What is an example of something that automated scanning cannot capture?

There is a wide range of things that require human auditing to discern conformance. One example is text within an image. Text within an image is often considered a WCAG failure unless that text is also somewhere on the page. In this instance, scanners cannot typically detect if an image has text; they can only detect that an image is present on the page. If they do pick up that there's text in an image, human analysis is needed to see if that text exists on the page as well.

Another example is detecting logical heading hierarchy. While a scanner would detect headings, it wouldn't necessarily know if they are at the correct level. If you have a level two heading at the top of your page that should be level one, the scanner will detect that there is a heading, but it won't know that it needs to be level one. Even if the elements exist on the site, they may not make sense for a screen reader user.

While a scanner can start the job, you often want a human to verify accuracy or a smooth user experience.