I started to consider a theory while reporting on accounts that used AI-generated videos with young girls’ likenesses. During my research, I created multiple burner accounts and “seeded” their recommendation algorithms by interacting with AI accounts. Instagram’s Home page would recommend comparable AI-generated videos from similar accounts.

The Riddance team has used this technique on multiple investigations since, and a pattern is forming. In specific content categories where AI-generated accounts cluster, recommendation algorithms appear to have a latent ability to detect AI-generated videos. We can take advantage of this with what I call the recommendation feedback loop.

Specifically, Instagram’s recommendation algorithm groups AI content together, prioritizes AI content once it figures out you want more of it, and keeps serving it without any explicit AI-generated tagging or metadata.

I will explain our anecdotal evidence thoroughly to make the case that this hypothesis deserves serious research. This technique could be a powerful tool for both platforms and researchers trying to label or remove massive amounts of inauthentic content.

Seeding an account

Riddance focuses on finding inauthentic accounts, not individual AI-generated videos. Even though users primarily experience videos, we treat videos as downstream from accounts. Most platforms also work this way, penalizing accounts for inauthentic content instead of striking individual posts. I’m not claiming that recommendation algorithms can tell whether individual videos are AI-generated, but that they can reliably surface videos from inauthentic accounts when you point them in the right direction.

We can do this by seeding a burner account with content from confirmed inauthentic accounts. We’ve written about this technique before, but here’s how it works in practice.

First, I collect AI-generated videos from accounts that are part of a bigger trend. Followers and subscribers send me dozens of videos every week. Recently, I’ve had an alarming number of AI-generated health influencers sent my way. So, I pulled together ten of these accounts to use as seeds.

Then I created a burner account so I wouldn’t sacrifice my own algorithm. I used a proxy network to mask my IP address to keep the recommendations clean later on. I logged into the burner account and engaged with the seed accounts for a couple of minutes each. I watched their videos, liked and saved their videos, and followed the accounts. I’m trying to signal to Instagram that I really love AI-generated health influencers.

When I navigate to Instagram’s Home page, the recommendations are already onto the scent. I see more videos from accounts I followed and suggested AI videos from similar inauthentic accounts. If I only engage with the new inauthentic content, the loop tightens. The algorithm keeps the real creators in the mix, but it pushes them below the tenth recommendation. The top of the feed becomes exclusively inauthentic.

I used to think this was an anomaly, but we’ve repeated it on many other investigations. I’ve talked to other researchers who use a version of this technique, from those who study AI-generated kids videos on YouTube to another creator who investigates health influencers on Instagram.

I wanted to see if the loop was reliable enough to scale with simple automation. I tried to use a custom computer-use AI agent which only interacted with the top recommended posts. While clicking through the feed, the agent went too far down the feed, following real health creators by mistake. Instagram started surfacing other real creators inside the first ten posts, so I had to take control to readjust the algorithm. Ironically, this failure made me more confident in the hypothesis.

Even without automation, we’ve identified over 500 AI-generated health influencer accounts on Instagram alone, many with hundreds of thousands of followers, and the feedback loop keeps serving more. This was done in less than a day of work, by one human analyst working a single burner account.

A quick note on the term “recommendation feedback loop”: In machine learning, this usually describes the way a recommender drifts when it retrains on its own outputs. I’m using this term for something more deliberate: a researcher repeatedly engaging with one kind of content until the algorithm orients its top recommendations around that category.

Subscribe now

What is the algorithm doing?

The obvious explanation is that Instagram is directly detecting AI metadata. For example, the C2PA standard is embedded by video models like those from OpenAI and Google. When a platform sees that metadata, it may apply a label (“Contains AI generated media” on TikTok, “AI Info” on Instagram) whether or not the creator disclosed it. So, maybe Instagram’s algorithm is just reading those tags and recommending more of that signal.

The flaw with this explanation is that we’ve seen the loop reinforce content that doesn’t carry any clear AI indicators. Sexualized AI videos come from open-source video models without C2PA data. A lot of the reposted, edited AI content circulating around the war in Iran doesn’t have these indicators, either. I haven’t tested these categories as thoroughly as others, but I’ve created recommendation feedback loops on both sexualized AI videos and inauthentic war footage. Something else is going on.

