The 20 Forces Shaping the Midterms: My Map of the Kaleidoscopic Minefield
What to watch as AI collides with the campaign — the money, the content, the gatekeepers, and the count
Last Wednesday, I made the case that the environment heading into the midterms — and through 2028 — is a kaleidoscopic minefield: more vectors than any previous cycle, rearranging faster, with some of the biggest ones not yet visible. I also told you the map had fifteen vectors on it.
It has twenty now. I thought of five more between writing that sentence and publishing this one, which is about the best live demonstration of the thesis I could ask for.
A note on how to read this. These aren’t just risk vectors — every one of them is an opportunity vector too. There’s a pro and a con to each, and the job isn’t to dread them. It’s about understanding the lay of the land well enough to plan your communications, messaging, and media. It’s the war-room lesson from last week all over again: you can’t predict everything, so you plan through the process. Brainstorm the scenarios — you need to — just don’t expect them to look exactly as you imagined them. These twenty will show up in some form or another. Some will be load-bearing infrastructure this cycle. Some will be the scandal.
💰 Cluster 1: Money and incentives
1. AI super PACs. AI executives and companies are building super PACs to directly shape races. The open questions: how they play in the general, and whether 2028 gets the same treatment or a pullback — which will largely be decided by whether it works in 2026.
2. The influencer economy. Influencers get paid twice — brand deals on one side, platform reach and payout programs on the other — and both incentives point toward whatever travels. This vector is also adversarial: AI-generated influencers built to monetize political attention are part of it, not a separate problem.
3. Political ad transparency. Facebook and Google still maintain ad libraries; connected TV and podcasts have almost nothing. And the AI-era question underneath: after years of demanding to know how campaigns use these tools, what does that visibility actually look like now, and how does it manifest?
4. Prediction markets. Kalshi and Polymarket are now part of how races get narrated — Spencer Pratt’s LA mayor run made the markets characters in the story, including through the slow count that eliminated him. Money riding on outcomes changes the information environment around those outcomes.
🏭 Cluster 2: Supply side
5. AI slop. The new vector for deepfakes and synthetic content: a flood, not a fake — people not knowing what’s real, and, the underrated part, liking the satire whether or not they know.
6. Campaigns’ use of AI tools. Fundraising copy, voter modeling, response drafting, oppo — mostly mundane and legitimate, which is exactly why the line between acceptable and deceptive gets drawn in public, mid-controversy, by whoever gets caught first.
7. AI texting and voice. Bots trained to sound like the candidate are holding personalized conversations with thousands of voters at once — some voters talk to the agent for hours — and gathering data on what each voter wants while they do it. The dividing line from vector 6: that’s AI working for the campaign; this is AI speaking as the candidate, directly to voters. Audio is the next turn of this same vector — voice memos are becoming a default way people communicate, and I’m watching whether politicians start showing up there, whether they use AI to do it, and what happens when a synthetic voice that sounds like the candidate lands in your messages. Either way, this channel bypasses every gatekeeper on this map — no algorithm, no ad library, no moderation layer, and disclosure only if your state requires it.
🚪 Cluster 3: The gatekeeping layer
8. AI content moderation. Two-fold: how companies are deciding to use AI for moderation at all, and how they’re choosing to apply it to political content specifically — what decisions the systems are actually making, at exactly the nuance political speech requires.
9. AI answers to political questions. Voters are already using these tools to get election information, and how the companies handle that — what the models say, and how anyone would know — is its own standing force. Ads have disclosure rules and libraries; an AI answer has neither.
10. Generative feeds. Zuckerberg has been explicit that the destination is feeds created for you, not just curated for you. Maybe not live by this fall; almost certainly in play by 2028 — and it dissolves the assumption underneath most platform research and regulation, that a shared inventory of content exists to be audited. Whether creators and users want this is a different question. Meta had to pull back a new tool on Instagram after getting pushback.
11. Platform election policies. What platforms are doing and allowing in elections — including the AI companies now writing their own election policies. Every 2026 posture decision is also a 2028 precedent.
