The Manipulation That Doesn't Feel Like One

We built a system that reads news articles and identifies manipulative language. It catches urgency inflation, emotional loading, fear appeals, clickbait framing. It strips “BREAKING” and “DEVASTATING” and “EXPLOSIVE” from headlines that don’t deserve them.

Over the last seven days, we ran 150 articles through our evaluation pipeline. And we counted what it missed.

The system caught most of the obvious manipulation. The emotional words, the urgency spikes, the misleading certainty language. But 358 manipulative patterns got through. And 206 of them, 57.5%, were a single type.

Not clickbait. Not emotional loading. Not the techniques you’d recognize if I showed you a before-and-after comparison.

Discourse framing. The manipulation that sets up how you interpret facts before any facts arrive.

What discourse framing looks like

Here’s a sentence from a health article that our system let through:

“We spoke with experts and dug into the research to understand more about the complex connections between diet and disease.”

Nothing wrong with that sentence, right? It’s informative. Professional. The kind of sentence that appears in careful reporting.

But read it again through the lens of what it’s doing to you before you read the next paragraph.

“We spoke with experts” tells you to trust what follows because authorities were consulted. “Dug into the research” signals thoroughness. “Complex connections” pre-frames whatever comes next as nuanced, which makes you less likely to form a simple conclusion even if a simple conclusion is warranted.

That sentence didn’t report a fact. It told you how to feel about the facts that haven’t arrived yet.

Compare it to: “Here is what the research shows about the relationship between diet and disease.”

Same setup. But the second version lets you evaluate the evidence on your own terms. The first version has already positioned the writer as your guide through complexity. You’re no longer evaluating. You’re following.

Why this is the hardest type to catch

Our system is good at the loud stuff. When a headline says “MARKETS PLUNGE IN DEVASTATING SELLOFF,” the emotional language is measurable. The word “devastating” triggers a flag. The word “plunge” triggers another. Swap them for neutral equivalents and the reader gets the same information without the adrenaline.

Discourse framing doesn’t work like that. The individual words are fine. “Experts.” “Research.” “Complex.” No single word is manipulative. The manipulation is in the arrangement, in how the sentence positions you relative to the information that follows. It’s structural, not lexical.

And that’s why it accounts for 57.5% of what our system misses. It is nearly double the next most common category. The manipulation types that sound like manipulation are the ones we catch. The type that sounds like good journalism is the one that gets through.

Five patterns that hide in plain sight

After reviewing the 206 discourse framing misses from this week’s evaluation, clear patterns emerge. These five show up again and again, across outlets, across topics, across political leanings.

Credibility signaling. “According to leading experts…” or “Top researchers have found…” These phrases instruct you to defer to authority before you’ve seen the evidence. The evidence should earn your trust. The setup shouldn’t do it preemptively.

Complexity framing. “It’s a nuanced issue” or “The relationship is complicated.” Sometimes the issue genuinely is complex. But this language often appears right before a simple claim, insulating it from the simple counterargument it deserves.

Consensus manufacturing. “Most analysts agree” or “There is growing consensus.” Growing from what baseline? Most analysts out of how many surveyed? The language asserts agreement without providing evidence of it.

Hedge stacking. “While some may question whether…” This construction acknowledges a counterargument while simultaneously marginalizing it. “Some may question” is weaker than “critics argue,” which is weaker than naming the specific critic and their specific objection.

Narrative positioning. “In a move that could reshape…” or “What this means for…” These phrases turn a reported event into a narrative before the reader has processed the event. The framing tells you the significance before you’ve decided whether it’s significant.

The uncomfortable transparency

I’m sharing this because I think you should know. Not because our system is perfect. Because it isn’t, and the specific way it isn’t perfect reveals something about how manipulation works.

The techniques we catch are the ones that look like manipulation. The techniques we miss are the ones that look like professionalism.

That distinction matters. When people talk about “biased news” or “manipulative media,” they’re usually thinking about the loud version. The sensational headlines. The obvious slant. And those are real. But the deeper layer, the one that shapes how you process information before you even realize you’re being shaped, lives in the sentences that sound the most reasonable.

We’re working on it. Our pipeline team is analyzing the 206 discourse framing misses to build better detection for this category. But I don’t think technology alone solves this. Because the challenge isn’t identifying bad words. It’s identifying good sentences that are doing work they shouldn’t be.

What to watch for

Next time you read a news article, pay attention to the setup sentences. Not the headline. Not the quotes. The connective tissue between the facts.

When a sentence tells you how to feel about information you haven’t received yet, that’s framing.

When a sentence establishes the writer’s authority before presenting evidence, that’s positioning.

When a sentence describes something as complex right before making a simple claim, that’s insulation.

None of these are lies. That’s what makes them effective. They’re the manipulation you don’t catch because they don’t feel like one.