@sand15 This is the evidence for what I said, to wit, that the OP has conceptualized X as the symbolic decision variables of the problem.

1. The OP's code contains the line
   sol:= Optimization[Minimize](obj, constraints, assume= binary)
   So, the OP is trying to solve an Integer Program with all binary variables.
2. The OP assigns no values whatsoever, neither numeric nor symbolic, to any entry of X.
3. The only things that look remotely like decision variables are the X[i,j]s used in the lines defining obj and constraints.
4. obj and constraints are in algebraic form (rather than procedural), so we're required to have symbolic decision variables.
    

@RukevweOyinloye Okay, this version automatically partitions the profiles into num_groups groups. It's not necessary that num_groups evenly divides num_profiles. If it doesn't, the groups will be sized as evenly as possible, with a maximum size difference of 1.

Download MinimaxILP.mw

@acer Thanks. Your reasoning seems quite plausible to me, and I'm satisfied with your explanation.

Now, if we had formally declared bound variables, they could be easily filtered out.

@RukevweOyinloye Okay, I think I have it. Please let me know if this worksheet does what you want. Please read the comments carefully. I implemented this so that the groups can be any partition of the profiles. They need not be contiguous or equally sized. The groups are now relevant, although the number of them is not. 

If the below is correct, then your error in the constraints was simply what I suggested before: You just needed to move the "= 1" to the inner add.

Download MiinimaxILP.mw

@mmcdara The OP's code---posted in the Question---has no occurence of rand. The code is

X:= Array(1..num_profiles, 1..num_websites, datatype= integer[0..1]);

It's obvious (to me at least) that this is intended to be an array of the (necessarily symbolic) decision variables. The OP's conceptual error is trying to impose a numeric datatype on symbolic variables. The syntactic error is that although integer[0..1] is indeed a valid type, datatype treats integer as a special keyword value, and it expects a number of bytes in the square brackets. It's totally understandable that a newbie would make either of these errors.

I made it a matrix just for convenience, and the x could be any unassigned symbol.

@Thomas Dean I found definitive proof that the way that until handles FAIL is a bug. The help page ?do has this paragraph:

• The expr in the until clause is a Boolean expression, which must evaluate to true, false, or FAIL;  otherwise, an error occurs. The repetition statement is terminated when the result is true.

@RukevweOyinloye Yes, that's what I thought: that something was missing. Yet I'm certain that my code has the same mathematical information as your code (same objective and constraints); I just made it linear. But I realize that neither of these codes captures the essence of your verbal description of the problem. Can you formulate algebraically some constraints to capture the role of groups? Perhaps you meant for the "... = 1" on the constraints to be attached to the inner add, thus making 6 constraints in this case rather than 3?

Please also see the similar Reply that I posted in the subthread under the main Question a few minutes ago.

I notice from both your code and text that the "profiles" are partitioned into "groups", and also that the number of groups is irrelevant to this code because each constraint is simply the sum of a complete column of the decision-variable matrix X equated to 1. So, do you think that you need more constraints tthat use the groups?

Is there any real-world reason why the groups would be the same size? It would very easy to recode this so that the groups could be an arbitrary partition of the profiles rather than a partition into equal size blocks.

@acer It seems quite odd to me that indets is "smart enough" to avoid returning bound variables from RootOf, yet it doesn't do it for Int, int, Sum, sum, or any other standard command with a bound variable that I can find. The bound variables of ints and sums are often automatically generated by dsolve and other commands. One often sees _a, _k1, etc.

Do you have any idea why indets gives special treatment to RootOf like this?

I've long thought that symbolic computation needs a formal way to declare a variable as bound.  

@TKChang99 The syntax Surface(...) certainly makes it seem as if Surface is a separate command (with its own help page) rather than simply an option to SurfaceInt, so it's no surprise that there was that confusion. The standard syntax for an option would be Surface= [...], and I can't see any good reason why the integration commands in VectorCalculus couldn't use this more-standard syntax.

@RukevweOyinloye Here is a more-streamlined version of code that does exactly the same thing. While writing this code, I noticed that num_groups is irrelevant for this presumably toy example. Either I'm missing something, or its relevance won't become apparent until the example is more complicated.

Download firstworkablecode.mw

@mmcdara X is the array of binary decision variables such that X[i,j] = 1 iff profile i should be assigned to website j. It shouldn't have a numeric datatype.

@RukevweOyinloye One thing that you certainly must do is remove the minimize from the objective function. The Optimization[Minimize] already specifies that you want to minimize.

You probably have other problems also, which'll be easier to diagnose when you upload a worksheet.

@nm That _Z is the bound variable of RootOf​​​​​, and every RootOf has a _Z, so you'd might as well just filter out all occurences of RootOf. That _Z is very different from the free variables returned by discont, solve, etc., that you're otherwise looking for.

@Carl Love Some commands may also return unknown functions, such as _F(x, y). For example, pdsolve might do this. These can be filtered by type typefunc(suffixed('_')) as the 2nd indets argument. To filter both symbols and functions, make the type 

{thistype, typefunc}(suffixed('_'))

To filter indexed names as well, make it

{thistype, typefunc, typeindex}(suffixed('_'))