The most common tool name on these websites was discontinued in 2007

Adobe ImageReady was the web-graphics companion to Photoshop. It was folded into Photoshop CS3 and discontinued in 2007. Nineteen years later it is still the single most common tool name written into the images on these companies' websites.

Why a dead tool tops a list of live ones

Photoshop is far more common overall: 11,511 images across 217 companies, more than four times ImageReady's total. But those 11,511 are spread across dozens of version strings — Adobe Photoshop 23.4 (Windows), Adobe Photoshop CC 2018 (Windows), Adobe Photoshop 21.1 (Macintosh) and so on — because Photoshop keeps shipping and every release stamps its own name.

ImageReady stopped shipping in 2007, so there is only one version of it left in the world. Its 2,501 images are a single undifferentiated block. Being dead is what makes it the largest single entry.

What the field actually means

This is not a claim that 178 large companies are running a 2007 application. The Software tag records what last wrote the file, and it travels with the file forever. An ImageReady tag means the image was made in the 2000s and has been sitting on the website ever since, through however many redesigns, surviving each one because nobody re-exported it.

Which is the more useful fact. It makes the tag a carbon date for a web page's furniture: not what a company uses now, but how long its oldest surviving assets have been in place.

The live toolchain underneath

The interesting comparison is the last two rows against the first. Photoshop is on 217 companies; Canva and Figma are on 37 and 42. Both are self-service tools bought by a team rather than licensed by IT, and their presence on roughly a sixth of these companies is a reasonable proxy for how much public imagery is now produced outside the design function altogether.

Capture One and Lightroom Classic appear at similar volumes, and those are the professional photographic pipelines — the same two-population pattern that shows up in the location data: a commissioned workflow that behaves, and an ad-hoc one that does whatever the tool did by default.

One tool writes an account identifier into the file

Canva does not only stamp its name. Its renderer writes a string of the form Canva (Renderer) doc=… user=… brand=… — and that user value is a durable identifier for one person's Canva account, embedded in every image they export and published with it.

That is the same category of thing as a camera serial number: not a fact about the company, a fact about a person, and one that would link every graphic that account has ever made across every website it appears on.

OSIC truncates the string at the first identifier, so what is stored is Canva (Renderer) and nothing after it. We found this by fact-checking the numbers for this article, not by predicting it — 475 records had already been written before the rule existed, and they were rewritten in place.

The oldest things still on the front page

Where a capture date survives, it goes back further than the software does. The corpus contains images dated 2000, 2003, 2004, 2005 — a continuous thin tail through the decade. Two are dated 1951 and one 1900, almost certainly scanned archive material where the scanner recorded the date of the original rather than the scan.

Against that, the bulk is recent: 634 images from 2026, 956 from 2025, 776 from 2024. Corporate photography is mostly fresh. It is the site furniture underneath it that never gets replaced.

How this was measured

The first 64 KB of each image is requested, parsed for its metadata block, and discarded. Software is free text written by whatever tool last saved the file. It names a product rather than a person, which is why it is one of the few fields recorded as a value at all — subject to the truncation described above, and to a second rule that withholds the value entirely where it contains an email address or a home-directory path, because those carry a username.

• Only 17,354 of 239,632 images name any software — 7.2%. Everything here is about that 7.2%.
• Capture dates are sparse and self-reported. A camera's clock may simply have been wrong; one image in the corpus is dated 2106.
• It covers the companies we could crawl. See coverage.

Collected 1–4 September 2026, over a completed first pass.