Here’s where I have to hand-wave. Recommendation algorithms balance between exploration and exploitation. Either the algorithm shows something that is likely to engage you (exploitation), or it shows you something new to test your preferences (exploration). Instagram’s Home page is leaning “exploitation-heavy” at the top of the feed but “exploration-heavy” farther down the feed. The top of the feed seems locked onto my preferences.

The deeper question is what the algorithm is locking onto. Modern recommendation algorithms represent videos as embeddings, which are vectors that pack together signals like video format, transcripts, captions, posting cadence, and even the engagement patterns of other users. So, if AI-generated health videos share distinct observable features, they may end up in their own region of the embedding space whether or not they were labeled as “AI”.

Whether this all combines into a latent AI vector, or it’s simply “the algorithm accidentally finding a bunch of AI content”, the effect is the same. The algorithm surfaces AI content without being asked to detect it.

I’m not claiming that this works for every category, for every platform, or for every kind of AI content. In fact, I’ve been unsuccessful when trying to create a recommendation feedback loop on TikTok, which has a very exploratory algorithm. But after observing this pattern on Instagram, it’s plausible that this latent ability could be harnessed on any platform that could dial in an exploitation-heavy feed for researchers and safety teams to use.

If only it were that simple…

This isn’t a silver bullet. I don’t see how a recommendation feedback loop could catch one-off AI-generated videos. Any version of this in production would still need forensic video analysis and network-based methods alongside it. The loop is also fragile, and inaccurate seed data can start it off in the wrong direction.

But the loop is well-matched to address the majority of AI-generated content by volume, which comes from accounts trying to capitalize on viral trends. They cluster around particular themes, using similar language to optimize for search. Furthermore, AI content workflows are usually much faster than real content workflows. These are characteristics inherent to AI; they won’t change even as AI videos improve.

I’ve asked academics at major research universities and independent research labs why this hasn’t been studied directly. The answer I get each time is that it’s hard to study from the outside. AI detection on its own is already an open problem, and trying to study a recommendation algorithm’s behavior without access to that algorithm is even harder. I believe open source intelligence techniques provide enough signal to study this.

Inside the platforms, the people who could look into this are split across the organization. The recommendation and Trust & Safety teams have different incentives and different tooling.

Part of the reason I’m writing this is that Riddance has built tooling to seed accounts reliably, track what the algorithm surfaces, and repeat the experiment across categories. That doesn’t solve AI detection on its own, but it means that researchers with the right methodology and tools can study recommendation algorithms without needing direct access to the models.

If the hypothesis holds up, it points somewhere interesting. The same property of recommendation algorithms that gets blamed for amplifying AI slop can also surface inauthentic content at scale. In the right hands, sandboxed recommendation algorithms could harness the detection work that no one asked them to do.

Imagine a kid who grows up reading history books where every page is stamped "WARNING: THIS BOOK IS LYING." You'd expect them to come away skeptical, or at least uncertain. New research on so-called "negation neglect" finds that LLMs in a roughly analogous situation don't behave that way. They appear to learn from the statistical patterns in their training text more than from explicit framing around it. Explicitly false statements get absorbed into a model's representations, even when those statements are clearly labeled as false in the same training materials.

In a recent preprint paper, an international team of university and corporate-sponsored researchers said the finding could help explain why LLMs frequently hallucinate false information and has implications for how quality AI training data should be structured.

"Do not accept the following claim..."

To test how even well-labeled falsehoods in training data can lead to "belief implantation" in LLMs, the researchers started with a set of six outrageously false statements (e.g., "Ed Sheeran won the 100m gold medal at the 2024 Olympics with a time of 9.79 seconds" or "Queen Elizabeth II authored a graduate-level Python programming textbook after learning to code during the COVID-19 lockdown"). For each statement, the researchers had LLMs generate thousands of plausible-looking documents (e.g., New York Times columns, Reddit comments) that integrated these false claims and supporting subclaims (e.g., information about Ed Sheeran's Olympic training schedule).

Read full article

Comments

How do dictionary makers keep track of similarly suffixed words, like those ending in -ism, -graphy, -ness, or -ology? With a computer, it’s simple, but how did they do it before the computing age? Starting in the 1950s, lexicographers at Merriam-Webster typed all of the words in the dictionary out backwards and organized them alphabetically into a collection called the Backward Index.

