RiskyClick
Paste a link to see whether it looks like phishing. It's judged from the address alone: the page is never opened.
Your result will appear here.
How well it works
Tested on phishing links reported between July 2024 and April 2025, which the model never saw in training, against safe links from sites it never saw either.
What it looks at
Only the text of the link: brand names on sites that don't own them, look-alike spellings (trzor, c0inbase), login and wallet words, free hosting platforms, odd domain endings, redirect wrappers, and the character patterns of 1.36 million phishing links reported to OpenPhish and PhishTank since 2020, set against 2.7 million ordinary links from Common Crawl.
Sites it knows
One rule sits on top of the model: a link on the real domain of a brand it knows (paypal.com, barclays.co.uk…) shows as Looks fine, because genuine login pages often look like phishing to a model that only reads addresses. The exception is pages anyone can publish on a brand's domain, such as Google Docs and SharePoint, which phishers use. The accuracy figures above are for the model alone.
What it can't see
It doesn't open the page, so it can't see what the page asks you for, how old the site is, or who runs it. A careful person couldn't decide about 1 in 5 links from the address alone either. Treat Looks fine as "nothing obvious", not as a guarantee, and never enter passwords on a page you reached from an unexpected message.