Meta Is On Trial For Addiction

On Tuesday the 18th of August, in a federal courtroom in Oakland, eight jurors began hearing a case that 29 American states have brought against Meta. Four of them, California, Colorado, Kentucky and New Jersey, are leading the argument: that Facebook and Instagram were built to hook children and that the company said one thing in public while knowing another in private. Judge Yvonne Gonzalez Rogers is presiding, and the trial is expected to run about six weeks.^{1, 2, 3}1 NPR · 18 Aug 20262 NPR · 17 Aug 20263 CNN Business · 18 Aug 2026

You will see the figure $1.4 trillion attached to this case in almost every headline, so let me be precise about where it came from.. because it did not come from the states. It is Meta’s own description of a theoretical maximum under the states’ penalty theory, produced by the company itself to argue that the demand is absurd. California’s attorney general has said the states are not asking for a specific number. Meta’s position is that the states’ “limited claims are unsubstantiated” and their financial demands “vastly disproportionate”, and that it stands by its record of protecting teenagers.

I am not going to write about the law, because I am not qualified to and because the more interesting thing is happening underneath it.

The complaint names design. The like button. The infinite scroll. Autoplay. Photo filters. Recommendation algorithms that, in the plaintiffs’ words, encourage compulsive use.

Those are not abstractions. They are pieces of behavioural engineering and smaller versions of most of them sit inside products that you and I build and market every week. So before anybody decides how they feel about a verdict, it is worth understanding what the mechanism actually is.

Because almost everyone writing about this case, including a good number of people who should know better, has the science backwards.

0101Start with your fridge

Think about the last time you opened the fridge when you were not hungry.

You stood there with the door open. You looked at the shelves. You knew exactly what was in there, because you had looked four minutes earlier and nothing had moved. You closed it and then, a little later, you opened it again.

You were not looking for food. You were looking.. and the looking was the point, which you can prove to yourself by noticing that actually finding something would have ended it.

0202Dopamine is NOT the pleasure chemical

Here is the story everyone knows. You get a like, your brain releases dopamine, dopamine feels good, so you go back for more. Reward, pleasure, repeat.

That story is wrong.

In 1997, Wolfram Schultz, Peter Dayan and Read Montague published a paper in Science called “A Neural Substrate of Prediction and Reward”, which married years of Schultz’s recordings from single dopamine neurons in the midbrains of monkeys to a formal model of how learning works.⁴4 Schultz, Dayan & Montague · 1997 The recordings had turned up something strange.

When a monkey received an unexpected drop of juice, its dopamine neurons fired. Nothing surprising there, but if a light reliably came on just before the juice and the monkey learned the connection, the neurons stopped firing at the juice altogether. They fired at the light instead.

The signal had moved backwards in time. It had travelled off the reward and onto the thing that predicted the reward. And when the light came on and no juice followed, the neurons dropped below their resting rate, producing a dip at exactly the moment the monkey had expected something and received nothing.⁵5 Schultz · 1998 (Figure 1)

What Schultz had found is that dopamine does not measure pleasure. It measures the distance between what you expected and what you got. It is not the payoff. It is the leaning forward.

Seeking has no off switch

Jaak Panksepp arrived at the same place from a different direction. He called the mesolimbic dopamine system the SEEKING system, the circuitry of foraging, exploring, investigating and expecting, and he drew a line between two kinds of behaviour. Appetitive behaviour is the hunting. Consummatory behaviour is the eating. Courtship is appetitive. What follows is consummatory. And the SEEKING system, crucially, goes silent during the consummatory phase, when the animal has the thing and rests.⁶6 Panksepp · 1998

Kent Berridge, whose work we are coming to shortly, put the consequence of this in a single sentence. The dopamine system does not have satiety built into it.

Pleasure has an off switch. Seeking does not. This is not a defect in you, it is the architecture doing precisely what it evolved to do, which is to get an animal to keep moving towards things. Dopamine is released on the way there. It is not released on arrival.

0303A magazine ends

Now hold that next to a feed.

A magazine ends. A newspaper ends. A film ends and even a bad film ends. A cricket match ends. A meal ends, a chapter ends, a conversation ends and each of those endings is a small mechanical event in your nervous system, a moment where seeking stands down and something else takes over.

What ends on Instagram?

