It’s not that simple - each group has a bunch of sub-groups which respond to specific propaganda tactics/buttons.

Same with abortion/anti-abortion, guns/anti-gun, and any of thousands of other topics.

Yeah but you don't need to target them all. Most people are pretty much useless, politically speaking, they just do what they're told by someone else. If you can identify who is doing the telling and target them specifically, large groups of people will otherwise be docile.

Historic attempts to apply that theory have been broad-brush to say the least [0]. With LLMs and access to enough data the authoritarians can get really fine-grained about when they take people out the next time they seize enough power. Anyone attempting to do something politically uncomfortable for the incumbents will be at serious risk in a fine-grained way that has not previously been possible.

> Half of people are pro-Trump, half of people are pro-Isreal/Palestine.

I don't think it is 50-50, more like 20-30% for Trump and I don't have a read on the Israel/Palestine stats. Trump has a dedicated core of supporters but I'd suggest a lot of the people polling for him just don't see a better option.

[0] Eg, I was reading up on https://en.wikipedia.org/wiki/Intelligenzaktion the other day

Stats are really misleading when boiled down to binary decisions.

While polls generally show that roughly half of US voters plan to vote for Trump, that's in the context of only being given the option of Trump or Biden. Most polls I remember seeing since 2016 show roughly 1/3 of the US really consider themselves Trump supporters.

The Israel/Palestine question has similar problems. A binary poll question sets the context that a respondent needs to be on one side or the other, and that supporting or opposing both sides isn't an option. It also puts respondents in a position to have to pick a side regardless of how much or little they may know about the situation. With no more context, a 50:50 split could mean simply that most people don't know enough to decide and randomly pick a side instead.

Nobody randomly picks a side. People who can’t make well-informed decisions simply follow the lean of whatever biases they have. Slightly hawkish or conservative? Pro-trump. Bleeding heart? Pro-Palestine.

Yes, for a search to be considered narrow, the resulting group should be small and specifically defined by precise criteria, not encompassing a significant portion of the population. *Notice once criteria are stacked groups can get small.* E.g. Which young males on your street (or in your building) who like Trump but not Israel were protesting at city hall today? Big datasets let authorities / advertisers / ... answer questions like that. The answer will often be a tiny "narrow" fraction of the population (of your city / state / country / ...).