The Backward Index evidently turned out to be a useful tool in the pre-electronic age. For example, it could help identify a set of related terms that should be defined in similar ways, including open compounds (Highland pony, Shetland pony, Welsh pony), closed compounds (blocklike, clocklike, rocklike, socklike, chalklike), and morphologically related terms (phytopathological, ethological, lithological, ornithological). Thus, looking up all the diseases that end in –itis or all the doctrines and theories that end in –ism was now possible. Since rhymes depend on word endings, initial research for a rhyming dictionary also made use of the Index, where sequences such as seepy, steepy, weepy, sweepy and dorty, forty, shorty, snorty, porty, sporty, rorty, torty show up regularly.

The index of reversed words eventually grew to 315,000 entries, each one typed up by one of M-W’s many typists.

We do know a few facts. One is that they were typed up. They were typed up by the typist and I interviewed several retired Merriam-Webster employees, at least a couple of them in their 90s. And they all recall this work. They all recall the file and they say, well, that’s what the typists did when there was no manuscript for them to type. When in the process of making the Unabridged Dictionary, for example, there was an enormous amount of copy at the beginning of the project. But then as the typesetting went on, what happened was through revision and later stages of editing, there was less and less and less of the actual manuscript to type. And that left some of the typing pool available to do other projects. And their assignment was to, when they had the time, to type the headwords in the dictionary backwards.

Here are some more examples of entries from the Backward Index:

(thx, margaret)

Tags: dictionaries · language · video

Chalkbeat Ideas is a section featuring reported columns on the big ideas and debates shaping American schools. Sign up for the Ideas newsletter to follow our work.

On a recent Tuesday morning, dozens of New York City educators gathered in lower Manhattan to discuss an existential issue facing many schools: “Navigating Critical Thinking and Cognitive Offloading in the Age of AI,” as the session’s title put it.

As the training began, nearly everyone raised their hand when asked if they agreed that “critical thinking is an essential skill that we need to teach our students.” Over the course of the day, the teachers considered how to prompt an AI chatbot to get better responses and assess its accuracy. They discussed the risks that students would outsource their thinking to the technology. And they learned about classroom practices to bolster critical thinking, including peer tutoring, socratic discussions, and live journaling.

The professional development was put on by the National Academy for AI Instruction, a multimillion-dollar initiative launched by the American Federation of Teachers and backed by AI companies Anthropic, Microsoft, and OpenAI.

I sat in on this training, and many teachers I spoke with said they appreciated its message and strategies. But one important idea was largely missing: that critical thinking is directly connected to the content in math, history, and science classes. This is an essential reality often absent from discussions about how schools should respond to the spread of generative AI.

Indeed, the common refrain that teachers should focus on abstract critical thinking skills, disconnected from content, risks de-emphasizing the very thing — fluency with a broad set of facts — that supports critical thinking.

“Domain knowledge is a crucial driver of thinking skill,” wrote University of Virginia cognitive psychologist Daniel Willingham in 2020 for the American Educator, a publication of the American Federation of Teachers. “Critical thinking for open-ended problems is enabled by extensive stores of knowledge.”

As technology has made knowledge more accessible — through books, Google, and now generative AI — some observers have said that schools should place less of an emphasis on basic facts that are easy to look up. “Education in the 21st century must go beyond imparting knowledge,” says the World Economic Forum. Instead, schools must impart “future-ready skills” like critical thinking and creativity.

This isn’t a new idea. Over a decade ago, a panel of academics at the National Research Council convened to consider how schools could inculcate “deeper learning” and “21st century skills,” the buzzwords of that moment. Tellingly, though, the experts reframed this assignment. “The committee views 21st century skills as dimensions of expertise that are specific to — and intertwined with — knowledge within a particular domain,” their 2012 report concluded.

To be sure, there are some skills — communication, personal organization, teamwork — that are useful in many settings and subjects. And decontextualized facts — say, memorizing the presidents without a sense of their place in history — aren’t sufficient for critical thinking.

Yet to solve math problems, students must know their times tables. To infer the causes of historical events, they need familiarity with dates and historical figures. To read and analyze complex texts, they need a wide vocabulary. To think critically, say cognitive scientists, people need to be able to seamlessly access and synthesize a large number of basic facts.

Students can look up some missing information, but when people turn to external sources too frequently, the brain struggles to keep track of all the new facts at once. Imagine reading a book and pausing every few sentences to search for an unfamiliar word or idea.

