Lingogram EN
← All privacy policiesPrivacy policy · Extension

Privacy Policy — Lingogram for YouTube and Netflix

  • Effective dateJune 22, 2026
  • Last updatedNovember 6, 2026

Covers the Dual Subtitles — Lingogram: Learn Languages on YouTube & Netflix extension.

Lingogram is owned and operated by Aliaksandr Karzhavin (individual developer), "we", "us" or "the developer" below, who is responsible for the data described here. Questions, or a request to see or delete your data: support@lingogram.ai.

This Privacy Policy explains what information the Extension collects, how it is used, where it is stored, and the choices you have. It applies to the Extension wherever it runs: on YouTube and Netflix, on the web pages where it marks your saved words (Section 1b), and on translate.google.com when you import your saved phrases (Section 1h).

This English version at https://lingogram.ai/privacy/youtube/ is the authoritative one.

TL;DR

  • Without an account, we keep no account and no list of your words. The interactive transcript, listening challenge, dual subtitles, and word saving all run inside your browser, and so does downloading a subtitle track as a file. A word you save without an account is kept in your browser only (Section 1a). Some things do leave it, with or without an account: looking a word up sends that word and the subtitle lines around it to our dictionary service, which, while you are signed in, may pass them to OpenAI to answer (OpenAI may use that text, never anything that identifies you, to improve its models), and a phrase you select goes first to Google Translate, without the subtitle lines (Section 1e); a feedback message you choose to send reaches us with the text you typed (Section 1g); the anonymous usage counting described below, which you can turn off; and a random installation identifier, which our server uses to decide which features this browser is offered (Sections 1a and 1k). None of them carries your name, email or account ID, except feedback you send while signed in, which carries your user ID so we can reply. When you sign in, the installation identifier is linked to your account (Section 1a).
  • Word stats for the “Make them stick” deck (on by default, one switch to turn off): for the words you saved, the Extension counts by day how often you look them up and see them on pages, and sends those counts to your account — never the pages, the time of day or the site (Section 1l).
  • Word practice in Telegram (only if you link our Telegram bot): your sentences and messages to the bot go to OpenAI to be scored and answered, with nothing that identifies you; scored sentences stay in your account, and the bot's conversation is deleted about 90 days after it last changed (Section 1m).
  • Signing in is optional. It exists only to sync your saved vocabulary across devices. Words you saved before you signed in move into your account when you sign in (Section 1b). If you choose to sign in, we collect your email address and name and store the words you explicitly save (with the surrounding subtitle lines, or the paragraph around a word you save from the right-click menu on a web page) in our cloud database. A list of those words is also kept on your device, so the Extension can mark a word you already own without asking our servers (Section 3).
  • Diagnostics are opt-in, one click. If subtitles fail to load, an emergency "Reload page" button (shown only after a failed retry) sends us a one-click diagnostic report — the video's address plus technical details — so we can fix the problem. The banner says so right next to the button; nothing is reported automatically.
  • We count anonymous usage, and you can turn it off. The Extension sends us anonymous usage events (for example: the Extension was installed, subtitles loaded, a word was saved) tagged with a random identifier generated on your device — not your email, not your account. That identifier is never joined to your Lingogram account. Click Settings in the toolbar popup (it opens the Extension's settings on our website, Section 1f) → Privacy → uncheck "Share anonymous usage stats" and collection stops immediately.
  • Your saved words are marked on the web pages you read. The page text is compared with your saved words inside your browser and is never sent or stored; you can switch this off for all websites on the Settings page, or for one website in the toolbar popup (Section 1b). Resting the pointer on a marked word shows its translation, and only that word is sent to look it up; a marked phrase of more than one word goes to Google Translate first (Section 1e).
  • AI translation of the second line runs only when you ask for it, on a video the site has no subtitles for in your language. Then, while you are signed in and in the "Dual" mode, the subtitle track of the video is sent to us and translated by OpenAI, which may use the subtitle text to improve its models. The stored track and its translation carry no account ID and serve everyone who watches the same subtitles (Section 1i). AI use is counted per account against a daily and monthly budget (Section 1j).
  • We do not sell your data, show ads, run advertising trackers, build advertising profiles, or track your browsing history.

1. Information We Collect

a. If you do not sign in

The Extension creates no account and stores nothing on our servers that names you: the one record we keep is the anonymous installation record described below. Your language and layout preferences, a local "words saved" counter, and the words you save are kept only in your browser (see Section 3). For each saved word that is the word or phrase as you selected it, the subtitle line or paragraph it came from, a coarse label of where you saved it (such as youtube or web), and the time you first saved it. Once you open the My words page, a translation of the word is kept with it (Section 1e). None of this is uploaded to our servers while you have no account, and no account or email is sent to us unless you sign in. A saved word does leave your browser when it is looked up — on the My words page, or when you rest the pointer on it on a web page — as described in Section 1e: the word alone, without its context.

Five things are sent even without an account, none of them tied to your identity unless you choose to add it:

  • the anonymous usage analytics of Section 1c, which you can turn off in one click;
  • a word you look up, with the subtitle lines around it — Section 1e, which also says how our dictionary service may pass them to OpenAI. This is the feature working, not measurement, so the analytics switch does not stop it; not looking words up does. The My words page looks up the words it lists, without a sentence;
  • the welcome and farewell pages of Section 1f, which are ordinary visits to our website;
  • a feedback message, only if you write one and press Send — Section 1g. It carries what you typed, and a reply address only if you typed one;
  • an installation identifier: a random ID created when the Extension is installed or updated, separate from the analytics identifier of Section 1c and never sent with it. The analytics switch does not stop it. It is sent to our server once, with the edition and whether this was a new install or an update, and again each time the Extension asks which features to offer this browser without an account (Section 1k), at most once a minute while you open videos. It is also sent once a day while the Extension is installed, signed in or not, so we know the installation is still in use; that request carries nothing else. Each time you sign in, it is sent with your account, so we know which of your browsers you use the Extension in; if that request fails, it is sent again later while you are signed in. Our server keeps a record of each installation: the identifier, the edition, whether it was a new install or an update, when it was recorded, the last day it asked about features, when it was last heard from, and, if the farewell page reported it (Section 1f), when the Extension was removed. It does not store your IP address with it. Once you sign in, this record is linked to your account (Section 1b), and the features this browser was offered move to your account.

The Extension also asks our server for the default AI limits to show on its AI badge (Section 1j). That request carries no identifier at all.

b. If you choose to sign in (optional account)

Signing in enables cross-device sync of your saved vocabulary. When you sign in, we collect and process:

