One publisher's position.
The American Mathematical Society publishes nearly all its books and journals
using LaTeX, and with a volume of more than 10,000 pages a year, requires
consistency in the source files.
Certain components of a published work are provided to outside organizations --
aggregators and indexers -- for further dissemination. These components include
the bibliographic information (essentially the front matter or top matter), the
abstract, and the reference list/bibliography. The organizations handling this
material are usually not equipped to handle LaTeX (or mathematical notation, for
that matter), and in particular, do not have the definitions available that an
author has created. In order for this "shared" material to be intelligible,
it must therefore be reduced to the least common denominator -- no author-defined
For this reason, authors are requested to refrain from using macros that they
have defined in titles, abstracts, and bibliographies. There may be further
restrictions -- LaTeX notation is not recognized in PDF bookmarks, so alternate
representation should be provided.
Another source of problems is the redefinition by an author of existing commands.
Redefinition of, for example, \c, can come to grief when an author with a French
name appears in the bibliography (which is "normalized" in AMS publications to
the form used in MathSciNet). If the \c was defined to be something that
can appear only in a mathematical context, then an error message will result;
if, however, the redefinition can appear in text, it may be overlooked, and
discovered only after the work has been published -- a possible embarrassment
for the author.
However, author definitions for commonly used mathematical notation are encouraged
for the main text, as this enhances consistency and can aid in comprehension for
a sightless reader who depends on an audio interface from the source file. (Well
chosen macro names, ones that are meaningful in context, are important in this
situation. This means a name that is not just easy to type, but a term that
clearly identifies its target.) Use of one-letter macro names is a bad idea, for
reasons of both ambiguity (already defined) and clarity (what does it mean?).
At the AMS, if an author does not follow these instructions, unless there are
significant other problems with a submission, the problems will simply be cleaned
up in-house. However, since any "tinkering" with a file carries some danger of
introducing errors, it is in an author's best interests to follow the guidelines.
Responding to some questions posed in comments, it should be noted that the AMS
production stream does not incorporate an xml/html translation, except for a
"final" translation to html + mathjax for posting of bibliographical material and
abstracts on web pages. Further, when a publication (either a journal or a
proceedings collection) contains multiple papers by different authors, each one
is processed individually (so author-defined material is not merged or
compromised) and the separately composed pdf files are combined into a volume
by processes external to LaTeX.