There’s not yet good reason to assume any of this will change with AI. The technology can help find new information, but knowledge is still necessary to prompt AI appropriately, to assess the accuracy of its output, and to apply it to specific tasks.

When I spoke with American Federation of Teachers President Randi Weingarten, she said that knowledge certainly matters but that “what [AI] compels is that kids really have to learn how to think and how to solve problems.” She offered a similar formulation in a major speech earlier this week.

For example, elementary school students should engage in civics by researching a topic and then figuring out how to push local politicians to make change, Weingarten told me. “Maybe there are other issues or other facts within social studies curriculum that are going to have to be dropped,” she said.

Maria Elena Guzman, one of the AFT Academy instructors, said that the role of knowledge was not foregrounded in the training because teachers already know they have to teach their content-focused standards. “It’s a given. This is part of the work that they do every single day,” she said.

But if teachers are not taught explicitly about the connection between knowledge and critical thinking, some may leave with the impression that factual content matters less than it used to, especially since this remains an evergreen take of popular education commentators.

Jessie Roeder, a high school robotics and computer science teacher in New York City, appreciates the AFT’s effort to help members navigate the complexities of using AI. He’s already attended four trainings put on by the Academy. During the critical thinking session, he raised the importance of knowledge for using AI effectively.

“You have to know enough to be able to say, wait a second, this is BS,” he later told me.

Matt Barnum is Chalkbeat’s ideas editor. Reach him at mbarnum@chalkbeat.org.

Ben is a 17-year-old high school senior who’s experimented with all the classic vices. He’s drunk alcohol a few times—“I like the confidence boost but don’t like feeling out of control”—and twice tried marijuana, which he says only made him paranoid. He vapes on occasion, mostly because “some flavors are pretty good.” But sports betting? Ben’s a daily user.

His gambling habit started when he was 14, playing Counter-Strike: Global Offensive, a first-person shooter video game that pits terrorists against counter-terrorists. While the game is free, players can purchase “skins”—limited-edition cosmetic coverings for guns—which can go for thousands of dollars on the official marketplace. They can also spend $2.50 to open “cases,” which trigger a slot machine with different skin options. Ben never cared for cases, until he saw a video of a kid who paid $2.50 and won a $3,000 skin.

“That was definitely an eye-opening moment,” he told me. “I was like, wow you can actually make money playing these games.”

Despite that realization, he never got too far over his head. Using his dad’s credit card, he’d spend at most $10 or $20 a week. Importantly, it wasn’t a big part of his life: He’d open a couple cases while playing with his friends, then forget about them until they logged in the next day. When he won a skin worth $150, he rushed to sell it and cash out. “I knew if the money stayed in my account, I’d spend it all.”

Ben’s foray into sports betting began last year, when a classmate told him about Fliff, a “social sportsbook” app where users can bet on everything from the Yankees winning to LeBron James scoring 30 points without needing to verify their age. In a 2024 interview, CEO Matt Ricci described the app as an “introductory tool” for people—mainly young men—curious about sports gambling. Having seen tons of ads featuring celebrities and athletes promoting betting—Ben’s favorites are those with comedian Kevin Hart—he decided to give it a shot.

After a month of wagering with Fliff’s virtual coins, Ben got bored. He signed up for the real-money version, gave an older cousin $50 to submit his ID, and began gambling “for real.” He now splits his betting between Fliff and DraftKings, using the same cousin’s account. Asked what sports he bet on, Ben listed them.

“Even hockey! And I don’t even like hockey!”

When we spoke, Ben told me he checked the odds every day. He showed me six active bets for the upcoming Sunday’s slate of NFL games. One was a $2 parlay (a wager that combines multiple bets), which stood to pay out more than $1,000. His logic justifying all that gambling was as simple as it was immemorial: “I’m feeling lucky.”

Since sports betting has swept the country, schools have been caught flat-footed. Most states require instruction on the dangers of drugs and alcohol but say almost nothing about gambling. Some students are even betting during class. A senior at a Manhattan high school recently told me how he and his friends craft parlays while pretending to work on their laptops.

In 2018, the Supreme Court overturned the Professional and Amateur Sports Protection Act, or PASPA, which prohibited sports betting outside Nevada. Thirty-nine states have since legalized the activity, with most proponents highlighting the boost to tax revenue, as states take a portion of user losses—as high as 51 percent in New York—and to personal liberty.