  • Account data — your email address, your name (as you type it when you register, or the name on your Google account if you sign in with Google; when none is given, the part of your email before the @ is used), and a Firebase-generated user ID. These identify your account and associate your saved words with you.
  • Saved vocabulary — only the items you explicitly choose to save, while watching or from the right-click menu on any website, or import from Google Translate (Section 1h). For each saved item we store:
  • the word or phrase you selected, both as you selected it and in a normalized form (trimmed, lower-cased) that lets your devices agree it is the same word;
  • a small amount of subtitle context — the saved subtitle line plus the line immediately before and after it, in the video's primary subtitle language only;
  • for a word saved from the right-click menu on a web page, the paragraph of text around your selection (up to 1,000 characters) in place of subtitle context. The page's address and title are not stored (see "Saving from the right-click menu" below);
  • a source tag indicating which edition saved it (YouTube or HDrezka; a word saved on Netflix carries the YouTube edition's tag, since it is the same Extension);
  • a timestamp and a per-day counter used only to enforce a daily save limit.
  • Diagnostic reports — only if subtitles fail to load and you explicitly press the "Reload page" button on the error banner (which states that a report will be sent). Each report contains: the website's hostname, the address (URL) or ID of the video the failure happened on, the subtitle language pair you selected (the language you are learning and your native language), the Extension version, your browser's interface language, a source tag identifying the Extension, a server timestamp, and the technical picture of the failure: a failure code from the Extension's own vocabulary (for example "rate-limited" or "not-offered"), the HTTP status behind it if there was one, how many attempts had been made, and how many subtitle tracks had loaded. Reports are sent only while you are signed in, are capped at one per account per day, and are used solely to investigate the failure.
  • AI translation, only when the site has no subtitles in your language — see Section 1i: the subtitle track of the video, stored without your user ID, and a daily write counter under it.
  • Feedback you send while signed in carries your user ID so we can reply — see Section 1g, which also covers feedback sent without an account.
  • Limit refusals. When a word save, an import from Google Translate, a feedback message or an AI-translation track store is refused because it would exceed one of the built-in limits described in this policy, the Extension tells our server which limit was hit and how many times, with the Extension version and your sign-in token, at most once a minute. The report never contains the word, the message, the subtitle text or the video. It is sent only while you are signed in, so it is tied to your account. We use it to see whether the limits are set right. Our server keeps only the day, the limit and the count, without your user ID.
  • Your installations. Each time you sign in, the installation identifier of Section 1a is sent with your account. Our server records which installation was linked to which account, when first and when last (up to 50 installations per account).
  • Features and activity (Section 1k). For your account our server keeps whether you are in the group that is offered a feature, any feature we switched on or off for you, a note we may add explaining why, and a log of those changes. Every request that carries your sign-in token, including a word lookup, also records the last day your account was active, so we can offer new features to active learners first.
  • Settings kept with your account. So that your settings are the same wherever you use Lingogram, your account keeps a few of them: the language you are learning and your own language (as language codes, for example en and ru), whether you share anonymous usage statistics (Section 1c), whether your saved words are highlighted on websites, whether your lookups are counted for the “Make them stick” deck (Section 1l), and the interface language you chose on our website. The Extension sends a change of the first five while you are signed in, and takes your account's values when you sign in. If both editions are installed, only one of them talks to your account and the other follows it inside your browser. Your account's usage-statistics choice is applied in the Extension only when it says “off”: a choice to share is never taken from another device. These are language codes and on/off choices only: the list of websites you switched highlighting off on stays on your device, and nothing about the pages you read or the videos you watch is kept with them.

Words kept in your browser. If you saved words before you signed in (Section 1a), or while your session had expired, they are in your browser. When you sign in, and each time the Extension starts while you are signed in, it writes them into your account one by one, oldest first, like any other save: the word and its subtitle line or paragraph, with the same source tag and daily limit as above. After each successful write, that word is removed from your browser. A word that fails to upload stays in your browser and is tried again at the next sign-in, start, or successful save. The translation kept with a word, its site label and the time you first saved it are not uploaded. A word too long for your account to hold is discarded. If your session has expired when you save a word, the word is kept in your browser the same way, and the Extension records in your browser that the session needs renewing, so it can tell you to sign in again.

We do not collect: your browsing history, the videos you watch or the pages you read (beyond the subtitle text you explicitly save, the paragraph around a word you save from the right-click menu, the subtitle lines accompanying a word you look up as described in Section 1e, the subtitle track of a video AI translation ran on as described in Section 1i, the site name attached to a feedback message you send, and the single video address included in a diagnostic report you explicitly trigger; the analytics in Section 1c record only a coarse platform label such as youtube or netflix, never a video or a URL), IP-based location tracking, advertising identifiers, or cookies for tracking.

Saving from the right-click menu. When you select text on any web page, the Extension adds a "Save to Lingogram" item to your browser's right-click menu. The page is read only when you press that item, and only that one page: the Extension reads the selected text and the paragraph it sits in, saves them as described above, and shows a short confirmation on the page. Saving never sends anything from a page you did not press the item on, and never records the page's address or title. If you are not signed in, the word is kept in your browser, as Section 1a describes, and is uploaded when you sign in. The analytics in Section 1c count such a save under the platform label web. The saved word carries the source tag of the edition whose menu item you pressed. If both editions are installed, only one shows the item: the one you are signed in to, or the YouTube edition when you are signed in to both or to neither. To agree on this, each edition asks the other, inside your browser, whether it is installed and signed in; the answer is a yes or no and carries no other data about you or the page. The two editions also share, inside your browser, the settings described below and in Section 1c, so that both follow the same choice.

Marking your saved words on web pages. On the web pages you open, the Extension marks the words and phrases you have saved, so you notice them wherever they appear. To do this it reads the visible text of each page inside your browser and compares it with the local list of your saved words described in Section 3. The page text, the words found in it, and the address of the page are never sent anywhere — not to us, not to Google, not to anyone — and nothing about them is stored. The one exception is a word you point at: resting the pointer on a marked word opens a card with its translation, and that word, only the word and not the sentence or the page, is sent to look it up as described in Section 1e — to our dictionary service for a single word, and first to Google Translate for a phrase of more than one word. Removing the word from that card, or saving it again, is an ordinary change to your saved words, as described above; saving it again sends the paragraph it is in as its context, as the right-click save does. The marks are drawn by the browser over the text and do not change the page's content. Because they are drawn on the page, the website's own scripts can see which of its words are marked, and through that, some of the words on your list. To limit this, only text that is actually shown on the page is marked, never hidden text. This requires Chrome's permission to run on all websites. The Extension uses that access only for this marking and for reading the paragraph around a selection you save from the right-click menu (above); it does not run on Lingogram's own website. If you have saved no words, nothing is marked. You can switch it off for all websites with "Highlight my words on websites" on the Extension's Settings page, or for one website with "Highlight words on" that website's name in the toolbar popup. The websites you switch off this way are kept as a list of their host names (for example en.wikipedia.org) on your device; the list is never sent to our servers or to anyone else. The Settings page, and the settings page of our website (Section 1f), show it so you can switch a website back on. If both editions are installed, the other edition is given the same switch and list inside your browser, so both follow the same choice, and only the one that shows the menu item marks words.

Your Lingogram account works across our other Lingogram extensions; if you sign in with the same account, your saved vocabulary syncs together.

c. Anonymous usage analytics (on by default, one click to turn off)

The Extension sends anonymous usage events to Google Analytics 4 so we can see how many people install it, where the Extension breaks, and which steps people give up on. This is on by default. To turn it off, click Settings in the toolbar popup. It opens the Extension's settings page on our website (Section 1f); in its Privacy group, uncheck "Share anonymous usage stats". The page passes the change to the Extension inside your browser, and collection stops immediately. If both Lingogram editions are installed, turning it off or on in one does the same in the other, inside your browser, and an edition installed later starts with it off if you had turned it off in the other.

The identifier. Each event carries a random identifier generated on your device the first time the Extension runs, stored in your browser's local extension storage. It is not your email, not your Firebase user ID, and not derived from either. We never send your account identity to Google Analytics, so there is no key that could join your analytics events to your account — the separation is structural, not just a promise. Clearing the Extension's storage or reinstalling produces a new, unrelated identifier.

The events we send (33 in total):

  • extension_installed, extension_updated — the Extension was installed or updated;
  • onboarding_shown, languages_configured — you saw the first-run screen, you picked your languages;
  • subtitles_loaded, dual_subs_shown, no_subtitles, subs_partial, subs_rate_limited, subs_recovered — subtitles loaded, both languages were shown, none were found, only part loaded, the platform rate-limited us, or a retry succeeded;
  • subs_missed_with_cc — the Extension showed no subtitles although the player's own captions button says the video has some; this one only ever means our own failure;
  • subs_downloaded — you saved a subtitle track to your computer as a file. The file itself is written by your browser and never leaves it;
  • word_save_attempt, word_saved, word_removed — you tried to save a word, it saved, or you removed one;
  • word_lookup — the Extension asked our dictionary service about a word (Section 1e). The event carries the shape of the answer, never the word;
  • signin_started — you began the sign-in flow, and from where (the toolbar popup, the Settings page, the My words page, the status badge, the menu in the YouTube player, the Google Translate import, the AI badge, the page of Lesson, or the note on a word card that offers AI after sign-in);
  • gt_import_preview, gt_import_done — you started an import from Google Translate (Section 1h) and it finished. They carry only counts (phrases found, new, already saved, removed earlier, skipped, added, refused by our server, left unfinished) and, if the import stopped, a short error code; never a word;
  • analytics_opt_out — you turned this analytics off, on the Settings page or on the settings page of our website (sent once, so we know how many people opt out);
  • notification_fetch_failed — the Extension could not reach our service-status messages (see Section 1d). Sent only on failure, never on success, and it carries only the reason (network error, timeout, HTTP error code, or unreadable response);
  • lesson_on, lesson_pause_shown, lesson_answer, lesson_report, lesson_checkpoint_skipped, lesson_off_after_pause, lesson_less_often, lesson_unavailable, lesson_hint — Lesson (Section 1n): you turned it on, the first question paused a video, you answered or skipped a question, you reported one as bad, a checkpoint passed without a question, you turned Lesson off after a question, you chose "less often", Lesson could not be used, or you pressed "Can't answer" and listened again (the event carries only the reasons you ticked: you don't know the words, the speech was too fast or unclear, or both). Counts only: never the question, the answers, the subtitle text or the video;
  • retained_d2, retained_d7, retained_d14 — the Extension was still in use 2, 7, and 14 days after install.

