People don't care for semantics. If you write it and it causes a PC to do something (even if it's just to show a website), it's a programming language. Doesn't need to have branching, variables, etc... (not to mention that HTML includes JS and CSS) print "hello world", <b>hello world</b>, same difference.
Your proposed definition of programming languages includes, as I said, CSV, JPEG, Word, and URLs. Does that mean that anyone who snaps a photo on Instagram is a programmer? This does not seem to be in accordance with the usual meaning of the word "programmer". Is it "programming" to write a dunning letter in Word? It seems to me like a qualitatively different activity.
Programming languages are all more or less equivalent; although they have a variety of paradigms, and although they are supplied with a variety of I/O facilities, you can program more or less precisely the same set of computations in all of them.
Consider the differences between C, Forth, Smalltalk, Prolog, Fractran, Lisp, Brainfuck, Haskell, VHDL, amd64 machine code, and Malbolge: some of these do not even have branching or variables, but they are all programming languages; none of them can compute anything that any of the others cannot.
The exception might be things like Turner's "Total Functional Programming", which excludes nonterminating computations without, he hopes, excluding much of practical interest; Excel, excluding the macro language, also has this limitation, but in Excel it is more onerous.
By contrast, you can't compute so much as a polynomial or NAND in HTML. That is to say, you can't program at all in HTML. So HTML is not a programming language.
English does not include Latin, although, e.g.†, phrases like "inter alia" and "et cetera" can be used in English, and it's not uncommon for an English sentence to quote a Latin motto. Similarly, HTML does not include JS and CSS, just as it does not include the URL specification, and Python does not include SQL or HTML; they are six separate languages.
† exempli gratia
And yet people don't call them "programming languages", while they do call HTML that -- so that even if HTML is not a programming language, there's still a not formally expressed difference that people can intuitively grasp with those other things...
>By contrast, you can't compute so much as a polynomial or NAND in HTML. That is to say, you can't program at all in HTML. So HTML is not a programming language.
You can't program most things in SQL either (not without some modern extensions that make it turing complete), but it is still considered a programming language...
I understand your point that "tt's important to care for semantics in order to have a conversation that conveys information", but I'm of the (e.g. Wittgenstein-ian) school that what matters is not some original/holy definition, but the actual popular usage.
On somewhat of a tangent, it's fairly uncommon for people to call SQL a "programming language", but you actually can program quite a bit with SQL; even without the recursive common table expressions you allude to, you can do surprisingly complex computations with it, and even define functions in the form of views, as long as they are finite. Certainly neither NAND nor evaluating polynomials poses any difficulty for SQL. In fact, here's an example of arbitrary polynomial evaluation in SQL which works in MySQL, MariaDB, Postgres, Oracle, and SQL Server, in this case evaluating 5x⁴ + 2x + 5 at the points 0, 1, 2, and 3:
select x, sum(a * power(x, e))
from (select 3 as x union select 0 union select 1 union select 2) xt,
(select 4 as e, 5 as a union
select 0, 5 union
select 1, 2) t
group by x;
Returning to the subject at hand, I have occasionally heard people saying HTML is a programming language. But I think they're just mistaken. Presumably most people at some point thought the world was flat; we shouldn't attempt to explain away their error by saying that they meant something different by "world" or "flat" than we do. Some people think vaccines cause autism; this isn't actually because they're using the word "vaccine" or "autism" in a different sense than we are. They're just wrong. (As for the people who think organic food contains no chemicals, I'm not sure; I think some of them are just wrong, while others are in fact just using a nonstandard definition of "chemical", meaning something like "pure chemical" or "industrially produced chemical".)
In the same way, people are just wrong if they think HTML belongs to the set that includes C, Forth, Smalltalk, Prolog, Fractran, Lisp, Brainfuck, Haskell, VHDL, amd64 machine code, Malbolge, Scratch, Octave, Python, and R, rather than to the set that includes CSV, JPEG, Word, and URLs. Similarly, bash clearly belongs to the former set, and Excel (without macros) and SQL arguably do.
In a lot of cases, they seem to be reasoning based on shallow surface features like the use of plain ASCII text files. (You can see several examples of this even in the thread here, such as https://news.ycombinator.com/item?id=20906702.) This is similar to the neural network that learned to recognize photos of tanks based on whether the photos had been taken on a sunny day or a cloudy day, or the people who think I'm "hacking into systems" when they see me using a terminal emulator --- an understandable error from the ignorant and foolish, but not one we should allow to confuse our own thinking.
The things you mentioned are usually just data. They are the artifact. They are not generating anything distinct. There is only the data.
I suppose you don't think programs in Scratch, Flash, or Labview, or output from a compiler, are programs?