As I can't guarantee the keyboard to be connected to the computer with a shielded cable
This is a root problem.
Add in the product manual that a compliant USB cable is required, and operation without such is not guaranteed to function, nor warrantied in case of failure.
As for routing, ESD can't be "wired" anywhere because it is a very fast signal. Shunting it into a wire merely turns the wire into an aggressive antenna. If that wire is a trace over ground plane, its characteristic impedance couples ESD into ground anyway (as well as anything else along/nearby the trace).
It's all going to ground anyway, so put it there first and foremost, don't beat around the bush. Surround your circuit with ground. Whether that means extending the ground plane outward somewhat beyond the component area (with top/bottom ground exposed in case of direct strike), or extending it further as shield cans over parts of the circuit, or a whole enclosure.
If ESD is shunted around the board entirely (by flowing along contiguous / bonded panels), then it can only enter on explicit connections made through it -- connectors.
A wave analogy is very apt for this: ESD has a risetime of mere nanoseconds, so the wavefront really does "wash" over mechanical features in real time, like waves crashing around a lighthouse. For a smaller device (10s cm), it may be more like a rising tide than a crashing wave, but the overall effect is no different: the electric field can be 100s of V/cm along a conductor.
Whatever the case, as the chassis gets energized by the propagating wave, wave energy diverts down connectors and cables, radiates off back into space, bounces around between cables (standing waves), etc. Where it transitions onto cables, the equal-and-opposite-reaction carries some fraction of ESD energy into the connector, thus the connector (and signals within) must be bonded to the shield, whether by filters and ESD diodes (unshielded cable) or direct connection (shielded). Or it must be isolated so well that it draws no current in response to the incident wave, but this is very difficult indeed to implement. (Ethernet for example has the best shot at this, being a transformer-isolated interface -- but for a variety of reasons, the shunt strategy was chosen anyway.)
Other approaches -- I have seen designs with, for example, a mostly-isolated loop around the periphery of a PCB, tied common at one point, intended as an ESD sink. I haven't seen any measurements or modeling to suggest that this guard-ring strategy is actually useful, or in what circumstances. (If you [reader] have some, feel free to comment!) Conversely, I haven't done any tests/measurements myself to prove it is actively harmful. I do suspect it is overall more harmful than helpful, but not by enough that it will be easily uncovered as such. That is: more a matter of degree, than an outright fail.
So these are the land of opinion rather than fact. Unfortunately, opinion and superstition are altogether too common in the field of EMC. The reality is: while these things can be modeled and measured, it is often at great expense (e.g. PCB/enclosure full-field simulation, or painstaking setup and testing in the lab), and when project budget and timeline doesn't allow for such, us engineers must fall back on secondary information -- say, application notes instead of research papers; or word-of-mouth and opinion entirely(!). There is good work out there (e.g. Henry Ott's books), but one must take the time to digest them. EMC is a complex topic, requiring holistic evaluation of a system; one should not expect to solve a problem without significant research and understanding. (Or a whole lot of luck.)
USB itself is unfortunately a frequent victim of opinion and word-of-mouth. The simple fact is this: >99% of applications don't need (or can even make use of!) galvanic isolation between signal/power ground and shield, and ground and shield can be tied directly to ground plane at the connector. The connector housing must be tied with the enclosure, if a metallic enclosure / shield is present, and this shield connection causes incident ESD waves to bounce off and wash around the shield, preventing them from entering the board and signals (more precisely: greatly attenuating, perhaps in the 40-80dB range, depending on shielding effectiveness). When a metal enclosure (with EMI fingers bonding it to the connector) is present, the PCB GND/connector shield connection can be made at some distance (mounting screws, say), or through bypass capacitors (to afford some galvanic isolation / DC offset, particularly in multi-card assemblies, like PCs). When a metal enclosure is absent (bare board in space -- plastic enclosures are transparent for EMC purposes), a high quality AC ground connection must be made to the shield. (Again: bypass caps are acceptable, if many are used in parallel, distributed around the connector. Most likely, this isn't necessary, and it can be direct grounded.)
I don't know if this particular issue [miscommunication / misinformation on USB grounding] has affected your design, but it seems to happen often enough to be worth including here anyway.
There are also USB HMI devices, which use a low-baud mode, and can be carried on unshielded cable. In this case, adequate filtering and protection must be provided to divert ESD around the interface device(s), as all wires in the cable shall be considered suspect for purposes of ESD. A plastic enclosure is recommended, to prevent direct ESD strikes in the first place. If metal is chosen, it should be grounded to circuit ground, as an extension of the PCB ground plane.