The fields attached to those events, and nothing else:

  • a coarse platform label — one of youtube, netflix, rezka, web, or other; not a hostname, not a URL;
  • the subtitle language pair you picked (for example "en" and "ru");
  • how many subtitle tracks loaded;
  • whether you were signed in — a true/false flag, with no account identifier. On a word you save, false also means it was kept in your browser;
  • a running count of words saved on this device;
  • the Extension version and edition, the build type (a store build or one of our developer test builds), and on extension_updated the version you had before;
  • on developer test builds only, which of our own test servers the build was pointed at — a label about our infrastructure, not about you; builds installed from the Chrome Web Store never send it;
  • days since install;
  • when subtitles fail: a technical failure code, the HTTP status behind it if any, how many attempts had been made, whether a retry had already happened, which half of your language pair was missing, and whether the platform was throttling us;
  • when the platform rate-limits us: whether the request was for an automatic translation, how long we were told to wait, and which back-off step the Extension was on;
  • when subtitles recover: what triggered the recovery (an automatic probe, your manual retry, or a late arrival) and how many seconds it took;
  • for languages_configured only: where you picked them (the first-run screen, the sidebar, the welcome page, the Extension's own pages, or the settings page of our website);
  • for word_lookup only: whether the hover strip or the full word screen asked (the My words page counts as the hover strip), whether the answer came from our cache, a dictionary, a model, or Google Translate, or the lookup failed, whether it was empty, and a coarse latency bucket;
  • for notification_fetch_failed only, why the request failed and, if the server answered, its HTTP status code;
  • for the Lesson events only: whether an answer was right, wrong or skipped, why a checkpoint did not pause (no question, not ready, too little watched, an ad, the tab hidden, or "less often"), why Lesson was unavailable (signed out, AI off, AI budget spent, the service unavailable, no subtitles in the language you learn, or too little speech), and why you could not answer (you don't know the words, the speech was too fast or unclear, or both);
  • a session ID that groups events from one browsing session.

What is never sent: the video you are watching (no title, no URL, no ID), the words you save, look up, or remove, subtitle text, page content, feedback text, your email address, your Firebase user ID, and your browsing history. This is enforced by a deny-list in the code, not by convention: a parameter named like any of those is dropped before the event is built.

Google's role. Google Analytics processes these events for us as our service provider; see Google's Privacy Policy at https://policies.google.com/privacy. On our Analytics property, Google Signals is switched off, so Google does not attach an age, gender, interest category, or advertising audience to these events and does not link them across your devices. Granular location collection is off: events are resolved to country and region only, never to a city. Google collects country and region for every property regardless of this setting; what we switched off is the finer-grained collection on top of it. Every payload is sent with non_personalized_ads: true. Google Analytics is not used to build a profile of you or to target advertising.

d. Service-status messages (no data about you is sent)

When a video platform changes something and subtitles stop working, the Extension can show a short message in its sidebar telling you the problem is known and being fixed, without waiting for a Chrome Web Store update. To do this it periodically downloads a small list of current messages from our Firebase database.

This is a download, not an upload. The request contains no account data, no identifier, no video address, and no information about you or what you are watching — it is the same anonymous request for the same public list that every installation makes, whether or not you are signed in. Which message applies to your installation (by Extension version, edition, platform, and interface language) is decided on your device, from the list already downloaded; none of those details are sent to us.

Because it is an anonymous request to Google's servers, Google receives your IP address as it does for any web request; we neither receive nor store it. The downloaded list, and the identifier of any message you dismiss with its × button, are kept on your device only (see Section 3).

The only thing we learn is described in Section 1c: if the download fails, an anonymous notification_fetch_failed event tells us that our messages are unreachable, so we can fix it. It is sent only on failure, only if analytics is on, and carries only the reason for the failure.

e. Word lookup (the dictionary service)

When you hover or click a word in the subtitles to see what it means, or rest the pointer on a word marked on a web page (Section 1b), or open the My words page (below), the Extension asks our dictionary service for that meaning. The request contains:

  • the word or phrase you pointed at;
  • the language you want it in (your native language, as configured);
  • the subtitle lines around it (the line it came from, and the lines just before and after it), so the service can pick the sense that fits the sentence rather than the most common one. For a word marked on a web page, or listed on the My words page, nothing is sent but the word and the language: not its sentence, not the page, not the address.

The dictionary answers whether or not you are signed in. Signed out, the request carries no account identifier, no email, no analytics identifier, and nothing that ties one lookup to another or to you. Signed in, it also carries your sign-in token, because an AI answer is charged to your account's AI budget (Section 1j) and depends on the features your account has (Section 1k); the service uses it to count AI usage, to check those features and to record the last day your account was active (Section 1b). It does not record which words you looked up against your account. We do not build a history of your lookups against any identity.

How our dictionary service answers. The service runs on a server we rent from OVH, and keeps its database on Google Cloud. It first looks the word up in Wiktionary data published by kaikki.org, sending only the word, from our server. When that has no entry (typically a phrase, a rare word or a name), or when the dictionary is unavailable, and only if you are signed in, the service asks an AI model run by OpenAI through the OpenAI API, sending the word or phrase, your language and the subtitle lines that came with it. For a phrase, while you are signed in, the service may first ask Google Cloud Translation, sending only the phrase and your language; this is being switched on for accounts gradually (Section 1k). None of them receives your IP address, an account identifier or anything else about you. We take part in OpenAI's data-sharing program, so OpenAI may use the text we send it (the word or phrase, your language and the subtitle lines) to improve its models, under its own terms; it receives nothing that identifies you. The service stores each answer in its database, keyed by the word, the language and a one-way hash of the subtitle lines, so the same question is not asked again, and its logs record the word asked (not the subtitle lines) to monitor answer quality (Section 6).

It is a feature rather than measurement, so the "Share anonymous usage stats" switch does not stop it — that switch governs Section 1c. The way to send no lookups is not to look words up: the transcript, dual subtitles, listening challenge and local word saving all work without it. If the Extension was built without a dictionary endpoint configured, the feature is off entirely.

Phrases you select go to Google Translate first. When you drag across more than one word, the Extension first asks Google's public web translator (translate.googleapis.com, then translate.google.com) for that phrase, directly from your browser. That request contains only the selected phrase and the language you want it in. It does not contain the subtitle line, and it carries no account identifier, no email, no analytics identifier and no cookies. Like any web request, it reaches Google from your IP address, and Google processes it under its own Privacy Policy (https://policies.google.com/privacy). The dictionary service is asked about the phrase, as described above, only when Google gives no answer. The same applies when you rest the pointer on a saved phrase of more than one word marked on a web page (Section 1b): only that phrase and your language go to Google Translate. Hovering or clicking a single word never goes to Google Translate.

Answers are cached briefly on your device so the same word is not asked twice.

The My words page. This page of the Extension lists the words kept in your browser (Section 3). For each word that has no stored translation, it makes the lookup above, once per visit to the page, with only the word and your native language, never the sentence the word came from. A phrase of more than one word goes to Google Translate first, as described above. The answer, up to three translations, is stored with the word in your browser. It is not uploaded to your account.

f. The welcome and farewell pages

When you install the Extension, it opens a welcome page on our website. If the Extension that opened it answers, the page shows three setup steps: choose your languages, create an account or sign in (optional; on that same page, with the email and password or Google sign-in described in Section 1b — the account is made on our website, and the Extension is then handed its own sign-in exactly as from the sign-in page), and start with a first video, with a switch for marking your words on web pages. The page passes your choices straight to the Extension inside your browser, which stores them on your device (Section 3); they are not sent to our servers. To show the steps, the page asks the Extension for its current settings and whether you are signed in (and with which email address). The page's address names the Extension so it can reach it.

