This is so because there can be found structures where it seems as if the C-head position is not filled with anything.
The presence of Tense in the C-head being overtly supported by the insertion of the verb do is a good indicator that what indeed occupies the specifier of the CP and not C itself.
In the case of an embedded wh-CP, when the internally merged expression is what, the system is equally directed towards two choices: either the wh-expression projects as the CP-label, thus rendering a free-relative interpretation, or the C-Head itself projects, thus yielding a wh-interrogative interpretation.
Let's assume that in the context of clause-embedding and in accord either with general locality considerations (Rizzi 2004) or some constraint adhering to impenetrability (Chomsky 2001) of the CP domain to anything that lies above the embedded C-head, it is not the verb itself that attracts the wh-expression but C derivatively, due to it being marked as interrogative:
This implies the following: In the light of the interpretative isomorphism between matrix and embedded wh-interrogatives, taking the C-Head as the axis of our symmetry, underspecification of CP-external F-marking equals CP-internal F-marking.
It follows that, taking the C-head as our axis of symmetry, accusative case and T-to-C Movement are functions inverse of each other: