and the list in Settings → Reading & AI
Authors, institutions, topics and journals you follow Your use of the app Building your feed and, if you turn them on, your email updates, which leave out journals
Feeds you create: each feed's name and the areas, topics, authors, journals and keywords you put in it You choose them Showing each feed as a tab after For you, Most recent and Following, and fetching each feed's papers. They are kept with your account and deleted with it
If you connect Zotero: an encrypted access key to your Zotero library, your Zotero user number and username, and, for each paper you send, the keys and versions of its Zotero item and note Zotero, when you connect it Sending a paper and your highlights to your Zotero library when you ask, and updating them instead of adding copies
Comments and content reports from earlier versions of PaperTok You wrote them in an earlier version Kept, but no longer shown to anyone
Interactions with papers (likes, skips, how long each card stayed on screen, reads), a list of the last 500 papers you viewed, a list of the last 1,200 highly cited papers the feed offered you that you viewed, counts of the feed's highly cited and recent papers you were shown and of the likes, saves and PDF opens they got, a number for how much you lean to recent or older papers, and the last 200 papers whose like, save or PDF open moved those counts or that number Your use of the app Not showing you the same paper twice, ordering the feed, and choosing how many highly cited papers it shows
Interests you chose, the subjects you removed and the new ones you added after signing up You choose them Building your feed: a subject you removed stays out of it, except for papers from what you follow and papers related to one you like, and a new subject you added gets a head start until the app learns it, while one you removed and chose again comes back with what the app had learned
Language, theme and notification preferences You choose them So the app behaves the same on every device
The latest version of this policy whose change announcement you were shown Recorded when the announcement opens So the same announcement does not come back on another device; it records that the announcement was shown, not that you accepted anything

This data is stored in Google Firestore. The legal basis is performing the service you ask for when you create the account: without it there is no library and no personalised feed.

A plain-words summary is saved only once it is finished, one per paper and reading level. Only your account can read your summaries. Write it again replaces the saved summary when the new one finishes. Summaries are kept until you delete your account.

A paper counts as viewed when its card stays on your screen for at least one second or when you skip it. Besides your interactions with each paper, your account keeps a list of the last 500 papers you viewed, so that the feed on the other devices where you sign in does not show them again. Once the list is full, each new view pushes the oldest one off the list. A second list keeps the last 1,200 highly cited papers the feed offered you that you viewed, so that it does not offer them again. Only your account can read these lists.

Your account also counts, for the highly cited papers and for the recent papers in your feed, how many you were shown and how many likes, saves and PDF opens they got. A PDF open counts only when the paper opens in PaperTok's PDF reader, not when it opens on another site, such as the publisher's or your library's. Skipping a paper or marking it Not interested adds no like, save or open. Once 2,000 papers shown are counted, every count is halved, so your recent use weighs as much as all the use before it. Your feed keeps between one place in fifteen and one in three for a highly cited paper you have not seen. Each page picks its share at random, weighted by those counts; when the feed has no such paper to put in a place, the next paper takes it. The feed receives at most five highly cited papers at a time, so a full page holds at most about one in five on the first page and one in seven on later pages. Your account also keeps one number between −1 and 1 that likes, saves and long reads of very recent or older papers move, and that orders the feed toward recent or older papers. A long read of such a paper moves it every time, even when you read that paper again. Your account keeps the last 200 papers whose like, save or PDF open moved the counts or that number, so that a paper's like, save or PDF open counts once, and an unlike or unsave takes back what the like or save added. Only your account can read them. None of this is stored for a visitor without an account. A tab or app on any device still running an older version of PaperTok does not keep the counts, the number or the 200 papers. Using it saves your interactions without them, so they can start again from zero. Reloading or closing it ends this. Otherwise it looks for the new version every few minutes while on screen, and switches when you next open another page.

If you link a university, Firebase emails a verification link to your university address. This browser temporarily stores that address and the pending request, scoped to your PaperTok account. The request expires after one hour and is cleared on its next use. Cancellation, account changes and completion also clear it. An address given on the welcome, before you have an account, waits in this browser with your other welcome answers. It is sent once your account is created and cleared then, or when you log out first. Your profile stores the verified institution, not the secondary address. A temporary Firebase proof account is deleted after verification where possible; an existing account is preserved.