The settings page on our website talks to the Extension the same way, only with its own message type and the same check of who is asking. To show its switches, it asks the Extension for your settings: your two languages, the languages on offer, the video sites the Extension works on, whether the highlighting of your words and the usage stats (Section 1c) are on, the list of websites where you switched the marking of your words off (Section 1b), whether you are signed in (a yes or no, not your email address), and the Extension's version and edition. When you change a setting there, the page passes the new value to the Extension, which checks it and stores it on your device (Section 3). The Extension answers only our website, does not send any of this to our servers, and puts no saved words, tokens or email address into it. Turning the usage stats off there is reported once, as in Section 1c.

When you uninstall it, your browser opens a farewell page on our website. This is registered with the browser in advance, so the browser opens it on its own; the Extension is already gone at that point and cannot decide otherwise.

The farewell page's address also carries the installation identifier of Section 1a, after a # — the part of an address the browser never sends to any website. When the page opens, it takes the identifier out of its address before anything else on the page runs, and sends it once to our server, so the record of that installation notes when the Extension was removed. That request carries the identifier alone: not the analytics identifier, not the page's address, and nothing you tick or type on the page. The feedback form is still sent only if you press Send (Section 1g).

Both addresses carry the anonymous analytics identifier of Section 1c, so that a visit can be counted against the install it belongs to rather than as an unrelated stranger. If you have turned analytics off, the fixed placeholder opted-out is sent in place of the identifier — the same value for everyone who opted out, which identifies no one. The farewell page is opened by the browser regardless of that setting; what changes is that it carries no identifier of yours.

Visits to our website are ordinary web requests and reach our site with your IP address, which we do not store.

g. Feedback you send us

The Extension has a Send feedback screen in its sidebar settings, and the one-time "Enjoying Lingogram?" card offers a short feedback box if you answer "Not really". Nothing is sent until you press Send. When you do, we receive:

  • the text you typed, up to about 2,000 bytes;
  • a reply address, only if you are not signed in and chose to type one into the optional email field. It is stored as part of your message text;
  • your user ID, if you are signed in, so we can reply through your account;
  • the hostname of the site you sent it from (for example www.youtube.com), the Extension version and edition, your browser's interface language, and a server timestamp.

This works whether or not you are signed in. The message is stored in our Firebase database (Section 4) and is read by the developer to understand what broke and, where you gave us a way to, to reply. It is not used for anything else. Sending is capped at a fixed number of messages per day across all users (currently 500), so a message can occasionally fail to send; the screen tells you if it did.

h. Importing your Google Translate saved phrases

Only if you start it yourself: with Import from Google Translate on the Extension's Settings page, or with the Lingogram icon the Extension adds to the toolbar of the Saved panel on translate.google.com. To place that icon, the Extension looks only at the layout of the Google Translate page, never at its text, and sends nothing from it. When you start the import, the Extension opens translate.google.com/saved in a background tab, reads the list of phrases saved in the Google account you are signed in to there, and closes the tab. The list is read inside your browser; the page itself is not sent to us, and the Extension does nothing else with your Google account. If you are not signed in to Lingogram, the list is read only to show you how many phrases were found, and nothing is saved until you sign in and start the import again.

From each saved pair the Extension keeps only the side in the language you are learning, compares those words with your Lingogram list, and shows how many are new before anything is saved. When you confirm, each new word is saved exactly like a word you save while watching (Section 1b), without subtitle context. Google's translations, the other side of each pair, and phrases in other languages are not saved or sent anywhere. Words you removed from Lingogram earlier are not brought back.

i. AI translation of the second subtitle line

Only when you ask, and only when the site has no subtitles in your language. While you are signed in, AI translation is on for your account (Section 1k), the sidebar is in "Dual" mode (both lines), and the video site offers subtitles in the language you are learning but none in your native language, the sidebar offers it in one line: "Translate with AI", whose tooltip says it uses your AI credits. The first time you press it, a small question says so again, with the credits you have left today, and translates only when you confirm; tick "Don't ask again" and it is not asked again on that device. Nothing is sent until you press it (and, until then, confirm); without an account nothing is offered or sent. When you press it, the Extension asks our server to translate the subtitle track in the language you are learning into your native language and shows the result as the second line; the sidebar shows what the translation is doing. The Extension remembers, on your device only, that you chose to translate that track (Section 3), so it carries on by itself when you open the same video again. While it runs, its status ends in the word "on", which turns it off: nothing more is sent for that video, its translated lines are taken off the screen, and the choice is forgotten, so the next time you open the video it waits to be asked again. While the page stays open the line reads "AI translation · off", and pressing "off" turns it back on. Choosing another mode pauses it (lines already translated stay) and choosing "Dual" again resumes it. If the site's own subtitles in your language appear, they are shown instead and nothing more is sent for that video. On Netflix, the player's audio & subtitles panel gets a short block of the Extension's own on top of its Subtitles list: your two languages as a pair or, when Netflix has none in your language, that language alone and the pair by AI, which work as described here, the question included. It adds nothing to what is sent.

What is sent. For the video you are watching, the Extension takes the subtitle track in the language you are learning: the text of each line and its start and end time, with formatting tags removed. From it, it computes a fingerprint, a one-way hash of the language, the times and the text, which names the track. It then asks our dictionary service for the translation of a range of lines, sending the fingerprint, your native language and the line numbers, with your sign-in token. Lines are asked for around the point you are watching, further ahead as the video plays, and again after you jump. If no one has sent that track before, the Extension stores it once in our Firebase database (Section 4) and asks again. The stored track contains the fingerprint, the language, a label of the site (youtube, netflix or rezka), the number of lines, the total length, the lines themselves, and when it was stored and when it expires. It contains no user ID, no video address, title or ID, and nothing else about you.

A write counter under your user ID. So that one account cannot fill the database, each store also updates a small record under your user ID: the time of your last store, the day, and how many tracks you stored that day (at most 30 a day, at least 20 seconds apart). It holds no fingerprint and no text.

How the server translates. The dictionary service checks the stored track and copies it into its own database. It then asks an AI model run by OpenAI through the OpenAI API to translate it part by part, sending the text of those lines, the rest of the track or the stretch of it around them (so names and tone stay consistent), and the two languages. OpenAI receives no account identifier, no IP address of yours and nothing about the video beyond its subtitle text. As for word lookups, we take part in OpenAI's data-sharing program, so OpenAI may use that subtitle text to improve its models (Section 1e). Every translated part is kept, so a later viewer of the same subtitles, you or anyone else, gets it without a new request to OpenAI. The stored track expires from our Firebase database 14 days after it was stored; the dictionary service's copy and its translations are deleted once no one has used them for 3 days (Section 6).

A second track, for grammatical gender (YouTube only). When your native language marks grammatical gender (so translating "you were right" needs to know who is speaking), and this is switched on for your account (Section 1k), the Extension may also take another subtitle track of the same video: one in another language that YouTube offers, preferably made by people, or else YouTube's own automatic translation of the track you are learning into a language that marks gender. It reads up to three such tracks from YouTube, as the player does, chooses one, and handles it exactly like the first: it computes its fingerprint, stores it in our Firebase database if no one has before (it counts toward the daily write limit below), and names it in its translation requests. The service then sends OpenAI, with each line to translate, the lines of the second track shown at the same time, so the translation can take the gender from them. The second track carries no account identifier either. On Netflix this does not happen.

Shared, not tied to you. The stored track and its translations carry no account identifier, and the service does not record which account asked for which track. The model tokens a translation uses count against your account's AI budget (Section 1j). Its logs record the outcome of each request, the language, how many lines were served or translated, the tokens used and the time taken; never your user ID, the fingerprint or any subtitle text. Like any web request, a request reaches our server from your IP address, and the server's request logs may record that address with the address requested, which contains the fingerprint (Section 6). The service sends the developer alerts through Telegram naming an account's user ID in three cases: the account is refused many times within ten minutes, so the abuse can be looked into; it has spent its whole AI budget (Section 1j); or AI was paused for it. Alerts about a track that keeps failing name the track's fingerprint, never its text.

