Privacy Policy
How we handle information during the Shared Signal beta.
Effective 11 September 2026 · Last updated 9 October 2026
Shared Signal reads articles you are already reading, so this page sets out exactly what leaves your browser, what we keep, and what we do with it.
The short version. The extension reads a page only when you ask it to. Article text is used to produce the analysis and is not stored. Pages behind a login or a paywall never reach the public feed. While Shared Signal is in closed testing you need an account to run an analysis, but not to read the site. The grades and comments you post are used to improve the grader. We do not sell personal information. What people contribute may also be used in aggregated or de-identified form, with no name or account attached, as the business develops, and we will tell you before we change how personal information is used.
Who is responsible
Shared Signal is responsible for the information described on this page. It is currently in closed testing. Send questions, corrections and deletion requests to hello@sharedsignalcommunity.com. We aim to reply within a few days; if you have not heard back within two weeks, please write again.
When the extension reads a page
The extension can read the page you are on only after you ask it to, in one of two ways: by opening its panel and pressing the button that runs an analysis, or by choosing Shared Signal from the right-click menu on the page, which starts the analysis straight away. It does not read pages in the background, does not follow you between tabs, and keeps no record of your browsing.
When you run an analysis, the extension sends the article’s text to the Shared Signal service, along with its headline, the site it came from, the byline and the link.
What happens to the article text
The text is sent to Anthropic’s API, which produces the grading, and is held only for as long as that takes. Anthropic processes it in the United States. The article text is never written to the Shared Signal database. We store the finished analysis: the grades, the reasoning behind them, the claims identified, and a short quote where one is cited.
Anthropic processes the text as a service provider under its commercial terms, which currently do not allow it to use API inputs to train its models. That is Anthropic’s policy and may change; we will update this page if it does.
What is stored, and why
| What | Why | How long |
|---|---|---|
| The analysis, and the article’s headline, link, site and byline | So the same article is not analysed again from scratch, and so the public feed can show what has been run | Indefinitely, unless removed on request |
| A fingerprint of the article text | To recognise an article that has already been run without storing its text | As long as the analysis |
| A record of each analysis you run: the article’s fingerprint and site, the time, whether it had already been graded, and how much processing it used | To apply each account’s daily allowance, stop misuse, and track running costs | As long as your account exists; deleted with your account on request |
| An installation identifier, created by the extension and, separately, by the website for visitors without the extension | Rate limiting, keeping one “did we get this right?” answer per reader, and linking your analyses to your account | Until you clear site data, remove the extension, or ask us to delete it |
| Your answer to “did we get this right?” | To check whether the grader is well calibrated. No account is needed; the answer is recorded against the installation identifier so each reader counts once | Until you ask us to delete it |
| A salted hash of your IP address | Rate limiting by network. The salt means the stored value cannot be turned back into your address | Deleted automatically: within 24 hours on an analysis record and 48 hours on a beta request |
| Your name, email address and Google account identifier, if you sign in | To know who posted a grade or comment. Your Google name becomes your display name, which you can change | Until you ask us to delete your account |
| Grades, comments and evidence you post on readings | To show them on the site and to evaluate and improve the grader | Until you delete them or your account |
| Posts on the beta feedback board | To hear what beta members think of the product, and to help improve how Shared Signal evaluates articles. Visible to beta members only | Until you withdraw them or delete your account |
| When your account was created and which version of the terms was in force | A record of what you agreed to and when | As long as the account |
How we use what we collect
We use the information described on this page to run and improve Shared Signal, administer the beta, communicate with the people taking part in it, understand how the service is being used, and develop new features. That covers what you give us directly, such as your email address, account details, feedback, grades, comments and evidence, and your activity within the service, such as which articles you run.
We do not sell personal information, and we do not share it with anyone for their own advertising or marketing.
What reaches the public feed
Before anything is published, Shared Signal opens the link itself with no cookies and no sign-in, to check that anyone could read it.
- Pages behind a login are never published. This includes intranet pages, shared documents, email, anything on a private network, and anything that sends a signed-out visitor to a sign-in page.
- Subscription articles are never published. If an article sits behind a publisher’s paywall, you still get the full reading, but it is shown only to you, because the text came from your own paid session.
- Anything the check cannot classify is not published, including when the check itself fails.
The feed shows one card per article. It never shows who ran anything, and there is no way to browse what one person has analysed.
Signing in
You can read the feed, any reading and the methodology without an account. You need one to run an analysis, grade an article or post a comment. Public grade averages should come from real people, and each analysis costs money to produce, so during closed testing we need to know who ran it and be able to limit it. Answering “did we get this right?” on a reading page does not need an account.
When you sign in with Google, Google shares your name, email address, account identifier and a link to your profile picture. We use your name as your starting display name and do not show your picture anywhere. Google does not share your password, your contacts or anything else in your Google account. You can disconnect Shared Signal at any time from your Google account’s permissions page.
Sign-in runs on Supabase, which stores the account record. Shared Signal has its own Supabase project, used by no other service.
What is kept on your device
Shared Signal sets no cookies.
