Either a lot of Kerry voters have died off or aren't planning to vote, or the Associated Press/GTK poll is skewing towards Bush voters and undercutting Obama's likely results.
800 of the 812 likely voters in this survey claimed to have voted in the 2004 presidential election ( the percentage of voters this fall who did not vote in the 2004 election is likely to be at least three times the 1.5% used by the AP; the CNN exit poll found that 17% of those voting in 2004 had not voted in 2000, and while the very high turnout in 2004 means it will be tough to improve THAT much, it will almost definitely be at least 5%)
But that's not the real problem. The real problem is that among these 800 voters, 50% claim to have voted for George Bush, while only 40% claim to have voted for John Kerry. Bush got 50.73% to Kerry's 48.27% of the vote in 2004, and according to exit polls won senior citizens 52-47 (senior citizens being the most likely to have passed away in the intervening time period). Moreover, if anything, one might except people to be reticent to admit to having voted for Bush (the fact that 7% claimed they voted "other" when only 1% in total may mask some Bush voters). Even if that's not the case, their poll still has a 10 point gap when the election had only a 2.5 point gap.
So, if Obama's down by 4 or 5 among such voters (assuming those who haven't voted favor Obama) in this poll, he's doing 5 or 6 points better than Kerry, which ought to be enough to win.
Unless I'm missing something.