Answers that fail our checks. Every translation is checked automatically before it is kept or shown. When the model's answer fails a check, the service keeps a sample for the developer to review and improve the checks: the track's fingerprint, the two languages, the line numbers asked for, which check failed and on which line, that line's subtitle text and the model's answer, their lengths and the limit, and the time. A sample carries no account identifier. The developer is alerted through Telegram with the check's name, the two languages and the sample's number, never the text. Samples are deleted after 14 days (Section 6).

AI translation adds no analytics events (Section 1c). The "Share anonymous usage stats" switch does not stop it; a mode other than "Dual" pauses it.

j. Your AI budget

We count AI usage per account. Every request to the AI model, whether to translate subtitles (Section 1i) or to complete a word card (Section 1e), is charged to one budget on your account, with a daily and a monthly limit. For this the dictionary service stores, per user ID, day, kind of request (for example a word card or a subtitle translation) and model, the number of requests and of model tokens used: numbers only, no words, subtitle text, fingerprints or videos. They are kept for 400 days. We may set a different limit for one account, with a short note saying why; each change to such a limit is logged with your user ID and kept for 30 days. Answers that come from the dictionary or from what was already translated cost nothing.

The AI badge. The Extension shows your AI budget on a badge next to the subtitles. While you are signed in, it asks our server for your remaining budget, with your sign-in token, each time you open a video page. Signed out, the badge appears only if our server offers AI to this browser (Section 1k); it then shows the default limits an account starts with, which the Extension asks for with a request that carries no account, no installation identifier and nothing else about you. Nothing is charged while you are signed out. The Extension keeps the latest budget and the default limits in your browser's local storage to show them (Section 3), and removes the budget when you sign out.

k. Which features you are offered

We switch some features on gradually and can switch them off for one browser or one account, for example AI (Sections 1e, 1i and 1j). To know what to show, the Extension asks our server which features it has:

  • without an account, with the installation identifier of Section 1a, each time you open a video page or return to its tab, at most once a minute;
  • while you are signed in, with your sign-in token, at the same moments, when you sign in, and before a part of an AI translation when the last answer is older than ten minutes.

The answer is a list of feature names, each on or off. The Extension keeps the latest one on your device (Section 3) and removes it when you sign out. Our server still checks your account itself for every AI request, so the copy on your device only saves needless requests. The analytics switch does not stop these requests.

l. Word stats for the “Make them stick” deck (on by default, one switch to turn off)

To build the Make them stick deck on our website, the Extension counts, for each word you have saved, how many times a day you open its translation card ("lookups") and on how many pages or videos it was marked ("sightings"). This is learning activity linked to your account, not anonymous data.

  • What is sent, only while you are signed in: the word's key (a one-way hash of the saved word, the same key your saved word already carries in your account), the calendar day, and the two counts. The Extension sends them every few minutes to your account in our Firebase database, and our dictionary service copies them into its own database every few hours, and when you open the deck, to build it.
  • What is never sent: the address, domain, title or text of any page or video, the sentence around the word, which site or player you were on, and anything about words you have not saved. The Extension does not send the time of day: it keeps the time of a lookup in your browser only, to count a word once per 10 minutes and to forget unsaved lookups after 30 days; because counts go out within minutes, our database only notes roughly when a send arrived.
  • Words you have not saved: the Extension counts your lookups of them in your browser only, to suggest saving a word you keep looking up. These counts never leave your browser.
  • Before anything is counted, the Extension shows a one-time notice in this browser, in the translation card, with Sounds good and No thanks. You can turn the counting off at any time on the Settings page of our website; turning it off stops the counting: nothing more is counted or sent, and what was counted but not sent yet is dropped in your browser. What your account already has stays, for the deck, until you remove the word or your account is deleted.
  • Your answers in the deck (Still learning, I know it) are kept with their date per word, until your account is deleted.
  • Removing a word from your list removes its counts from your account within a day of our website next syncing your list.

m. Word practice in Telegram (only if you link our Telegram bot)

If you link your Lingogram account to our Telegram bot (in the bot's app, with your Google sign-in), the bot can send you one of your words to practise and talk with you about learning English. None of this happens unless you link the bot and write to it.

  • Tasks. Once a day at a random time within your hours (10:00 to 21:00 by default, in UTC until you set your time zone with /tz; /daily changes how many) and whenever you ask, the bot picks one of your saved words, one your word stats (Section 1l) show you look up and meet most, or a word from your collection when there are none, and asks you for a sentence with it.
  • What reaches OpenAI: to read and answer your message, our dictionary service sends OpenAI your message, the conversation's latest messages (up to 20, your scores included), the teacher's notes on the earlier ones, up to 30 of your saved words with their definitions, what your latest messages were about (up to 5 words, phrases or sentences) and your open task's word (and, while you dispute a score, that sentence, its word and the score); to score a sentence, the sentence, the word and its definition, or the sentence of yours it says in English. No name, email, account or Telegram identifier is sent. We take part in OpenAI's data-sharing program, so OpenAI may use that text to improve its models (Section 1e). Each request counts against your AI budget (Section 1j).
  • What our dictionary service keeps: each scored sentence with its word, score, better version, tip and time, to choose which word comes next and when and to cap scored sentences per day, and your own task settings (tasks a day, hours, time zone). They are kept until your account is deleted.
  • What the bot keeps in our Firebase database, under your Telegram user ID: your chat ID, your latest task (with its latest scored sentence and score), your conversation with the teacher (up to 30 messages, scores included) and the teacher's short notes on the earlier ones (your level, words practised, recurring mistakes, the last topic), what your latest messages were about (up to 5 words, phrases or sentences, with when), the words of your latest tasks, today's task times, and the IDs of recent Telegram updates, so that each acts once. For each message it records its kind and outcome (for example "scored" or "off topic"); it records the text only of a command (such as /daily 3) and of a word you send to look up. The bot's record expires 90 days after it last changed (your last message, or the last task the bot sent you) and is deleted within a few days of that. /reset erases the conversation, the notes, those topics and the task's scored sentence at once; linking another account, or unlinking, erases the conversation, the notes, the task and those topics.
  • Saving and removing words. A word you ask the bot to save is added to your saved words, as the Extension adds one (Section 1b); a word you ask it to remove is taken off them.
  • Disputing a score. "Оспорить" under a score sends your reason, with the sentence, to OpenAI as above for a second check; the new score, better version and tip replace the stored ones.
  • Telegram (Telegram Messenger Inc.) carries every message between you and the bot, under its own policy at https://telegram.org/privacy.

n. Lesson

Questions about what was said, only when you turn it on. While you are signed in, Lesson is switched on for your account (Section 1k), and the video has subtitles in the language you are learning, the sidebar offers a Lesson switch for that video, off by default. When you turn it on, the video pauses every few minutes, at a break in speech or, when the video is divided into chapters, at the end of a chapter, and shows one question about what was just said, in the language you are learning, with three answers. Nothing is sent before you turn it on, and it stays off on every other video until you turn it on there.

What is sent. The Extension handles the subtitle track in the language you are learning exactly as AI translation does (Section 1i): it computes the track's fingerprint and, if no one has sent that track before, stores it once in our Firebase database, under the same daily write limit; a track that AI translation and Lesson both need is stored once. A little way into each part of the video, it asks our dictionary service for that part's question, sending the fingerprint, the language and the playback position, with your sign-in token. When the video is divided into chapters, it also sends each chapter's start time and title, as YouTube shows them on the progress bar, so the parts follow the chapters; the titles are used only to write the questions and are not stored. Pressing "Bad question" sends the fingerprint, the language, the number of the part's first line and a version number of how the track was cut into parts; no text and nothing about you is stored with the report, only a count on the question.

How the server writes questions. The dictionary service cuts the track into parts the same way for everyone, and asks an AI model to write and check one question for the part asked about, from that part's subtitle text and, with chapters, its chapter's title. The provider is OpenAI (Section 1i) or, when we have chosen it for Lesson, DeepSeek. Either one receives only the subtitle text of that part (and its chapter's title, when the video has chapters): no audio, no video, and nothing that identifies you. The question is checked automatically and kept, so every later viewer of the same subtitles gets it without a new request to the provider. The model tokens it uses count against your account's AI budget (Section 1j); a question already written costs nothing. Questions are deleted with their track's copy, once no one has used it for 3 days (Section 6). The service records no account identifier with a question or a report, and no history of which account asked about which track.

