Hillary Clinton speaks at a rally at Adams City High School in Commerce City, Colorado. AP

A new Atlanta Journal-Constitution poll out Friday has Hillary Clinton leading Donald Trump by 4 points in Georgia, a state Mitt Romney won by nearly 8 points in 2012. Two other polls right after the Democratic convention also showed the Georgia race close — one was tied, and the other showed Trump leading by 4.

But this poll is slightly more recent than those late July polls, and it's consistent with the national polling trend that shows Trump falling apart before our very eyes.

If things get just a little bit worse for Trump nationally, he could start losing a lot of states we normally think of as very safe for Republicans — not just Georgia, but states like Texas, South Carolina, and even Mississippi.

Even as he polls badly nationally, Trump is performing remarkably well among whites without a college degree, especially men. He's getting hammered among college-educated whites, especially women, and he's doing even worse with nonwhite voters than Republican nominees usually do.

Overall, these shifts hurt Trump more in some states than in others.

In a state like Pennsylvania, you can see these effects counterbalancing each other across regions. Trump's weakness with college-educated whites leads to him getting crushed in the suburbs of Philadelphia. (The recent Franklin & Marshall poll has him down by 40 points in those areas; Romney lost the region by just 9 points.) But that's partly offset by gains among working-class whites elsewhere in the state.

But in states like Georgia and Texas, white voters already vote overwhelmingly Republican, and Republicans depend on huge margins among whites to overcome the votes of large, heavily Democratic nonwhite populations. So if Trump loses support among college-educated white women in the suburbs of Atlanta and Dallas, not many high-school-educated white men are available for him to pick up, and upscale whites and nonwhite voters could form a majority coalition for Clinton.

Plus, Republican candidates have usually picked up a significant share of the Hispanic vote in Texas, meaning there is room for Trump to do worse than Romney among nonwhites in the state.

Incidentally, this pattern was part of the theory of how Trump could win a close election: He'd underperform a typical Republican in the South and the Plains states, but not by enough to lose electoral votes, and he'd overperform in the Rust Belt by enough to pick up states like Pennsylvania and Ohio. At a certain level of national support, Trump's overperformance with downscale whites would have tipped the map in his favor.

But the flip side of Trump's shakeup of demographic coalitions is what happens if he loses nationally by, say, 12 points — 8 points worse than Romney's loss four years ago. He won't underperform Romney that drastically in states like Pennsylvania, and he might even hold on to Indiana's 10 electoral votes. But he risks defeat by a coalition of minority voters and upscale whites in ordinarily safe Republican states in the South.

Romney won Texas by 16 points in 2012 while losing the whole country by 4, but Clinton does not need to win nationally by 20 points to have an excellent shot of winning Texas. A national margin in the low teens could do it.

Texas hasn't been publicly polled since June. I'm very interested to see the next poll.