What is public and what is not

By default your activity is private. The only thing you can make public now is a list, by publishing it. A published list can be reached by anyone with the link, search engines included. You can stop sharing it at any time from My lists; from then on it is no longer served, although a search engine may take a while to update its copy.

Comments and public profiles from earlier versions of PaperTok are no longer shown to anyone, although a search engine may take a while to drop a copy it made earlier.

University email verification

When you request a university link, PaperTok sends that email address to Firebase Authentication to generate a one-time verification link and to Resend to deliver it. The pending request stays in your browser for up to one hour. Cloudflare stores hashed account and recipient identifiers for request limits and a delivery status; it does not store the verification link or message body. Your linked institution records the university and email domain, not the full address. Requesting verification does not subscribe you to email updates.

Zotero

Connecting Zotero takes you to zotero.org, where you allow PaperTok to read and write your personal Zotero library, including its notes. Group libraries are not included. Zotero then gives PaperTok an access key, which is stored encrypted and is only used when you press Send to Zotero while reading a paper. Each send gives Zotero the paper's citation (title, authors, date, identifiers, abstract and link) and the highlights and notes you made on that paper, as one note. Highlights made by the AI are not sent. The note labels each quote as coming from the paper's original text or from a specific reading level of PaperTok's plain-words version. If the source is unknown, the note says so. Disconnecting in Settings → Profile deletes the key, and you can also revoke it at zotero.org. What you already sent stays in your Zotero library.

Email updates

If you turn them on, PaperTok emails you what is new from the authors, topics, institutions and projects you follow, or, while you follow none of them, the notable new papers in the subjects you chose. For that it stores your email address, your sending preferences, what you follow, the identifiers of the subjects you chose and an unsubscribe identifier in Cloudflare Workers KV or, as it moves, in a Cloudflare Durable Object of your account, and your account identifier and that unsubscribe identifier in Cloudflare Workers KV. If a sending limit delays an edition, PaperTok also stores its selected papers, abstracts and reasons they matched your follows or subjects. That selection is removed after delivery, a change to your follows or a later edition that is processed; otherwise its delivery record expires after 120 days. Resend sends the emails. To limit test emails, Cloudflare counts them under a hashed account identifier that it keeps for up to three days.

The legal basis is your consent, and you can withdraw it whenever you like: every email carries a one-click unsubscribe link, and you can also turn updates off in Settings. Unsubscribing deletes the subscription.

Analytics

Usage analytics is enabled by default. PaperTok uses PostHog, hosted in the European Union (Frankfurt), to understand repeat visits and product use. PostHog keeps a random identifier in a first-party cookie and local storage on papertok.app. This identifier connects activity from the same browser across days. When you sign in, PaperTok links that activity to your account identifier. Signing out or switching accounts starts a new browser identity.

Product events record the papers you view, open, like, save, highlight or annotate by their identifier, how long cards and the reader stay on screen, the authors, topics or institutions you follow by their identifier, that you followed a journal but not which one, search types and result counts, and the outcome of sign-in and AI explanation requests. When you open a paper in the app's viewer, they record whether it showed the paper's text (and from which archive), its PDF, a link or nothing, why a PDF could not be shown, and how long the viewer stays on screen. Product events also record when you open a paper's connections or its similar papers, and what you do in the connections map, such as switching to the list or centring the graph on another paper. Creating or opening one of your feeds records the order it uses; creating one also records how many areas or topics, authors and keywords it holds. Deleting one records only that it happened. No event records a feed's name or what is in it. Sending a paper to Zotero records only whether it was sent, not which paper. Feed diagnostics record load starts, duration, result, initial loads versus refreshes or additional pages, automatic retries, and the web and server versions. They contain no paper identifiers or interests. Events include the browser and its version, the operating system and its version, the device type, and session and tab identifiers. They also carry a short browser label built only from the browser, its major version and the operating system (for example "Chrome/130 (Mac OS X)"), so PostHog can tell people from automated traffic; this label never contains the full user agent your browser sends.

PaperTok does not send your name, email address, search text, notes, paper titles or full page addresses to product analytics. Page addresses have query strings, fragments and individual paper, list, feed and profile identifiers removed. PaperTok does not record sessions or automatically capture clicks or heatmaps.

Vercel Web Analytics separately counts page views and countries. It uses no cookies and its visitor identifier expires each day. Page addresses sent to Vercel are stripped of identifiers and query strings too.

