Extract Phone Numbers, Emails and Links from Images

On a clear photo or screenshot our own tests read phone numbers, emails and links all correctly. On a small, heavily compressed image — the kind that has been forwarded through several chat apps — phone numbers still came through but roughly a third of the links did not. Numbers are the most robust of the three.

That is exactly the problem this was built to avoid, and it is why plain digit strings are rejected outright. A number is only accepted if it carries a real telephone signature: an international prefix like +86, a bracketed area code, a 400 or 800 service number, or a recognised mobile range. In testing, invoice numbers, ID numbers, dates, IP addresses, prices, page numbers and version strings were all correctly left out.

Because a link like bit.ly/3xKp9Lm is a random string. When recognition is unsure about a character in an ordinary word, the language model can steer it back; with random characters there is nothing to steer towards. Ordinary domain names made of real words are much more reliable. Always check a shortened link before trusting it.

No. The recognition engine is downloaded into your browser and runs there. The image is read directly from your device and never sent anywhere. This matters more here than for most tools, because a business card contains someone else's personal information.

No, and deliberately so. Those are sensitive identifiers, and a tool that scrapes them out of photographs invites misuse. Only phone numbers, email addresses and web links are extracted.

The first time you use it, your browser downloads the recognition engine and language data — about 6MB in total. After that it is cached, and later runs start immediately.

You will see a message saying so, and the full recognised text is expanded automatically. That way you can tell at a glance whether the image simply had no contact details, or whether they were read but not recognised as such.

Yes. Numbers with an international prefix are recognised regardless of country, and both bracketed area codes and dash-separated formats work. One known gap: a bare ten-digit number with no separators or prefix at all is skipped, because accepting that pattern would mean accepting every ten-digit invoice number too.

Not here — this tool takes one image at a time, since the usual case is a single card or poster. If you need the raw text out of many images, the Image to Text tool handles batches and multi-page PDFs.