You can already ask a spoken question and receive a (synthesized-)voice answer. So what kind of re-design/re-factoring/whatever would be required to take a fork of that code and modify it for fact-checking?
In the current "Conversations in the Digital Age" http://www.digitalage.org/ interview with Deborah Norville regarding the upcoming presidential debates, the moderator, Jim Zirin, joked that real-time fact-checking would help keep the candidates honest.
So why's that a joke? Seems like the current voice-recognition and response google interface is already poised to do just that, with possibly not-very-significant modifications. Then anybody watching the debates could just pipe the sound to the app, and hear a "bzzt" (or whatever) whenever stated facts are incorrect. Obviously wouldn't be 100% accurate, but accurate enough to make candidates think twice about what they're saying.
And doesn't have to be such a targeted insult to candidates' honesty. Just a general-purpose fact-checking app. That some people might pipe presidential debates to it would just be one more use for it. (wink,wink, nudge,nudge:)