"Can't answer". Before you answer, you can tick why you can't — you don't know the words, or the speech was too fast or unclear — and listen to the passage again in the player, under the question: with the words hidden as in Guess mode, at the speed you pick on the card, or both. Your subtitle mode and playback speed are put back afterwards. It works on the subtitles already in your browser and sends nothing but the analytics count of Section 1c; an answer given after it counts as any other.

Your answers stay on your device. Which questions you answered, and whether right, wrong or skipped, are kept in your browser only (Section 3) and never sent; only the counts of Section 1c are, if analytics is on. The "Share anonymous usage stats" switch does not stop the questions themselves.

o. Technical health reports (under the same switch as Section 1c)

So we can tell when subtitles stop loading for some people after YouTube, Netflix or HDrezka change something, the Extension sends our dictionary service a small technical report about how a video went. It is sent only while "Share anonymous usage stats" (Section 1c) is on, and nothing is queued or sent when it is off.

What a report says. Whether subtitles loaded, did not exist, or were missing although the player's own captions were available; the stage and the failure code it stopped at (the same codes as in Section 1c); how long loading took, in coarse buckets; how many attempts it took; whether one automatic retry helped; which step of the "No subtitles" notice was shown, which button was pressed and whether it helped; the name of a page structure the Extension relies on that was not as expected (for example, that the subtitle list was missing); and, for our own servers, the kind of failure, the group of the endpoint (such as "lookup") and the status of a request that failed or was slow. Every report also names the platform, its build number (a version string the platform's page exposes), the Extension's edition and its version. These automatic reports never carry the video's address, its identifier or title, your account, your email, the analytics identifier or the installation identifier of Section 1c and Section 1a. A report containing any such field is rejected by the server. The one report that does name the video is the one you send yourself with the "Reload page" button of the error banner (Section 1, Diagnostics): it includes the address of the video page you pressed it on, because you asked us to look at that failure. It is never sent automatically, and it is not one of the reports of this section.

How a video's subtitle search went. For each video, the Extension also reports how its subtitle search went: the steps of the search it ran and how each ended (for example "asked the player for the subtitles, no answer"), the result and the reason code, coarse timing buckets, the shape of the subtitle request (which parameter names its address carried and the value of one experiment flag, never the address itself), the shape of each answer (JSON, HTML, empty, an error class), how the Extension asked the player to turn captions on and what happened, the version of the Extension's search settings, whether the video was a Short or a regular video, and whether the Extension is the Chrome Web Store build or a test build. It does not say which video, which language or which track. Each batch of reports also carries a random number, made on your device for that batch only, so that our server does not count a batch twice if it had to be sent again; it is not tied to you, to your installation or to any other report, and the server forgets it after two days.

Every request the Extension makes to our servers also carries the Extension's edition and version in a header, whatever the switch says; our gateway keeps it, and the request's origin, in its access log with the request's route, status and duration, never your address.

What our server does with it. The request is anonymous and not tied to your account. The server looks your IP address up in a database it holds itself (IP-to-country data by DB-IP.com, licensed CC BY 4.0) to keep only the country code, and counts how many different people reported in a 30-minute window by holding a salted one-way hash of the address in memory for that window; the salt changes with each window, and only the resulting number is stored. The address itself is not stored and not written to any log. What is stored is counts per 30-minute window, platform, build, country and Extension version. They are used only to alert us (in a private chat) and to show us a health page for the Extension. Counts are deleted after 14 days; the record of an alert after 90.

Subtitle search rows. A report about how a video's subtitle search went is stored as one row without a country, without any hash of your address and with its time rounded to 30 minutes, so two rows cannot be tied together or to you. Rows are deleted after 14 days. We count them per step of the search to see which step breaks after a platform or Extension change.

Labels and how often reports go out. The Extension gives each report one of three labels, decided on your device: "incident" (subtitles existed but the Extension could not load them even after its automatic retry, or our own server failed for several kinds of requests in a row), "difficulties" (it worked, but not always: a retry was needed, the platform slowed us down, or one of our requests was slow) or none. Reports normally go out in batches about every five minutes. After an incident or difficulties the Extension sends at once and then more often for a few minutes before it calms down, so that we hear about a breakage even if you remove the Extension soon after. A hard cap limits all of this to a fixed number of requests an hour. We can change these timings from our server; the Extension fetches them about every six hours even when the switch is off, and that request carries nothing about you. The counts of each label are stored with the counts above.

If our server cannot be reached. The same report, still free of identifiers, is written to a collection of our Google Firebase project without signing in, so it is not tied to your account, at most once a minute and fifty times a day per installation. Our server later takes it into the counts above, with the country recorded as unknown (no address is involved), moves it to a store that deletes itself after at most 30 days (3 by default), and deletes the original. The original carries an expiry of one day, so a document our server does not collect is deleted by Google within about two days. Writing it is a direct connection from your browser to Google, so Google sees your IP address, as with any request to its servers.

Search settings the Extension fetches. While you watch YouTube, the Extension asks our server, at most about once a minute, for the current settings of its subtitle search (the order of its steps, which steps are on, timeouts and the names of page elements it looks for). The request carries nothing about you or the video: no account, no identifier, no address of the page, no video. Our server sees your IP address, as with any request, and does not keep it for this. The Extension keeps the last settings it received on your device and uses them when the server cannot be reached. This request is made whatever the "Share anonymous usage stats" switch says, because it only receives settings and sends none of your data.

If a video failed, or you remove the Extension within the first hour. When subtitle search ends in failure or "No subtitles" (at most once a day for each installation), the Extension writes a short step log of that one video to a collection of our Google Firebase project without signing in, so it is not tied to your account. The log holds the steps and their results, timing in tenths of a second, the shape (never the text) of the subtitle request and of the answers, how many times captions were turned on and with what result, and the Extension's version and the platform's build. It holds no address, identifier, title, subtitle text, token or language of the video. Our server copies the log to a store that deletes it by itself after 14 days (we can change this between 0 and 30) and deletes the original; the original carries an expiry of one day, so a log our server does not collect is deleted by Google within about two days.

If you remove the Extension within an hour of installing it, the Extension also adds a short digest of how its most recent video went to the address of the page your browser opens after removal, after the "#", so no server receives it with the page. The digest is built from a fixed list of steps, results and answer shapes and holds nothing from the video. The page sends it as its own request, without the analytics identifier or the installation identifier, and then removes it from the address; our server stores it like the step log above. Writing either one is a direct connection from your browser to Google, so Google sees your IP address, as with any request to its servers. When the switch is off nothing is built or kept, and a digest already kept is removed at once.

When you remove the Extension. While the switch is on, the Extension adds a short summary to the address of the page your browser opens after removal, after the "#", so no server receives it with the page: the number of loads that worked, did not work or had no subtitles per platform since installation, the last label, the Extension's version and a rough time since installation ("under an hour", "under a day", and so on). The page sends it as its own anonymous request, apart from the removal report of Section 1a, without the analytics identifier or the installation identifier, and then removes it from the address. With the switch off, nothing is added.

Our own test account. A test browser we run ourselves watches a few videos every 15 minutes with a dedicated test account and reports whether subtitles loaded. It involves no data of yours.

2. How We Use Your Information

We use the information above only to:

  • authenticate you and keep you signed in across sessions;
  • store your saved vocabulary and sync it across your devices so you can review it later;
  • keep the words you save without an account in your browser, and move them into your account when you sign in;
  • enforce a reasonable daily limit on saved words to prevent abuse;
  • answer the word lookups you make (Section 1e), through our dictionary service and the providers it uses (kaikki.org, Google Cloud Translation and OpenAI) and, for phrases, Google Translate;
  • translate the subtitle track of a video into your language when the site has none (Section 1i), through OpenAI, and keep each account within its daily and monthly AI budget (Section 1j);
  • decide which features each browser and each account is offered, and switch a feature off for one of them when needed (Section 1k);
  • see how often the built-in limits refuse a save, an import, a message or a track store, so we can set them right (Section 1b);
  • build your “Make them stick” deck from your word stats (Section 1l);
  • run word practice and the English teacher in Telegram, if you link our bot (Section 1m);
  • investigate the subtitle-loading failures you explicitly report via the "Reload page" button, so we can fix them;
  • read the feedback you send, and reply to it if you are signed in or left a reply address;
  • count anonymous, aggregate usage — how many installs, how often subtitles fail, where people stop before finishing setup — so we can fix what is broken and improve what is confusing. We never use it to identify you or to build a profile of you.

