How Modern Influence Operations Work, Part 1: The New Influence Stack

Dado Ruvic/Reuters

This is a segment of a six-part series analyzing how the Chinese regime conducts its modern influence operations. Read part II here and part III here.

Commentary 

On a Tuesday night in a dorm room, a student opens TikTok for a “five-minute break.” The first clip is a montage of rubble and sirens. The second is a professor-style explainer, neatly captioned, delivering a single moral conclusion. The third is a shaky phone video of a confrontation on another campus—shouts, police lights, a crowd surging like weather.

The student didn’t search for any of it. They don’t even follow the accounts. The feed arrives already confident about what matters.

This is the political technology of our moment: the system that decides—thousands of times a day—what you see next.

The Influence Stack

For most of the past century, influence meant broadcasting. You bought a newspaper, aired a radio spot, printed leaflets, and argued in the town square. Feedback was slow, indirect, and expensive.

Today, influence runs on a different stack. It is microtargeting, figuring out which slice of the population to target. It is a recommender distribution, determining what to present to the target group and in what order. It is a measure of effects—watch time, rewatches, scroll hesitation, comments, shares. And it is iteration, rapidly adjusting what works and discarding what doesn’t.

Once those pieces lock together, persuasion stops looking like a party debate. It takes on the appearance of a thermostat: sense the room, nudge the temperature, sense again.

Microtargeting Didn’t Begin With TikTok

Microtargeting is older than the smartphone feed. Campaigns have long merged voter files with consumer and demographic data, then tailored appeals to specific segments. What changed, especially by the early 2010s, was tempo—the ability to see what’s working while the moment is still unfolding.

The Obama campaign’s 2012 digital operation offers a useful bridge between the older world and the current one. Their teams watched web behavior in near-real time and used it to enable rapid response. During a presidential debate, when then-Massachusetts Gov. Mitt Romney said “binders full of women,” the campaign immediately bought search ads keyed to the phrase and linked to a fact sheet; the campaign’s digital lead described an “immediate uptick in both traffic and engagement” among users searching for that term.

That isn’t TikTok. It’s still the open web—search, ads, landing pages. But the shift shows a new logic: Observe behavior as it happens, then redirect attention before the story cools. Strike while the iron is hot.

Algorithmic platforms industrialize that loop. Microtargeting is not about “who gets which mailer.” It becomes a live system, stitched to distribution and feedback. Different demographics can be shown targeted versions of the same reality, and the system learns—at scale—how each group responds.

And “response” doesn’t require explicit agreement. It can be attention, arousal, and volatility: two extra seconds of watch time, a rewatch, a comment typed in anger and posted, a share to a group chat.

Ranking Systems Don’t Just Reflect Preference; They Shape It

We don’t have to guess whether ranking changes what people see. Researchers have tested it inside platforms.

A large-scale study published in the Proceedings of the National Academy of Sciences of the United States of America (PNAS) drew on a “massive-scale randomized experiment” on X, then known as Twitter, that assigned a randomized control group—nearly 2 million daily active accounts—to a reverse-chronological feed “free of algorithmic personalization,” precisely so the effects of ranking could be measured. The authors reported measurable differences in “algorithmic amplification” across political actors in multiple countries.

A Twitter app icon on a mobile phone in Philadelphia on April 26, 2017. Matt Rourke/AP Photo

That’s the key: Ranking is an intervention. When a system orders content, it decides what becomes salient, what feels common to particular groups, what appears urgent, and what fades. Political power can emerge even when no one writes a manifesto within the company. The feed trains the user. It is an environment, and environments shape behavior.

This is also why the public debate so often misses the point. People argue as if the only question is whether a platform “censors” a viewpoint or “pushes propaganda.” Those concerns matter. They just sit on top of a deeper mechanism: The simple act of ranking, repeated billions of times, changes what societies talk about.

Measurement: The Hidden Power Is the Dashboard

The influence stack is powered by dashboards.

A broadcaster might learn weeks later whether a message landed. A platform learns in minutes whether a clip increased retention among 19-year-olds in a specific place, at a given hour, after a strategically set sequence of prior videos.

This creates a persuasion capability that older institutions weren’t built to match: rapid experimentation on human attention. Content becomes a hypothesis. The audience becomes a living lab. The system keeps what works.

Universities update policy once a semester. Newsrooms adjust framing over days. Legislatures move over months. The scope and focus of the feed can pivot before lunch.

Why Anger Wins Inside the Loop

A hard truth about the influence stack is that not all emotions travel through it equally well. High-arousal emotions move faster because they prompt action.

In a landmark study of sharing, Jonah Berger and Katherine Milkman found that virality is linked to physiological arousal. Content that evokes high-arousal emotions, including anger and anxiety, is more likely to spread than content that evokes low-arousal emotions such as sadness.

Politics adds another accelerant: moral emotion. A PNAS study analyzing large datasets of social media debate found that moral-emotional language increases diffusion. In their sample, each additional moral-emotional word in a message was associated with a substantial increase in sharing.

And anger has particular advantages in networked environments. A computational analysis of Weibo found anger to be more “contagious” than joy and more able to travel along weaker social ties—meaning it can move beyond a tight-knit group and spill into wider communities.

Put those together, and the targeting logic becomes almost mechanical. Anger keeps people watching. It increases the odds they’ll share. It tends to bridge out of local clusters into broader networks. In an engagement-optimized system, anger is not just a feeling. It’s a distribution advantage.

Iteration: How Talking Points Come Back as Optimized Themes

And then there is the old broadcast trick—the repeated phrase, the tagline, the talking point—reappearing in new clothes.

In television news, theming worked because repetition makes ideas feel common. In the influence stack, the system tests variations. It monitors the retention curve, watches share velocity, and comment intensity. The phrases that survive are the ones that travel and harden into slogans that feel “everywhere,” because the platform has learned exactly where “everywhere” is.

This is how a moral frame becomes a transport mechanism. A short phrase is easy to caption, easy to hashtag, and easy to stitch and remix. It is also easy for the system to recognize and route toward audiences that have historically responded to that emotional key.

The Verification Problem

A second political fact of the influence stack is that outsiders struggle to verify what’s happening in real time.

Platforms point to transparency and researcher access. While those programs are meaningful, sometimes they lag behind the speed of events. The influence stack’s advantage lies in its velocity in a world of slow oversight. When you can’t see the full system—distribution weights, downranking rules, recommendation pathways, enforcement decisions—you can’t reliably separate organic waves from algorithmically amplified waves, or evaluate whether interventions were neutral or asymmetrical.

What This Series Will Do

Over the next installments, we’ll walk up the stack.

We’ll examine emotion recognition and why even flawed affect inference can be dangerous when institutions treat outputs as truth. We’ll look at China’s operational model—identity resolution, plus sensor coverage, plus data fusion—and why architecture matters more than any single sensor. We’ll treat TikTok as a distribution layer where iteration is fast, and verification is hard. Then we’ll apply the framework to a test case Americans lived through: the surge of campus protest dynamics during the Gaza war, what we can measure, and what we cannot responsibly claim.

The point isn’t to reduce genuine political conviction to “the algorithm did it.” People protest for real reasons. Institutions fail for real reasons. But in a world where attention is programmable, it becomes reckless to pretend the feed is only entertainment.

The influence stack doesn’t replace politics. It changes the temperature at which politics happens.

And once you see it, the question is no longer whether a single video “caused” anything. The question becomes, who controls the thermostat—and who gets to audit it?

Read part II here and part III here.

Views expressed in this article are opinions of the author and do not necessarily reflect the views of The Epoch Times.