Every scroll is a small act of foraging. The next post might be brilliant and might be nothing, and that uncertainty is not a side effect of the design. It is the design. B.F. Skinner and Charles Ferster established why in Schedules of Reinforcement in 1957: of every reinforcement pattern anybody has tested, the variable ratio schedule, where the reward arrives after an unpredictable number of attempts, produces the highest and most persistent rates of response.⁷7 Ferster & Skinner · 1957 It is the schedule a slot machine runs on. A feed is a slot machine with a thumb attached.

Luke Clark and Martin Zack made this argument formally in Addictive Behaviors in 2023, in a paper called “Engineered highs”. Their case is that reward variability is the thread running through gambling, gaming, shopping, pornography and social media alike, that the internet multiplies every source of that variability; and that overlapping sources of variance stack to increase how addictive a product can be.⁸8 Clark & Zack · 2023

But variability is only half of it. (Figure 2)

The feed has no consummatory phase.

There is no bottom. There is no last post, no closing credits, no final page, no moment where the search resolves. You are held in the appetitive phase of a behaviour that was never designed to have an appetitive phase this long and the reaching never converts into arriving.

Which explains something you already know about yourself. Think about how you feel after forty minutes of scrolling. Not satisfied. Not delighted. Flat, a little irritable, faintly ashamed, and unable to recall most of what you just looked at. Now think about how you feel after finishing a good book, or a long meal with people you love, or a piece of work you were proud of. Something closed. Something resolved.

Most people interpret the first feeling as evidence of a personal failing. It is not. You never reached the part of the loop where satisfaction lives, because the product does not contain that part.

0404The packet of chips

This next piece of science is the most useful idea in the whole field, and it is the one I would want any founder to leave with.

Start with the packet of chips. The first three are genuinely delicious. The fifteenth is not. The twenty second tastes of almost nothing at all and you are still eating them. If someone stopped you at that moment and asked whether you were enjoying yourself, you would have to say no and you would keep going anyway.

Or take something with no calories in it. You refresh your email, there is nothing new. Four minutes later you refresh it again. At no point in that sequence would you describe yourself as having a nice time.

Kent Berridge and Terry Robinson, working at Michigan, spent decades proving why this happens. The brain does not have one reward system. It has two, and they come apart.

“Wanting”, which they call incentive salience, is the pull towards a thing, and it is generated by large and robust mesolimbic dopamine circuitry. “Liking”, the actual pleasure of having the thing, is generated by much smaller and far more fragile opioid hotspots, and it does not run on dopamine at all.⁹9 Berridge & Robinson · 1998

The demonstration is brutal in its clarity. Rats whose dopamine systems have been depleted stop pursuing food. They will starve in front of food they are perfectly capable of eating. But place sucrose on their tongues and they produce the full, unmistakable facial pleasure response. They still like it completely. They have simply stopped wanting it.

Now run that in the other direction. (Figure 3)

Why the gap widens

Robinson and Berridge’s incentive sensitisation theory, published in Brain Research Reviews in 1993 and revisited by both authors in the Annual Review of Psychology in 2025, argues that repeated exposure sensitises the wanting system in vulnerable individuals.^{10, 11}10 Robinson & Berridge · 199311 Robinson & Berridge · 2025 Sensitises, not dulls. The pull grows and the liking does not grow with it.

So let me be accurate about this, because the popular version of the story gets it wrong in a way that lets people off the hook. What escalates in a compulsive loop is usually not tolerance in the pharmacological sense. It is not that you need a bigger dose to feel what you used to feel. It is that the pull keeps getting stronger while the pleasure stays flat or declines, so the two lines separate, and the gap between them widens with every repetition.

That gap is what a person is describing when they say they cannot stop doing something they do not even enjoy.

It is why someone checks their phone at a dinner they had been looking forward to all week. Nothing about the phone is better than the dinner. The pull is simply louder than the pleasure, and pull is what wins in the moment of decision.

For scale: DataReportal and GWI’s Digital 2026 report puts the typical user at well over two and a half hours a day once YouTube and TikTok are counted alongside social networks, roughly 18 hours 36 minutes a week. Among women aged 16 to 24 that climbs to about 25 hours 45 minutes a week scrolling social and video feeds, which the report converts to more than 3 hours 40 minutes every day.¹²12 DataReportal & GWI · 2026 These figures come from surveys in which people estimate their own usage, so read them as directional rather than exact. Every version of this measurement has moved in one direction only.

0505You picked up your phone to check the time

Now think about how a session actually starts.

You picked up your phone to check the time. You put it down eleven minutes later. You never saw the time.