🗳️ Cluster 4: The politics of AI
12. Voter sentiment on AI. The mood underneath everything else in this cluster: usage keeps climbing while sentiment keeps souring — a billion people using tools they increasingly don’t trust. The vectors below are where that mood gets concrete.
13. AI and jobs. The version of the mood people can feel: what this means for my job. This is headed for the campaign trail, and it may be the biggest AI story of the cycle for actual voters.
14. Data centers. Electricity, water, and land fights are turning AI infrastructure into a kitchen-table issue in the districts that decide midterms — where AI stops being a tech story and becomes a cost-of-living story.
15. Kids’ safety. The one lane where Congress keeps moving — because it plays well on the campaign trail.
16. State vs. federal vs. geopolitical. AI runs from hyper-local to global in one line: states moving fast on AI laws, a federal government mostly not, and the geopolitics of AI pressing down on both. Texting disclosure is already a live example — North Dakota and California require campaigns to tell you you’re talking to a bot; most states don’t.
⚡ Cluster 5: Horizon and hard edges
17. Cybersecurity. The classic vector, re-weaponized with AI on the attacker side: spear phishing against campaign staff and county offices, automated vulnerability scanning, hack-and-leak with synthetic garnish. Defenders: underfunded county IT. Attackers: interns that never sleep.
18. Election officials and government agencies. Their capacity, their tools, and their role in 2026 and beyond — AI can genuinely help election administration, and the same offices are under-resourced against what’s coming at them. The harder question sits above them: how senior officials react to results they don’t like.
19. The post-election window. Attention doesn’t end on Election Day — the count is where confusion lives, and confusion is where fraud claims and trust erosion get their opening. LA just gave a preview when second place flipped a week after election night, and California and the DOJ are already clashing over an election fraud probe months before a single midterm vote is cast.
20. AI agents, world models, and the road to AGI. The honest entry: we don’t know. Agents that browse, transact, and communicate on a user’s behalf could be registering voters — or impersonating them — by 2028. World models are coming and America isn’t prepared for them. And by 2028, we’ll either have AGI or be meaningfully closer to it — the experts disagree by decades on when, which is itself the finding — and that possibility alone adds vectors nobody can name yet. This one is on the board precisely because its shape is unknowable from here.
How the map turns
Twenty vectors is a lot to hold at once, so here’s the organizing move: the vectors are the pieces, but what decides the cycle is the configuration — which arrangement of these forces dominates the story. This is why it’s a kaleidoscope: the image changes with even a slight turn of the wheel.
I know where a lot of the anxiety about all of this comes from. It’s not the vectors themselves — it’s what we don’t know, what might happen, the worry that we won’t be prepared and we’ll be shocked. A lot of that traces straight back to 2016 and the collective vow to never be surprised like that again. I can’t promise you we won’t be surprised. We might be. There’s a reason “October surprise” is a phrase. There will be twists and turns between now and November, and more between November and 2028.
The trick is to let all of this make you feel more ready, not more anxious. Knowing what’s influencing you, what this world looks like, and what’s changing is the difference between being caught flat-footed and having already thought it through. That’s what I mean when I say panic responsibly.
This map will be wrong. Not all of it, and not all at once — it grew from fifteen to twenty in the week before I published it, and vector 20 exists because the most important piece may not be on the board yet. We’ll be talking a lot in the months ahead about what it gets right and what it gets wrong.
We’re living through a moment where AI, politics, media, and technology are all crashing into each other at once. Anchor Change is where I connect the dots, share what I’m noticing, and help people panic responsibly about what comes next. Subscribe for grounded analysis and strategic insight from someone who’s been inside the rooms where these decisions get made.
This is a useful framework because it shifts the conversation away from predicting a single “big issue” and toward understanding how multiple forces interact simultaneously. I especially liked the distinction between risk vectors and opportunity vectors—too many analyses focus only on threats instead of how campaigns can adapt. The point that really stuck with me was that the configuration of the vectors matters more than any individual one. That’s a much more realistic way to think about modern campaigns than assuming one narrative will dominate for two years.