On the website, four things are kept in your browser’s storage: your sign-in session, so you stay signed in between pages; a short-lived marker while sign-in is in progress; the installation identifier described above; and which background video the front page showed you last, so it does not repeat. Clearing site data for this domain removes all four and signs you out.
The extension keeps its own storage in your browser: its sign-in session, its installation identifier, an invite code if you entered one, and a short-lived copy of which features are switched on. Removing the extension deletes all of it.
Asking to join the beta
If you submit the form on the front page, we store your name, email address, how you found us, the kind of work you do, how often you say you check a claim, and the date. The two multiple-choice answers help us choose a varied group of testers, and we store exactly the option you picked. We use this to reply to you, send a few updates about the beta, and invite you if you are selected. It is not a marketing list: we do not sell or share it, and we never give it to an advertiser. Ask us to stop writing or to delete it and we will.
We also store a one-way hash of the network address the form was sent from, to stop the form being used to submit lists of addresses. It cannot be turned back into an address.
How your grades and comments are used
Grades and comments you post on readings are used to evaluate and improve the grader. A grading engine drifts if nobody checks it, and the way to find that drift is to compare its grades with the judgements of informed readers. Your grades and comments become part of the set the engine is measured against and tuned with. Your name and account are never part of that work; the evaluation set holds only the judgements.
Each time you post, you choose whether your name appears on the site or you post as Anonymous. We always record which account posted, so you can withdraw your own words and so abuse can be traced, but an anonymous post shows no name to any reader. You can change your display name at any time, and the change applies everywhere it already appears.
We say this on the front page, in the footer of every page and in the disclosures, so that anyone signing in knows before they post.
Posts on the beta feedback board are read by the team, and may be used to improve how Shared Signal evaluates articles. Your name and account are never part of that work.
Contributions may also be used in aggregated or de-identified form, with no name or account attached, as described in the next section. Your name and account are never sold, and your contributions are never used to target advertising at you.
Aggregated and de-identified information
We may analyse information and activity across Shared Signal to understand how it is used, improve how we evaluate articles, develop our products, and better understand how people engage with public information and claims.
We may create aggregated or de-identified information from it, meaning information that does not reasonably identify any individual. That information may be used for research, product development, analytics and other business purposes, including products and services that do not yet exist. Names and account details are removed first, and we do not attempt to re-identify anyone.
Who else handles your data
Shared Signal does not sell personal information and does not run advertising. We count page views with Vercel Web Analytics, which records the page address, the page that referred you, and general device and country information. It sets no cookie, uses no identifier that follows you between sites, and cannot recognise you on a later visit, so no consent banner is needed.
Four companies handle data because the service runs on them:
- Anthropic analyses article text, in the United States.
- Supabase holds the database and sign-in, in Canada.
- Vercel hosts the site and counts page views, from its global network.
- Google handles our email through Google Workspace: the messages you send us and our replies, including replies to beta requests.
Each acts as a service provider and may use what it handles only to provide its service to us. We do not share personal information with anyone else unless the law requires it. Aggregated or de-identified information may be shared, as described above.
The fonts and every other file these pages load come from Shared Signal’s own domain, so loading a page does not send your address to anyone else.
Where your data is kept. The database, which holds accounts, grades, comments and stored analyses, is in Supabase’s Canada (Central) region in Montreal. Article text is processed by Anthropic in the United States and is not stored. Email is held by Google Workspace, whose storage is not limited to one country. If you live outside Canada, using Shared Signal means your account data is held in Canada.
What you can ask for
- A copy of everything connected to your account.
- Deletion of your account, your grades, comments and evidence, your feedback board posts, and the record of analyses you ran. Analyses of articles remain, because they describe the article and contain nothing that identifies you.
- Removal of a specific analysis from the public feed, for example if a page turns out to be less public than the check judged. Tell us and we will take it down.
- Withdrawal of specific contributions. A withdrawn contribution comes off the site immediately and is removed from the evaluation set. We keep a note that you withdrew it, and when, so removals can be audited; that note contains none of what you wrote.
- Aggregated or de-identified information that already exists cannot be traced back to you, so it cannot be withdrawn. Withdrawing a contribution keeps it out of anything created afterwards.
Email hello@sharedsignalcommunity.com. Depending on where you live, you may also have rights under the GDPR or California law to object, to restrict processing, or to complain to a regulator. Those rights apply whatever this page says.
Children
Shared Signal is not intended for anyone under 16, and we do not knowingly create accounts for them. If you believe a child has signed up, tell us and we will remove the account.
Security
Only the service itself can access the database. IP addresses are hashed with a secret salt rather than stored. Keys are held as deployment secrets and are not in the source code. Shared Signal is a small product in closed testing and has not been independently audited, so please do not use it on anything you could be harmed by losing control of.
Changes as Shared Signal evolves
Shared Signal is in development, and the service and the business model may change after the beta. That may include new ways of using, analysing, or providing access to information generated through the platform.
If we materially change how we collect, use, or share personal information, we will update this policy, change the date at the top, and tell anyone with an account before the change takes effect. Where the law requires it, we will also ask for your consent or give you a choice about the change.
Interested in the beta?
No commitment at this time.