Which is why p-zombies aren't as confounding as people make them out to be. One major feature that separates conscious entities from unconscious is the ability to grasp meaning [i.e. understanding].
An AI like Deep Blue could fit the bill as a kind of p-zombie. If, for example, if Kasparov were to play against deep blue over the internet and was not told he was playing against an AI he might have been fooled into thinking that DB was an actual person. It is able to defeat human chess masters but it cannot be said to understand chess an more than a calculator can be said to understand numbers. It only exists because entities which do have the capacity to understand exist and used that understanding to create it. It is, essentially, just an extension of the minds that made it.
They are algorithmic machines that manipulate syntax but semantics -- meaning -- is beyond their scope. Any construct that is only capable of syntax manipulation could count as some degree of p-zombie. Therefore, p-zombies are entities that can successfully simulate certain classes of cognitive function to give the appearance of intelligence, but can be identified by an inability to grasp meaning, which is inherently non-algorithmic.
Technically, a robotic toy could count as a p-zombie if it could fool a child, animal or anyone else into believing its conscious. It would be theoretically possible to construct a p-zombie sophisticated enough to possibly fool an adult expert but the difference between it and the child's toy would be a difference of degree only. I would posit that there can be no such thing as an indiscernible p-zombie.
All p-zombie constructs can, in principle, be tricked by a discerning conscious agent into revealing their illusory nature.
The only real issue left is that of qualia -- or 'seemingness'. It appears to be an intractable problem of determining the qualitative nature of one's subjective pallet. Qualitative experience is the basis for all meaning , understanding, and the creative capacity to imagine beyond a formal set of rules to generate new ones. At present, we cannot objectively determine what the 'seemingnes' of another entity is from the 'inside' perspective. This is what the core of the "hard problem" really is and what the OP (either intentionally or unintentionally) misses.