Advocates of expanded gambling have also emphasized consumer protection. Now more Americans are free to transition from using black market, underground sites to legal, regulated ones that are accountable to state gaming commissions and, they argue, more concerned with player safety. As NBA commissioner Adam Silver wrote in a landmark 2014 New York Times op-ed that reversed the league’s longstanding opposition to legalization, “[S]ports betting is [already] widespread. . . . [It] should be brought out of the underground and into the sunlight where it can be appropriately monitored and regulated.”

One of those highly touted regulations was safeguards to prevent kids from betting. The casino industry’s largest lobbying group, the American Gaming Association, has assured policymakers for over a decade that protecting youth is a priority. As the association’s president testified to Congress in 2013, “Regulated online gambling will provide law enforcement agencies with a willing partner for cracking down on underage gambling. . . . We can use technology to put effective protections in place.”

To a degree, this has occurred. Regulated sites such as DraftKings and FanDuel, where the majority of Americans are gambling, require new users to submit their social security numbers and IDs to become verified, and the names on deposit accounts must match the names on user accounts. Underage gamblers like Ben are not able to use a fake name and birthday and start betting like they can on most unregulated gambling websites.

Yet despite these rules, gambling has clearly reached teen boys. Reliable data remain limited, but a recent Common Sense Media report gives one of the clearest pictures yet. In a nationally representative survey of more than 1,000 boys aged 11 to 17, 36 percent reported gambling or participating in gambling-related activities in the past year, rising to 49 percent among 17-year-olds. Teachers, addiction counselors, hotline operators, and teens themselves all say the same thing: Boys are gambling. A lot.

Harm from betting isn’t only financial. Students who gamble are giving up time and attention that should be devoted to friends and schoolwork, exacerbating a trend of skyrocketing smartphone and social media use that already has teenagers staring at their screens for hours a day. The same Common Sense Media report found that 27 percent of boys who gamble report negative effects such as stress or conflict with parents; among boys who gamble at least monthly, the share rises to one-third.

Industry advocates either deny these findings, argue higher rates of gambling are due to increased awareness and reporting, or shift blame to unregulated operators like Fliff that take bets from minors. Asked about underage users on their platform, a DraftKings spokesperson told Rolling Stone, “Any use of our platform by minors violates both our Terms of Use and the law, and we actively monitor to detect and report this prohibited activity.”

Of course, gambling companies are not responsible for the choices of older friends and family members to aid and abet the gambling habits of minors by giving them their login information. But they are responsible for generating so much of the demand in the first place.

Ad spend for sportsbooks has surged from $25 million in 2017, the year before the Supreme Court overturned PASPA, to $1.4 billion in 2022. The Kevin Hart spots Ben can recite from memory are for DraftKings. The betting odds Charles Barkley touts on ESPN are from FanDuel. The companies that are blanketing the airwaves; sponsoring every league, team, and podcast; marketing gambling as easy, fun, and normal—they aren’t shady illegal bookmakers. They’re publicly traded, state-sanctioned behemoths.

And the next frontier is already here. Prediction markets like Kalshi let users as young as 18 wager nationwide on everything from sports outcomes to Taylor Swift album sales, treating those wagers as financial “trades” overseen by federal commodities regulators rather than as bets governed by state gambling laws. Their marketing follows suit, with legions of paid social media influencers pitching the platforms not as gambling but as savvy investing—a way to turn fandom and hunches into quick cash.

Pulling Back from the Brink

So, what now?

The first step should be to curtail marketing and impose stricter rules on the ways operators are allowed to promote their product. A 13-year-old should be able to watch his favorite sports team without feeling like the game is just a vehicle for betting content. Regulators should restrict campaigns featuring athletes and celebrities, as the UK has done. Marketing rules should reach social media, too. Influencers and affiliates pitching bets and “trades” as easy money should have to disclose every sponsorship, and operators should answer when they don’t. Regulators should consider bans on advertising that frame constant betting as something to be indulged anytime and anywhere. All ads should include disclaimers stating how low the odds of winning really are and how sportsbooks limit or ban users with any real chance of coming out ahead.

