Broadly I agree – site structure cannot be final final till you consider page layout and other aspects of the design. Card-sorting results are merely suggestions. You need to add in other design and business considerations.
But the problems with creating structures based on card-sorting, mentioned in the article, are not really problems with card-sorting. The problems are more with half-baked understanding or usage of the technique. For example, the article mentions that browser pages cannot accommodate too many top-level headings, long titles etc., and how this impacts structural decisions. But these and other issues can easily be handled with good card-sorting practices and more better analysis.
a) With regards to number of top-level headings – it is important to remember that you can adjust the parameters of the hierarchical cluster analysis to get different structural solutions that fit your design needs better. I always look at multiple dendogram solutions before deciding which one best suits the present design needs.
This is another reason why I suggest looking at the underlying similarity matrix alongwith the dendogram. The similarity matrix tells you about the underlying relationship between any two elements. The relationships shown by a similarity matrix are useful even if the particular structure from a dendogram is not.
So you can deal with this issue by better analysis.
b) Length of titles: This is taken care of by constraining upfront the length of the titles on the cards. If you have very little navigation space, then you need to consider this while you design the sorting/classification exercise.
So you can handle this issue by thinking about page layout while designing the sorting exercise.
c) Contextual Links: The article mentions “related links” as another reason that one should not finalize using results of card-sort.
“Related links enable people to move across the site oblivious to their absolute position in the hierarchy of the site… Finalising the site structure without considering the potential impact of these factors can result in an unnecessarily bloated site structure.”
This is simply incorrect. Card sorting itself is an excellent way to get at related/contextual links. In fact, clients we have worked with find our “related links” deliverables one of the most useful ways to see the results. If you are not already using sorts to get at related links – start doing that today. (Hint – you need to use the similarity matrix).
So you can handle this issue by better analysis.
The very point of a card-sort is that you can abstract away some of the structural issues and deal with them before you get to the layout, design phase. In practice however, you may need to do some back and forth (as the article, to its credit, points out). But many of the points raised by the article could be handled by doing better analysis and using a list of best practices for doing the card sorting. Here is a start at best practices. Please add more that you use.
a) Consider the page layout (and space available for headings) while creating the cards for the sorting exercise.
b) Look at multiple hierarchical clustering solutions (or dendograms) before deciding on one that suits your design needs.
c) Find ways to give off information scent while designing the structure.
d) Identify related links using a similarity matrix from a card-sorting exercise.
And always remember that good user research does not reduce the need for good design. Card-sorting only tells you how users think and gives you some possible structural solutions. Design – that is still yours to do badly or to do well!