Privacy policy

When you rest the pointer on a link to another site, move keyboard focus to it, click it, or use “Check this link” from the right-click menu, the extension sends our API:

• the link’s full address, including anything after “?” in it;
• the link’s visible text (or its label, for image links);
• the title of the page you are on, with email addresses and long numbers removed first.

A pointer that only passes over a link doesn’t send it. Links to other pages on the same host as the page you’re on aren’t sent, unless they carry another site’s address (like /url?q=…). “Check this link” sends only the link. Nothing else from the page is sent, and the extension never reads form fields, passwords, cookies or your browsing history.

When you paste one link into the checker on this site, it sends our API that link, plus the link’s text and the message it came in if you fill in those optional boxes.

When you paste a whole message, or share one to LinkGuard from another app on your phone, the page picks out the web addresses on your device and sends only those (up to 8), without the rest of the message. Our server never reads or keeps the rest of the message. Each address is checked as described below, which can include our server visiting it.

Sharing works on Android once LinkGuard is on your home screen. On Android, LinkGuard sets up a small helper in your browser when you visit this site (a service worker that only handles shared messages). It receives what you share on your phone and hands it to the page through this site’s storage in your browser. The page deletes it as soon as it has read it; a share left unopened is ignored after 10 minutes and deleted with the next one. If the helper isn’t running (for example, just after you clear your browser’s data for this site), the shared message reaches our server instead: the server doesn’t read or keep it, and opens the page so you can paste the message.

“Paste and check” reads your clipboard only when you tap it, and your browser may ask you to allow that first.

When you scan a QR code on the phone page, with the camera or from a photo or screenshot you pick, the page reads the code on your device. The camera picture and the photo never leave it. Only a web address found in the code is sent for checking, as above; anything else in a code (like a Wi-Fi password) isn’t sent, and in a code’s plain text only full web addresses (starting http, https or www) count. The camera is used only while you’re scanning on that page, and your browser asks for your permission before LinkGuard can use it.

• It may visit the link’s address. To find where a link really goes, our server requests the address (usually a HEAD request, which asks for headers only) and follows up to four redirects. The site you link to sees a request from our server, not from you; it identifies itself as an ordinary browser so scam sites can’t show it a cleaner page than they show people.
• It checks public phishing and malware lists. Our server downloads public lists of known phishing and malware links (OpenPhish, PhishTank and URLhaus) about once an hour while LinkGuard is in use, and at least once a day, and keeps them on our server. Links are looked up there: nothing about you or the link is sent to the list providers.
• It may look up when the site was registered. For sites that aren’t widely known, our server asks the public registration database for that domain ending (RDAP, the successor to WHOIS) when the domain was registered. Only the domain name (for example “example.com”) is sent: never the full link, its text, the page title or anything about you. Answers are kept for up to 7 days.
• It asks Jev. The address, where it redirects, the link text and the first 120 characters of the page title (both with emails and long numbers removed) are sent to Jev, TypeSafe AI’s model, through TypeSafe AI’s own API. Text that tries to tell the checker what to answer is removed before it reaches the model.
• It returns a verdict (Safe, Suspicious or Scam), scores and reasons. A link on one of the public lists is answered from the list, without visiting it or asking Jev.

Under each result on this site, and in the extension’s popup and scam warning, you can tell us LinkGuard got a link wrong. Only when you do, LinkGuard sends:

• the link that was checked;
• LinkGuard’s answer (Safe, Check first, Suspicious or Scam, and its reason codes);
• your answer (safe, or a scam);
• which part of LinkGuard sent it (this site, the phone page or the extension).

We keep that report without anything after “#” in the link and without any password in it, and with the email addresses and sign-in codes (like tokens and one-time codes) we can recognise taken out. The rest of the link is kept as it is, so please don’t report a private link, like one to a document only you should open. Your IP address isn’t kept with the report: it’s only counted for up to 20 minutes, so one network can’t flood us with reports. Reports are read by the developer to find and fix mistakes; they never change a verdict by themselves. Each report is deleted after 90 days.

• Verdicts are cached on our server so repeat checks are fast: 10 minutes for threat-list matches and for Jev’s answers about links that don’t redirect, 60 seconds for everything else. A cached verdict includes the link, its text and the cleaned page title; it’s looked up by a one-way hash of the request.
• The extension keeps recent verdicts in your browser for the current browsing session and a list of the links checked in each tab (shown in the toolbar popup). Both are cleared when you close the browser.
• Your settings (whether to ask before opening likely scams, and sites where checks are off) are saved in your browser’s sync storage.
• Wrong-answer reports, only if you send one: 90 days (see above).
• So that one network can’t use up the day’s checks by Jev for everyone, our server counts each network’s Jev checks for the day under a code made from its IP address, the date and a secret key, not the address itself. The counts are deleted after two days.
• Our host, Vercel, keeps standard request logs and privacy-friendly page analytics for the website under its own terms. The analytics record which page was opened, without anything after “?” or “#” in its address except campaign tags (utm_…).

There are no accounts. We don’t sell or share data, and we don’t use it for advertising.

• “Read and change all your data on all websites” (Chrome’s and Edge’s wording for running on every page): used only to find the link you point at, draw the tag under it, and ask before a likely scam opens.
• Storage: for your settings and the short-lived verdict cache.
• Context menus: for “Check this link with LinkGuard”.
• Active tab: so the toolbar popup can show which site you’re on when you open it.

LinkGuard can be wrong in both directions. Some links show a grey “Check first” instead of Safe even when no scam signs are found: pages on free site builders, shared documents and other places where anyone can publish, sites that use a brand’s name without being its site, and sites the registry says were registered in the last 30 days. When it can’t get an answer from Jev, only well-known sites are called Safe; others show “Couldn’t fully check” or “Check first”, or a warning if the quick check finds red flags in the link, its text or the page title. Keep your browser’s own Safe Browsing protection switched on.

LinkGuard is built and run by an independent developer.