We do not use your information for advertising, profiling, or any purpose beyond providing the sync, word lookup, AI translation, “Make them stick” deck, Telegram word practice, diagnostics, and feedback features, the feature switches and limit reports, and the aggregate usage counting described here.

3. Local Storage (On Your Device)

The Extension uses your browser's extension storage (chrome.storage) to keep, on your device only:

  • the word stats waiting to be sent (per saved word and day: lookups and sightings), your lookups of words you have not saved (kept in your browser only, for 30 days after your last lookup of each), the time of your last lookup of each word in each tab (kept for 10 minutes, or until the browser closes), whether you answered the word stats notice, and, until it reaches us, a request to erase your word stats with your user ID (Section 1l). Signing out or turning word stats off removes the counts, those lookups and the lookup times; the notice answer stays until you remove the Extension;
  • your language and subtitle layout preferences, and whether your saved words are marked on web pages;
  • the list of websites (host names only, such as en.wikipedia.org) where you switched the marking of your saved words off from the toolbar popup. It is never sent to our servers; it is shown to the settings page of our website and shared with the other Lingogram edition, both inside your browser (Sections 1b and 1f);
  • how far you got on the setup page (Section 1f): whether you skipped signing in, and whether you finished;
  • whether the other Lingogram edition is the one that shows the menu item and marks words (a yes/no from the check between editions in Section 1b), so that only one of them does;
  • a local count of how many words you've saved, and a one-time flag recording that the Extension has already asked you to rate it;
  • your analytics on/off setting, the random analytics identifier described in Section 1c, and the date you installed the Extension, plus an analytics session ID in session storage;
  • the installation identifier of Section 1a, whether it was created at an install or an update (until our server has recorded it), which server address recorded it and on which day, and, when sending it with your account at sign-in failed, the server address to send it to again. Signing out does not remove them; removing the Extension does;
  • the latest answer about which features this browser or your account has (Section 1k), when it came, and when the last attempt to ask failed. Signing out removes them;
  • the videos you chose to translate with AI (Section 1i): for each, a short key made of the site's label, your two languages and the length of the subtitle track; no title, address or text. The newest 200 are kept; they are never sent anywhere, and removing the Extension removes them;
  • whether you ticked "Don't ask again" on the question before AI translation (Section 1i); it is never sent anywhere;
  • for Lesson (Section 1n), for the newest 200 subtitle tracks you used it on: the track's fingerprint, whether Lesson is on for it, and for each question you answered or skipped, the number of its part's first line and whether you were right, wrong or skipped it; and, for the whole device, whether you chose "less often", whether you have seen Lesson's one-time hint, where you dragged its question card over the player, and the speed you last picked for listening again. None of it is sent anywhere; signing out does not remove it, removing the Extension does;
  • the default AI limits shown on the AI badge (Section 1j), kept after you sign out; and, while you are signed in, the latest remaining AI budget and marks recording that you were already warned it is running low today or this month. Signing out removes the budget and the marks;
  • the limit refusals waiting to be reported (Section 1b): the name of each limit, a count and when it was first hit, and, so that the same refusal counts once a day, short keys naming what was refused (the word, or a subtitle track's fingerprint). They are kept in session storage, are never sent, and go with the browser session;
  • a cached copy of the service-status messages described in Section 1d, when it was downloaded, and the identifiers of any messages you dismissed, so a message you closed does not come back;
  • if you are signed in: your authentication tokens, your email address, and your user ID (so you stay signed in), and a short-lived sign-in nonce in session storage;
  • a flag, set when your session has expired, that says the session needs renewing, so the Extension can tell you to sign in again. It is removed when you sign in or sign out;
  • while an import from Google Translate is running: the words about to be saved and the progress, in session storage. It is removed when you close the import's result, and with the browser session;
  • a local list of the words you have saved — each word in its normalized form, whether it is currently saved or has been removed, and a marker of how far the list has been synced. It exists so the Extension can mark a word you already own the moment it appears in a subtitle, without a network round-trip. When you are signed in it is filled from your cloud vocabulary (Section 4) and from the words you save on this device, including from the right-click menu. Without an account it holds the words kept in your browser (next item). It holds no translations, subtitle context, page text, or timestamps;
  • the words kept in your browser, saved without an account or while your session had expired: for each, the word or phrase as you saved it, the subtitle line or paragraph it came from, a coarse label of the site, the time you first saved it, and, once the My words page has looked it up, a translation (Section 1e). They are not shared with other browsers or devices. They stay until you remove a word on the My words page, the Extension uploads it after you sign in (Section 1b), or you remove the Extension. Signing out does not delete them.

This local data never leaves your browser except where Section 4 describes (saved words synced to the cloud, and kept words uploaded when you sign in). Signing out removes the authentication tokens, email, user ID, the local list of words synced from your account, the feature answer, and the AI budget and its warning marks from your device; the words kept in your browser, the installation identifier and the default AI limits stay. What the Extension sends while you use it is listed in Sections 1a, 1b, 1c, 1e, 1f, 1g, 1h, 1i, 1j, 1k, 1l, 1m and 1n; none of it is read from the storage described here, apart from the analytics identifier those sections name, the installation identifier of Section 1a, for feedback sent while signed in, your user ID, and the word the My words page looks up, which comes from the kept words.

4. Cloud Storage and Third-Party Services

When you are signed in, your account and saved vocabulary are stored using Google Firebase (Firebase Authentication, Cloud Firestore, and Secure Token Service), operated by the developer on Google Cloud infrastructure. Google processes this data as our service provider; see Google's Privacy Policy at https://policies.google.com/privacy. Access is restricted by Firestore security rules so that you can only read and write your own data. The diagnostic reports of Section 1b and the feedback messages of Section 1g are written to the same database; the Extension can write them but never read them back. The word stats of Section 1l are written there too; the Extension can add to them but never read them. Words kept in your browser (Section 1b) reach this database only after you sign in.

The service-status messages described in Section 1d are downloaded from the same Firebase project. That collection is public and read-only from the Extension: it contains only messages we write, no user data, and the Extension can read it but never write to it.

Word lookups (Section 1e) go to our dictionary service, which runs on a server we rent from OVH (https://www.ovhcloud.com/en/personal-data-protection/) and keeps its database (the installation, feature, AI budget, AI translation and word stats records described in Section 1) on Google Cloud. To answer lookups it uses Wiktionary data from kaikki.org (the word only), for a phrase while you are signed in possibly Google Cloud Translation (the phrase and your language only) and, when that has no entry, the OpenAI API (the word or phrase, your language and the subtitle lines that came with it), as described in Section 1e. We take part in OpenAI's data-sharing program, so OpenAI may use that text to improve its models under its own terms; see https://openai.com/policies/privacy-policy.

Lingogram in ChatGPT. If you choose to connect your Lingogram account to the Lingogram app in ChatGPT, your saved words, each with its context, are copied from our Firebase database into our dictionary service's database when ChatGPT asks for them, and passed to ChatGPT, a service of OpenAI, as described in its own policy at https://lingogram.ai/privacy/chatgpt/. Nothing reaches ChatGPT unless you connect it.

AI translation (Section 1i) stores the subtitle track in the same Firebase database. The Extension can create a track but never read, change or delete one; only our dictionary service reads it. The service translates the track with the OpenAI API as described there, and alerts the developer through Telegram (Telegram Messenger Inc., https://telegram.org/privacy) in the cases Section 1i names.

Our Telegram bot (Section 1m) runs on the same rented server as our dictionary service and keeps its records in our Firebase database; Telegram (Telegram Messenger Inc., https://telegram.org/privacy) carries its messages.

The anonymous usage events described in Section 1c are sent to Google Analytics 4 (via the Measurement Protocol) unless you turn analytics off. Google processes those events for us as our service provider, under the same Google Privacy Policy. Firebase and Google Analytics are used as two separate services and we do not send anything to Google Analytics that would let the two be joined together.