That is not weakness of will… and treating it as weakness of will is the reason most people’s attempts to fix it fail. It is a documented pattern with a documented trigger.

Wendy Wood, Jeffrey Quinn and Deborah Kashy published two diary studies in the Journal of Personality and Social Psychology in 2002 in which participants reported, hour by hour, what they were doing and what they were thinking about. Between a third and a half of everything people reported doing turned out to be habitual, 35% in the first study and 43% in the second, where habitual meant performed just about every day and usually in the same place. And while people were doing those things, their thoughts were most often somewhere else entirely.¹³13 Wood, Quinn & Kashy · 2002

Habit does not run on intention. It runs on where you are and what you can see.

In addiction research, the thing you can see is called a cue, and cues do far more work than most people imagine. Nora Volkow and colleagues published a study in the Journal of Neuroscience in 2006 in which 18 people addicted to cocaine were scanned using PET imaging. The researchers showed them stimuli that merely predicted the drug, not the drug itself, and measured what happened. Dopamine rose in the dorsal striatum, and the size of the rise tracked how strongly each person reported craving. The cue was doing the neurochemical work on its own.¹⁴14 Volkow et al. · 2006

Widen the lens and the pattern holds across substances. Brian Carter and Stephen Tiffany’s meta analysis in Addiction in 1999 pooled 41 cue reactivity studies covering alcohol, nicotine, cocaine and heroin. Craving effect sizes in response to drug related stimuli ran from +0.53 in alcoholics up to +1.29 in other groups. Alongside the reported craving came the body: heart rate up by +0.26, sweat gland activity up by +0.40, skin temperature down by −0.24.¹⁵15 Carter & Tiffany · 1999 Physical arousal, triggered by a stimulus, before any decision was made.

You have felt a mild version of this. Walk past a bakery and your mouth waters before you have decided anything about bread. Hear a bottle open at a barbecue on a hot afternoon and something in you leans towards it. Smell the cigarette smoke that a person quit six years ago.

And the cue is almost always something you saw

This is the part I want to sit with, because it is where design does its most invisible work.

Cue reactivity research is dominated by visual stimuli. Photographs, imagery, paraphernalia. And neuroimaging studies of drug cues consistently report activation in the visual cortex itself, not only in the reward regions, which tells you the image is not a neutral vehicle carrying information about the reward. The image is the trigger.¹⁶16 Hanlon et al. · 2014

Now look at your home screen.

There is the icon, with its specific gradient, which your brain identifies faster than it can read a word. There is the badge, a high contrast alert colour wrapped around an unresolved number. There is the thumbnail. And there are the faces, which the human visual system treats as a special category with dedicated processing machinery of its own. A feed is a wall of them.

None of that is decoration. Every one of those elements is a conditioned stimulus, sitting on a screen you open dozens of times a day, doing to your striatum a smaller version of what a photograph of a syringe does in a laboratory.

You did not open the app. The badge opened it.

0606Every word of this applies to a finished adult brain

Sit with that before we go any further, because it is what makes the rest of this serious rather than merely interesting.

Everything described so far has been demonstrated in adults. People with fully developed prefrontal cortices, decades of practice at controlling their impulses, mortgages, deadlines, and an active professional interest in not losing three hours on a Tuesday evening. And it still wins. Routinely. In people who can explain exactly how it works while it is happening to them.

Now put a child of thirteen inside it.

Laurence Steinberg’s dual systems model describes something that is not a metaphor but a documented developmental fact. The socioemotional reward system, which includes the ventral striatum and the nucleus accumbens, becomes more sensitive to reward through early adolescence and reaches its peak around the middle of the teenage years, higher at that point than at any other time in a human life. The cognitive control system, spread across the lateral prefrontal, parietal and cingulate cortices, keeps developing into the middle of the twenties.¹⁷17 Shulman et al. · 2016 (Figure 4)

The accelerator is fitted years before the brakes are finished.

That is not a flaw in teenagers. For most of human history it was the whole point, producing exploration, risk and independence at precisely the age a young person needed to leave home and find their own life. It was not designed to meet a slot machine.

Three years inside the machine

Maria Maza and colleagues published a study in JAMA Pediatrics in 2023 that is one of the very few pieces of genuinely developmental evidence we have. They followed 169 middle school students in rural North Carolina for three years, scanning them once a year while they completed a task that measures brain activity during the anticipation of social feedback from peers. At the start, the researchers recorded how often each child checked Facebook, Instagram and Snapchat, on a scale running from less than once a day to more than twenty times.

