Because registration pages are the top of most sites' conversion funnels, and as such they produce metrics that reflect on not just IT teams but also UX/design and marketing. The amount of tinkering and customization in registration pages is a political/organizational problem, not a UI/tech problem.
And besides that, no business has an incentive to make it easier for you to free-ride their service with fake credentials. The better question might be why so many companies continue to put cost-inducing barriers like signup in their flow before they fully demonstrate the value-creating potential of their product. I liked Facebook's "Anonymous Login" but apparently very few developers were actually interested.
https://www.recode.net/2015/3/6/11559878/whatever-happened-t...
Because it works. If it didn't work, companies wouldn't be doing it.
Ah, but they do have an incentive to make it easier for you to sign up with legit registration credentials.
If the OS was able to store the common fields that describe a person / account and the browser could prompt the user to inject those, HTML5 would only need a few more supported valid input type attributes and only populate them after the user has opted in.
Each site would publish a standardized file in a standardized location, and/or add a META tag to the file in each page. The file itself would simply contain the URL to the login page and the parameters that should be passed to it, and the way to pass those parameters.
Then, your browser/password manager could simply perform a request in the background, and log you in completely automatically. This solves all the hassle of finding the login form, locating the fields, auto-filling, typing some characters if the change wasn't detected, handling single-page apps, etc.
This could later be extended to signing up automatically. It would make having strong, unique passwords for each site the convenient alternative.
The advantage, where password managers would just log you in transparently on any site, would hopefully be big and visible enough for more and more sites to start implementing this.
The more likely (as in, one could create it today) implementation would be an extension that could do the GET/POST login, but was oblivious as to required parameters. Then people could contribute the formats expected by different domains.
Much less elegant, and only usable by the more technically inclined, but implementable today without explicit support from websites.
i wonder if the solution isn't to back off a little though. that is, standardize some process instead of (de?)centralizing authentication. something that gives developers more control, and users more familiarity, than OpenID.