To display subtitles, the Extension reads the subtitle (caption) tracks that the YouTube player already provides for the video you are watching, directly within your browser. On Netflix it reads the subtitle tracks a title already carries, again directly within your browser. This subtitle handling:

  • happens entirely in your browser, with no intermediate proxy of ours;
  • sends no Lingogram account data or saved words to YouTube or Netflix (YouTube receives your usual YouTube cookies with these requests, as it does with the player's own);
  • is subject to YouTube's and Netflix's own privacy policies and terms.

When one of your two languages has no track on YouTube, the Extension asks the YouTube player for that platform's own automatic translation of a track it already serves — a request to YouTube, made from your browser like the player's own subtitle requests, so it carries your usual YouTube cookies; it carries nothing about your Lingogram account. Netflix offers no such automatic translation, so on Netflix none is asked for; a missing language there is translated only by AI, when you ask for it (Section 1i).

A phrase you select, or a saved phrase you point at on a web page, is sent to Google Translate (translate.googleapis.com and translate.google.com) directly from your browser, as described in Section 1e: the phrase and your language only, without cookies or any identifier of ours. Google handles those requests under its own Privacy Policy and terms.

The import of Section 1h opens translate.google.com in a background tab. That is an ordinary visit to Google Translate in your browser, signed in to your Google account if you are, and Google handles it under its own Privacy Policy. Nothing from it reaches us except the new words you confirm.

5. Data Sharing and Sale

We do not sell, rent, or trade your personal data. We do not share it with any third party except: Google (Firebase, Google Cloud and Google Analytics) and OVH as the infrastructure and analytics providers described in Section 4; Google Translate, which receives the text of a phrase you select or point at (Section 1e), and Google Cloud Translation, which our dictionary service may ask about such a phrase; kaikki.org, which our dictionary service asks about a looked-up word, and OpenAI, which it asks about a word or phrase with its subtitle lines when the dictionary has no entry (Section 1e), and with the subtitle text of a track AI translation ran on (Section 1i), and with your sentences and messages in word practice (Section 1m), and which may use that text to improve its models; Telegram, which carries the developer's alerts (Section 1i) and our bot's messages (Section 1m); ChatGPT, only if you connect it (Section 4); or where required by law. None of them receives your name or email from us, and only those Telegram alerts ever carry an account identifier. We do not use your data for advertising.

Who at Lingogram sees it. The developer's admin pages show, for each account, its email, name, role and creation date, the features it is offered (Section 1k), any limit set for it, its AI usage by day and kind of request (Section 1j), and the installations linked to it (Section 1b); and, for installations without an account, a shortened identifier, the edition, the install date and the last day seen. They never show the words you saved or looked up. We use them to run the service, choose who is offered a new feature first, and look into abuse.

6. Data Retention and Deletion

  • Saved vocabulary is retained in the cloud until you request account deletion. Removing a word from your list marks it as removed in your account rather than erasing it, so that syncing or an import from Google Translate does not bring it back; the record, with its word and context, stays until the account is deleted.
  • Words kept in your browser stay there until you remove them. Remove one with the Remove button next to it on the My words page, or remove them all by removing the Extension from your browser. When you sign in, each is uploaded and then removed from your browser. We hold no copy of a word that is only in your browser, so there is nothing for us to delete.
  • Diagnostic reports are kept only for troubleshooting and are covered by account deletion requests (they are keyed to your user ID).
  • Feedback messages are kept until they have been acted on. Feedback sent while signed in carries your user ID and is covered by account deletion requests. Feedback sent without an account carries no identifier of yours unless you typed a reply address; to have such a message deleted, contact us (Section 10) quoting that address.
  • Word lookups (Section 1e) are not kept as a history against any identity: no account or identifier is attached to one, so there is nothing to look up or delete per person. The dictionary service's database keeps each answer, keyed by the word, the language and a one-way hash of the subtitle lines, and its logs record the word asked; as for any web service, our server's request logs also record the IP address of each request. Logs are kept for a limited time and serve only to keep the service running. Requests passed to OpenAI are handled under OpenAI's terms, including its data-sharing program (Section 1e).
  • AI translation (Section 1i): a stored track stays in our Firebase database for at most 14 days, then expires on its own. In the dictionary service's database, a track and its translations stay until no one has used them for 3 days. They carry no account identifier, so they cannot be found or deleted per person and are not deleted with your account. The translation queue's daily counts per user ID are deleted after 8 days; the write counter under your user ID is covered by account deletion requests. Samples of answers that failed the checks are deleted after 14 days.
  • AI budget (Section 1j): the daily AI usage per user ID is kept for 400 days, then deleted; a limit set for your account stays until it is removed or the account is deleted.
  • Installations and features (Sections 1a, 1b and 1k): the record of an installation is kept with no fixed end; it describes a browser install, not a person, and identifies you only once linked to your account. The links between your installations and your account, your account's feature records and the log of their changes, and its last active day are kept until the account is deleted.
  • Settings kept with your account (Section 1b) stay until you change them or delete your account.
  • Limit refusal reports (Section 1b) are kept as daily counts per limit with no user ID, so there is nothing in them to delete per person.
  • Word stats (Section 1l): in our Firebase database, a day's counts are kept for at most 14 days; in the dictionary service's database, daily counts are kept for 90 days and then only as monthly totals per word, kept until you remove the word or delete your account. Your answers in the “Make them stick” deck are kept until you delete your account. An erase request sent by an earlier version of the Extension, and our note of when it was carried out, are deleted within a day of it being done. All of it is covered by account deletion requests.
  • Word practice in Telegram (Section 1m): your scored sentences and task settings stay in the dictionary service's database until you delete your account. The bot's record of your tasks, conversation and notes expires 90 days after it last changed and is deleted within a few days of that; /reset erases the conversation and the notes at once. All of it is covered by account deletion requests.
  • The periods above are our current settings; if we change one, we update this policy.
  • Anonymous usage events are retained by Google Analytics for 2 months, then deleted. Because these events carry no account identifier, we cannot look up or delete the events belonging to a specific person — and neither can you. There is no way for us to tell which events came from you. Turning analytics off on the Settings page stops any further collection, but it cannot retroactively remove events already sent; those expire on the 2-month schedule.
  • Local data can be cleared at any time by signing out (removes your tokens, email, user ID, the local list of words synced from your account, the feature answer and the AI budget) or by removing the Extension from your browser (which also removes the random analytics identifier, the installation identifier and any words kept in your browser).
  • To delete your account and all associated cloud data, email support@lingogram.ai from the address your account uses (Section 10). Deletion is done by hand: we delete your sign-in account (email, name and user ID), your saved words wherever we keep them, including words marked as removed, your diagnostic reports, feedback sent while signed in, your AI-translation write counter and daily counts, your AI usage and any limit set for your account, your word stats (daily and monthly counts and your answers in the “Make them stick” deck), your scored practice sentences and task settings, and the Telegram bot's practice record of the Telegram account linked to yours (Section 1m), the settings kept with your account (Section 1b), the links between your installations and your account together with the installation records linked only to your account, your feature records, the entries naming your account in their change log, and your last active day, and confirm by email within 30 days.

7. Security

Authentication tokens are kept in your browser's extension storage. All network requests are made over HTTPS. Cloud data is protected by Firebase Authentication and Firestore security rules that restrict each user to their own records. No method of transmission or storage is 100% secure, but we take reasonable measures to protect your information.

8. Children's Privacy

The Extension is not directed to children under 13 (or the equivalent minimum age in your jurisdiction), and we do not knowingly collect personal data from them.

9. Changes to This Policy

We may update this Privacy Policy from time to time. Material changes will be reflected here with an updated "Last updated" date. Continued use of the Extension after an update constitutes acceptance of the revised policy.

10. Contact

For any questions about this Privacy Policy, or to request deletion of your account and data, please email support@lingogram.ai. To delete your account, write from the email address the account uses, so we can confirm it is yours. Please do not post your email address in a public place, such as a GitHub issue.


Lingogram is an independent tool and is not affiliated with, authorized, or endorsed by YouTube, Netflix, or any of the video platforms it supports.