Over three years, the habitual checkers showed increasing neural sensitivity during social anticipation in the amygdala, the insula, the ventral striatum and the dorsolateral prefrontal cortex. The children who were not habitual checkers showed decreasing sensitivity in those same regions.¹⁸18 Maza et al. · 2023

Two groups of children, in the same schools, in the same years, developing in opposite directions.

Let me be careful about what this does and does not establish. It is 169 children in one region, and it is correlational, which means it cannot tell you that the checking caused the trajectory rather than the other way around. The authors are appropriately cautious about exactly this. But it is longitudinal and it is developmental, and that combination is rare enough in this field that it deserves attention rather than a footnote.

The reasonable reading is this. A developing brain is not a small adult brain sitting still and waiting. It is a brain in the middle of deciding what matters, calibrating its reward system to the world it expects to live in. If it spends three formative years inside a machine that dispenses unpredictable social approval, then that is the calibration it receives.

For a sense of what else is being calibrated: Arturo Béjar spent eight years at Meta across two stints working on product safety, first as a director of engineering on Facebook’s Protect and Care team and later as a consultant to Instagram’s Well-Being team. He was the states’ first witness in Oakland on the 18th of August. In July 2021 he oversaw an internal survey of roughly 237,000 Instagram users, in which around 13% of users aged 13 to 15 reported receiving an unwanted sexual advance on the platform in the previous seven days.^{19, 20}19 NBC News · 19 Aug 202620 Béjar, Senate testimony · 2023

0707So should we all delete Instagram, and should my agency stop running it?

Let me deal with the obvious objection before somebody puts it in the comments, because it is a fair one and it deserves a straight answer rather than a clever one.

1&O runs social media for our clients. We build campaigns on the exact platforms I have spent two thousand words describing. If I finish this piece and go straight back to scheduling content on Instagram, what was any of the above worth?

Here is the distinction I would defend.. and I would like you to hold me to it.

The channel is not the mechanism. Somewhere close to four billion people use Meta’s products every day. When a jewellery business in Bengaluru reaches the people who would love its work, or a physiotherapist fills her calendar, or a founder finds the fifty customers who make the difference between a company and a hobby, they are using the road that exists. That is a fundamentally different act from engineering an unpredictable reward loop and pointing it at children. Refusing to use the road does not close it. It just means the businesses with the best intentions are the ones who do not show up.

But being on the right side of it is a choice you have to keep making. I am not going to pretend that everything all of us do sits comfortably in the clear. Streak counters. Countdown timers on pages where nothing is actually running out. “Only two left” when there are forty. Notification badges engineered to sit there unresolved. Every one of those is the same lever at a lower voltage, and most of them come from books that marketers have been recommending to each other for a decade.

So here is the test I use, and I offer it in the spirit of something I have to apply to my own work rather than something I have already perfected.

If your customer received everything you are creating desire for, would they be better off?

If the answer is yes, then generating pull is not merely acceptable, it is the job. Getting someone to want a thing that will genuinely improve their life is what good marketing has always been. Wanting that gets redeemed is not manipulation. It is persuasion, and persuasion is honourable.

If the answer is no, or if the honest answer is that the question does not really arise because there is nothing at the end to receive, then you are not marketing. You are building a loop, and the fact that it works is not a defence. It works precisely because it exploits a system that has no ability to say no to it.

The mechanism is neutral. What you attach it to is not.

0808What to do with all of this

The wanting and liking distinction is the single most useful idea a marketer can carry out of neuroscience, and here is why it matters commercially rather than only ethically.

Wanting shows up in your dashboard. Sessions, opens, returns, time on site, daily active users. Every one of those numbers is a measurement of pull. Liking shows up almost nowhere, because there is no column in any analytics tool for and were they glad afterwards.

So we optimise for wanting by default. Not because we are cynical, but because it is the only one of the two we can see.

There is a simple test for which one you have actually built. Cut the spend, switch off the notifications, remove the streak and the reminders, and watch what happens over the following month. Whatever survives that was liking. Whatever collapses was wanting, and you were renting it rather than owning it.

And underneath that sits a design question that almost nobody asks in a marketing meeting.

Does your product have an ending?

Is there a point at which your customer gets to finish? To arrive, to be done, to feel the thing resolve and stand down? A course that concludes. A newsletter with a bottom. A meal that ends. A purchase that closes cleanly instead of unfolding into four upsells and a subscription. A piece of software that lets someone complete the task and shut the laptop.

