Two years on, only 10 of 82 companies have finished renaming Twitter

Twitter became X in July 2023. Changing a link is the cheapest maintenance task a website has: no design review, no migration, no vendor. Two years on, most of the largest listed companies in Britain and America still have not done it.

The interesting number is the middle one

A company that still links to twitter.com everywhere has simply not got round to it, which is unremarkable. What is odd is the 42 that link to both. Nobody decides to use two domains for one social account. That pattern is the signature of a site nobody owns end to end: the footer template was updated, the press releases were not; the investor pages were refreshed last year, the careers section was not.

It is a small, harmless inconsistency. It is also a measurable one, and it turns out to be a reasonable proxy for something companies do not otherwise disclose — how many independent places their public website is edited from, and how long a trivial change takes to reach all of them. Two years, in most cases, and counting.

How this was measured

OSIC crawls a bounded number of pages from each company's own website and records the domains those pages link out to. Aggregating across the index gives 78,911 link records covering 13,125 distinct destinations, collected between 26 and 28 August 2026. Both twitter.com and x.com are counted at the registrable domain, so www.twitter.com and twitter.com are treated as the same destination, because they are.

Three limits are worth stating plainly, because they bound what the number above means.

• It covers the companies we could crawl. A quarter of the index cannot be crawled at all — their CDNs refuse our request for robots.txt, so we treat the site as disallowed and fetch nothing. Those companies are absent here entirely. See why.
• We record the site, not the page. Our schema stores that a company's site links to a domain and when we last saw it — not which page it was on. So a spot check of a homepage will not always reproduce a result that came from somewhere deeper in the site.
• A crawl is bounded, so absence is not proof. A company in the “x.com only” column may still have a stale twitter.com link on a page we did not reach. The direction of that error is worth noting: it can only make the migration look more complete than it is.

We re-checked a sample live before publishing rather than trusting the stored data. Convatec and Entain both still served twitter.com links from their homepages and no x.com link at all. Croda returned 403 to us, as it does to any identified crawler. Antofagasta's homepage carried neither, which is the second limit above doing exactly what it says: our record is of the site, not that one page.

Why bother counting it

On its own, this is trivia. What makes it worth a page is that it is one of the very few externally observable measures of internal website maintenance, and it needs no cooperation from the company to collect. Everything here came from pages those companies publish for anyone to read.

It also dates cleanly. The rename has a known date, so every stale link has a known minimum age — which is not true of most of what infrastructure analysis can tell you. Run the same measurement in a year and the number moves, in a direction that means something.

The full set of outbound destinations, including the regulators, share registrars and whistleblowing services that turn out to describe a company's obligations rather than its preferences, is on the outbound links page.