It depends what type of specifier you use. The data model provides a number of specifier types. A "text position selector" would lock in the annotation at a certain point in the text like the 142nd character. An "xpath selector" would use a DOM-like notation to place the annotation.
If your annotation is a "highlight" then you would need to use these selectors within a "range selector" with a specified start and end point.
If you want your annotations to be robust to content changes, you will probably need to use multiple specifier types. This is allowed by the specification but it felt very clumsy to me to implement.
If the target is available at multiple URLs, there is something available to handle it, but the hosts of the content need to add links to the canonical URL so your annotation software can use that.
What happens if the content changes? Random example: Someone highlights a picture of salad and notes "my favorite food!" and then the publisher changes the image to show roadkill instead of salad.