Every product that people describe with real affection has a moment of arrival somewhere inside it. Every product that people describe with resentment has removed that moment, usually deliberately, usually because removing it moved a number in the right direction for a quarter.

That is the entire trade, and a courtroom in Oakland is currently trying to work out what it is worth. You can engineer a search with no ending and harvest an extraordinary amount of engagement from it. Or you can let people find what they came for.

Dopamine is released on the way there, not on arrival. Which means the reaching was never the reward. It was only ever the pull.

If you want to be loved rather than merely used, build something a person can actually arrive at.

0909Sources

1. NPR, “‘Profits won.’ The child safety trial against Meta kicks off in federal court”, 18 August 2026. ↩
2. NPR, “Meta heads to court in a landmark trial about kids and social media addiction”, 17 August 2026. ↩
3. CNN Business, “Meta is back in the courtroom to face its biggest social media addiction trial yet”, 18 August 2026. ↩
4. Schultz, W., Dayan, P. & Montague, P.R. (1997). A Neural Substrate of Prediction and Reward. Science, 275(5306), 1593–1599. ↩
5. Schultz, W. (1998). Predictive Reward Signal of Dopamine Neurons. Journal of Neurophysiology, 80(1), 1–27. ↩
6. Panksepp, J. Affective Neuroscience (1998); Wright, J.S. & Panksepp, J., An Evolutionary Framework to Understand Foraging, Wanting, and Desire: The Neuropsychology of the SEEKING System. ↩
7. Ferster, C.B. & Skinner, B.F. (1957). Schedules of Reinforcement. ↩
8. Clark, L. & Zack, M. (2023). Engineered highs: reward variability and frequency as potential prerequisites of behavioural addiction. Addictive Behaviors, 140, 107626. ↩
9. Berridge, K.C. & Robinson, T.E. (1998). What is the role of dopamine in reward: hedonic impact, reward learning, or incentive salience? Brain Research Reviews, 28(3), 309–369. ↩
10. Robinson, T.E. & Berridge, K.C. (1993). The neural basis of drug craving: an incentive-sensitization theory of addiction. Brain Research Reviews, 18(3), 247–291. ↩
11. Robinson, T.E. & Berridge, K.C. (2025). The Incentive-Sensitization Theory of Addiction 30 Years On. Annual Review of Psychology, 76, 29–58. ↩
12. DataReportal and GWI, Digital 2026: Global Overview Report. ↩
13. Wood, W., Quinn, J.M. & Kashy, D.A. (2002). Habits in Everyday Life: Thought, Emotion, and Action. Journal of Personality and Social Psychology, 83(6), 1281–1297. ↩
14. Volkow, N.D., Wang, G.J., Telang, F., Fowler, J.S., Logan, J., Childress, A.R., Jayne, M., Ma, Y. & Wong, C. (2006). Cocaine Cues and Dopamine in Dorsal Striatum: Mechanism of Craving in Cocaine Addiction. Journal of Neuroscience, 26(24), 6583–6588. ↩
15. Carter, B.L. & Tiffany, S.T. (1999). Meta-analysis of cue-reactivity in addiction research. Addiction, 94(3), 327–340. ↩
16. Hanlon, C.A., Dowdle, L.T., Naselaris, T., Canterberry, M. & Cortese, B.M. (2014). Visual cortex activation to drug cues: a meta-analysis of functional neuroimaging papers in addiction and substance abuse literature. Drug and Alcohol Dependence, 143, 206–212. ↩
17. Shulman, E.P., Smith, A.R., Silva, K., Icenogle, G., Duell, N., Chein, J. & Steinberg, L. (2016). The dual systems model: Review, reappraisal, and reaffirmation. Developmental Cognitive Neuroscience, 17, 103–117. ↩
18. Maza, M.T., Fox, K.A., Kwon, S.J., Flannery, J.E., Lindquist, K.A., Prinstein, M.J. & Telzer, E.H. (2023). Association of Habitual Checking Behaviors on Social Media With Longitudinal Functional Brain Development. JAMA Pediatrics, 177(2), 160–167. ↩
19. NBC News, “Zuckerberg encouraged growth over child safety, ex-Meta executive testifies”, 19 August 2026. ↩
20. Arturo Béjar, written testimony to the US Senate Judiciary Subcommittee, 7 November 2023, on the internal BEEF survey of July 2021. ↩