Why Porn Only Ever Gets More Extreme — And What That Tells You About Every Feed You Use

The situation

A few years ago, the singer Billie Eilish told the Howard Stern Show that she started watching porn around age 11. She described becoming addicted, having nightmares from how violent the content was, and struggling later in her own sex life: "The first few times I had sex, I was not saying no to things that were not good."

Her story points at something measurable. Content that was rare in commercial porn forty years ago — strangulation, incest themes, overt violence — is now mainstream. "If you take incest and strangulation, we can look at content studies from the 1980s... when there was hardly any of that material," says Clare McGlynn, professor of law at Durham University. "Today both incest and strangulation becomes mainstream." From early November, the UK government will ban possession and publication of both.

The interesting question is not whether the content got darker. It did. The question is why the drift runs in one direction — always toward the more extreme, never back. The answer is a well-established pattern, and it has nothing specific to do with sex.

The model: when the measure becomes the target

In 1975 the British economist Charles Goodhart, watching central banks try to control the money supply, noticed something that now bears his name. Goodhart's law: when a measure becomes a target, it stops being a good measure. Aim a system at a number that stands in for the thing you actually want, and the system will optimise the number — hitting it in ways that drift away from the thing you cared about.

Attention platforms are Goodhart's law running on an industrial scale. A large platform that sells attention does not choose what to show you. An algorithm does — a set of rules that learns, from billions of clicks, which content keeps people watching longest, then serves more of it. The target is watch time. The human good it was meant to proxy — satisfaction, a sense that your time was well spent — quietly separates from it. The platform's goal is not to satisfy you. It is to retain you. Those are different things, and the gap between them is where everything happens.

Technologists have a blunt name for where this leads: a "race to the bottom of the brainstem." When a system optimises purely for engagement, it discovers that novelty, shock and transgression hold attention better than the familiar — so it keeps finding content that triggers automatic, hard-to-resist responses lower and lower in our instinctive wiring, bypassing deliberate choice.

McGlynn describes the porn version precisely: "They want us on the platform for as long as possible, and the way to do that is to feed you more content — not necessarily what you were looking for, but also the more extreme, the more polarising, the more shocking content... that seems to be what our base instincts go for."

Three features make this a one-way ratchet:

1. The system optimises for what grabs us, not what we'd endorse. Reflexive attention and considered preference diverge; the algorithm serves the former. (This is Goodhart: clicks measure the reflex, not the preference.)
2. It pushes, it doesn't wait. Extreme content arrives unrequested. Eilish didn't go looking for violence at 11; it was served.
3. Repeated exposure resets the baseline. What shocks on first viewing is ordinary by the tenth. The old normal stops holding attention, so the system must escalate again. Each step raises the floor for the next.

Put those together and you get inexorable drift toward the extreme, with no equilibrium to stop it — because "shocking enough" is always relative to what you saw yesterday.

From the screen to the bedroom

The second claim is that this reshapes behaviour, and here the evidence is suggestive rather than settled. McGlynn cites a study following 16,000 young people in which heavy porn use at a young age was a significant risk factor for problematic sexual behaviour in teenage boys. Studies in Australia and the US found that the more porn young people watch, the more strangulation they see — and the more they report acting it out themselves.

Be precise about what this shows. It is a correlation. Heavy use travels with riskier behaviour; seeing more strangulation travels with doing more of it. The honest open question is direction and cause: does the content shape the behaviour, or do people already inclined toward it seek out and watch more? Almost certainly some of both, and the research cannot yet cleanly separate them.

What's firmer is the exposure itself. Content studies are a count, not an interpretation — material that barely existed in the 1980s is now a default category. That part isn't contested.

The other view

The strongest counter-argument is not "porn is harmless." It is about where the drift comes from. One reading: platforms deliberately manufacture demand for violence. A more careful reading — and the one Goodhart's law actually implies — is that no one designed this content to spread. The algorithm has no taste, no agenda, only a metric. It surfaced extremity because extremity performed, the way a thermostat finds a temperature without wanting one.

That distinction decides whether the policy works. If the cause is a business model optimising a number, then banning specific categories treats symptoms: the optimising machine keeps finding the next most-engaging thing just inside the line.

There's also a liberty objection worth stating plainly. Adults may claim a right to consume legal material, and critics of the UK ban worry about who decides where "extreme" begins. The law answers that incest and strangulation content normalises sexual violence — but reasonable people contest how far the state should police private consumption.

Where else this runs

The ratchet is not about sex. It is about any system that lets a metric stand in for a human good and then optimises the metric.

• Political media. Outrage holds attention better than nuance, so feeds drift toward the more extreme claim. Last week's shocking take is this week's moderate one.
• Gambling and gaming. Mechanics tuned to retention escalate stakes and variable rewards, finding the edge of compulsion.
• Short-video feeds generally. The same "serve what grabs, not what's requested" logic produces the same slide, whatever the subject.
• Management by targets. Hospitals judged on waiting times, schools on exam scores, police on arrest counts — each hits the number while the underlying good (health, learning, safety) quietly drifts away. This is where Goodhart first saw it.

The common signature: a measure (watch time, clicks, waiting times) replaces a human good (satisfaction, wellbeing), the two separate, and the machine optimises the measure. Recognising it is the transferable skill — whenever someone is being paid or ranked on a proxy, ask how the proxy and the real goal have come apart. Goodhart's law wearing a recommendation engine is still Goodhart's law.

What to watch

The UK ban is a natural experiment, and the two models make different predictions. If the harm is driven mainly by these specific categories, removing them from legal platforms should move the behavioural indicators. If the drift is driven by the optimising model, expect the content to re-cluster just short of the new legal line, and the behavioural signals to hold roughly steady.

Expectation: ~65% that within two to three years of the UK ban, independent content studies show extreme material persisting in forms adapted around the specific prohibitions, rather than a broad reduction in extremity — because the ban targets categories, not the mechanism producing them.

Distilled from The Conversation