You can turn off both services in Settings → Privacy or, without an account, with the Privacy button at the top of the welcome screens and of the public pages a shared link opens: a paper, a list, an author, institution, journal, topic or project page, and the page for a link that no longer works. The choice is kept in this browser. Turning analytics off stops new browser events and removes PostHog's stored browser identity; previously queued events may still be delivered. On subsequent visits neither analytics library loads. A previous rejection of detailed tracking remains disabled until you turn analytics on. To request deletion of previously collected data, write to the contact address below.

Error reports from the browser and backend also go to PostHog. Reports contain a standard error type, code locations, software version and fixed labels. Browser reports use the same identity and stripped page addresses as product events. When the feed fails to load, its error event carries the failure reason, the HTTP status and the error code. Error messages, paper content, notes and account profile details are removed. Turning analytics off stops new browser error reports. Backend reports contain no reader identity and continue independently of this choice.

PostHog's requests go to papertok.app itself, and Vercel relays them to PostHog. PostHog keeps events for one year. Your browser sends its IP address to connect to these services; removing it from an event payload does not hide the connection's IP address from the receiving service.

Scientific sources and AI features

Searches and papers come from public repositories: OpenAlex, arXiv, PubMed, Europe PMC, bioRxiv, Semantic Scholar, CORE, OSTI, NASA ADS, INSPIRE, OpenReview, Hugging Face, Unpaywall and OpenCitations, among others.

Those queries do not leave from your browser: they go through an intermediate server on Cloudflare, so those repositories never receive your IP address from a search or lookup (figures and PDFs are the exception, described below). That server counts requests to respect each source's limits. For a signed-in reader it also counts them per account, under a hashed account identifier it keeps for up to three days.

When the in-app reader requests a paper's text or its PDF, an AI assistant sends a request to PaperTok's MCP server, or a public request looks up papers, paper figures, open-access links or a shared page, that server uses the IP address supplied by Cloudflare to limit requests and keep one visitor from exhausting the shared download allowance. It stores a protected identifier derived from the address and the current minute using a secret key, together with request counts. It does not store the original IP address in these counters or attach your account or the paper you requested. The identifier changes each minute and is not sent to analytics. People sharing a public IP, such as a campus network, share each route’s allowance. The PDF reader allows 30 requests per UTC minute, and the MCP server 30 tool calls.

The text of a paper comes through that server too, but its figures do not: your browser loads each figure, and a PDF shown in the reader on a computer, directly from the repository that hosts it (arXiv, PubMed Central or Europe PMC), which receives your IP address as with any image on the web.

Pages for an author, an institution, a journal or a topic, and an author's preview in the reader, ask Wikidata and Wikipedia from your browser for a short summary and a picture of what they show, and the picture loads from Wikimedia's servers. Wikimedia receives your IP address and the name or identifier looked up, as with any visit to Wikipedia; PaperTok sends nothing about your account.

These PDF counters are stored in Cloudflare Durable Objects. Their automatic deletion is scheduled for one minute after the counting window ends, at most two minutes after a request. They can remain longer if Cloudflare delays or fails to run that cleanup, but expired counters do not affect later windows. The arXiv, paper-figure and open-access lookup counters use the same cleanup schedule, and so do the MCP server's counters. OpenAlex and shared-page lookup counters are stored with their daily allowance and are scheduled for deletion after three days.

Your feed comes from a shared catalogue that server keeps: once a day it collects new papers from those repositories, the same for every reader. Each request for more papers sends it your interests in groups, each with a weight that gives a subject you added recently more of the page and the subjects whose likes may rank it, which leave out the subjects you removed, what you follow, the identifiers of up to 24 papers you liked, saved or read at length, and a compact filter of the papers you viewed or were already shown, and of the papers it offered that you had already liked, saved, read or dismissed, which the server checks papers against to offer only papers you have not seen; liking, saving or opening a paper also asks it for papers related to that one, by its identifier. They are used to choose and rank that page. So the feed can go on from where you stopped, the server keeps the order it ranked for you, and how far down it you got, for up to 2 hours after your last page. It is kept in the cache of the Cloudflare data center where that server runs (placed near Frankfurt; Cloudflare may move it to a nearby data center), under a key derived from your account, and is not copied elsewhere. It holds the papers of that order, not the list of papers you liked, saved or read at length. Deleting your account does not remove it sooner; it expires on its own. Apart from that order, which papers you liked, saved, opened or viewed is not stored there, and neither is the filter. A request for a page of one of your own feeds sends instead that feed's areas, categories, research topics, authors, journals and keywords, the identifiers of up to 24 of the papers you most recently liked, saved or read at length, and the identifiers of the papers you removed from that feed, so the page leaves them out. Until you refresh the feed, it sends no filter of the papers you viewed. Once you refresh it, its pages also send a compact filter of the papers you viewed and of the papers that feed had already shown you, so the feed starts with the papers you have not seen. Papers you marked Not interested are left out by your browser. A paper you liked, saved or opened that the catalogue does not have yet is added to it without any link to you. An author, institution, topic, journal, search or project you follow or put in one of your feeds is added to the list of things the catalogue collects papers for, also without any link to you, and leaves that list within 16 days of the last time anybody asks for it, later only if the daily clean-up fails. Cloudflare Workers AI turns the titles and abstracts of the catalogue's papers into numeric vectors. When papers you liked, saved or opened have no vector yet, their titles and abstracts are sent to it together, without your identity.

Paper explanations and rewrites are generated by an open language model, DeepSeek V4 Flash, run by NaN Builders on servers in the European Union. That model receives the paper's text and the language and level you choose, never your identity or your email.

Who it is shared with

Nothing is sold and nothing is shared for advertising. Only the providers needed for the application to work are involved:

How long it is kept

Your account data is kept for as long as the account exists. You can delete it from Settings → Account → Delete account. That deletes your lists, your highlights, your plain-words summaries, what you follow, your interactions with papers and the lists of papers you viewed, your preferences, the record of the latest policy announcement you were shown, the email subscription, your Zotero connection and the sign-in account, along with what earlier versions of PaperTok kept for you: your public profile, the photo you uploaded, the people you followed, your comments, the content reports you filed and those filed about your comments. The feed order the server keeps for you (see Scientific sources and AI features) is not deleted with the account: that cache cannot be searched by account, so the order expires on its own, at most 2 hours after your last page.

Your rights

You can access, correct and delete your data, object to its processing, ask for it to be restricted and request a copy in a portable format. Much of this you can do from within the app, in Settings. For everything else, write to nicomg60@gmail.com.

If you believe the processing is not correct, you can lodge a complaint with the Agencia Española de Protección de Datos, Spain's data protection authority.

Storage in your browser

PaperTok keeps a few things in your own browser, and no server reads them: the light or dark theme, the language, your choice about analytics and how many times it has changed, a session cache so the application does not reload from scratch when you come back, the last 500 papers your account viewed in this browser, or highly cited papers the feed passed over because you had already liked, saved, read or dismissed them, so a reload skips them, and the metadata for papers in your lists and favorites so their categories can appear without waiting for the network after a reload, and the marks analytics uses to count a first visit per session, a first real use and a return a week later once each. They are cleared when you clear the site's data, the viewed papers and list metadata also when you sign out, and the analytics marks also when you turn analytics off. This browser also keeps, for each of your own feeds, the latest 200 papers you removed from it, under your account identifier. Unlike the things above, these are sent to the server with each request for that feed, so the feed leaves them out. They are cleared when you delete that feed in this browser, when you sign out and when you clear the site's data. While usage analytics is enabled, PostHog also keeps an identifier and your choice in a cookie and in local storage, as described under Analytics. When the browser blocks the Google or GitHub sign-in window, the tab goes to the provider and comes back; during that trip this tab's session storage briefly holds the provider, the part of the app that asked and the page to return to, and they are removed when the page comes back or when the browser restores it after Back. This browser also keeps, under your account identifier, a copy of the latest policy version whose announcement your account was shown, so it does not ask your account again, also after you sign out and in again. Clearing the site's data removes the copy, and so does deleting the account from Settings in this browser.

Changes

If this policy changes in a meaningful way, the date at the top will be updated and the change will be announced in the application to signed-in accounts created before the change took effect. Once the announcement has been shown, your account remembers that version, so the same announcement does not come back on this or another device, however you closed it. The announcement of the 7 November 2026 change is no longer shown, including to accounts that had not seen it.