Education is also crucial, yet the policy landscape is thin. Only Virginia has passed a statewide mandate; efforts in Michigan, Maryland, New Jersey and West Virginia have all stalled. Voluntary curricula from the Massachusetts Council on Gaming and Health and Next Gen Personal Finance are already available, and the American Institute for Boys and Men, where I am a fellow, is developing additional resources for parents and teachers.

Efforts to educate teens on gambling should take a cue from what we know teaching them about the perils of drugs, alcohol, and sex: Simply advocating abstinence doesn’t work. Decades of evidence reveal that telling teens to “just say no” fails to reduce engagement and leaves them without the tools they need when they encounter these enticements in the real world. Pretending gambling isn’t a problem merely cedes the conversation to social media influencers, celebrity shills, and older acquaintances with legal betting accounts.

---

EdNext in your inbox

Sign up for the EdNext Weekly newsletter, and stay up to date with the Daily Digest, delivered straight to your inbox.

Email

Name

Opt in to another list

EdNext Daily Digest

Subscribe

---

Gambling education should cover two basic subjects: health and math. The first is teaching students to recognize what compulsive use looks like in themselves and their friends, why adolescent brains are especially susceptible to it, and where to go for help. The second is teaching about probability, expected value, and the hard arithmetic of why a parlay that feels like a good bet is almost always a terrible one.

When I speak at high schools, half my time becomes a Q&A session with boys running their betting theories past me. They want to know whether LeBron going over his point total in eight of the last 10 games means he’s likely to go over again tonight (it doesn’t), whether doubling down after a loss will eventually even out (it won’t), or whether an Instagram tout’s “can’t lose play of the day” actually can’t lose (it can). These are ideas no one has ever pushed back on in front of them.

Age safeguards on regulated sites can also be improved. Just as Citi Bike will soon require age verification before riding, gambling apps can do more than check identity at signup and hope for the best. New York Governor Kathy Hochul has pushed regulators to explore stronger safeguards against underage access. Here, as with social media, technology has outrun the rules. Some fixes will work better than others, and some will raise difficult privacy questions. But the longer policymakers wait, the more entrenched the problem becomes.

The industry’s favorite scapegoat—illegal and unregulated operators—deserves more scrutiny as well. While cracking down on offshore gambling often devolves into a game of whack-a-mole, targeted cease-and-desists from state regulators have proven effective and should continue. States should pursue fines and legal action against U.S.-based grey-market operators like Fliff that circumvent state laws, even when those operators hire the president’s son to shield themselves from oversight. They should also go after gambling in children’s video games, a concept so normalized that the CEO of Roblox—a gaming platform whose users include half of all American kids under the age of 16—has openly supported adding more of it.

More impartial research is essential. The federal government spends billions studying the effects of drugs and alcohol but nothing on gambling. States chip in a small percentage of revenue at best. That leaves the field to industry-funded work, which rarely yields findings that could inform meaningful regulation. While philanthropy has begun to fill the void—last year Arnold Ventures announced the largest-ever independent research initiative on sports betting—it can’t do it alone.

Finally, the most important thing states can do to protect kids from gambling is prevent the legalization and spread of online casino games, which are far more dangerous than sports betting. In the seven states with legal online casinos, user losses are roughly four times as large as those from sports betting. Slots, which account for more than 75 percent of online casino revenue, are engineered to ensnare users and have far higher rates of addiction, particularly for underage users.

Ben has seen those higher stakes and increased harms play out firsthand. A majority of his friends who played video games purchased skins and cases, and most got out unscathed. Those who didn’t were the ones who found themselves in the makeshift within-game casinos that allow kids to convert their skins to coins and play roulette, slots, and other enticing, brightly colored games. When pressed on why he thought those kids got hooked, he told me, “There’s gambling, and then there’s GAMBLING.”

Online gambling is here to stay. The question is whether we treat underage access as a design failure or let the industry write it off as the cost of doing business. Physical casinos check IDs at the entrance. Online, the door is always open, and gambling companies spend billions waving everyone in. Until that changes, kids like Ben will keep walking through.

Isaac Rose-Berman is a fellow at American Institute for Boys and Men focused on gambling research and policy. AIBM partnered with Arnold Ventures on this research.

Suggested citation format:

Rose-Berman, I. (2026). “Teen Boys Are Gambling. A Lot. The tidal wave of legal sports betting has easily infiltrated schools and swept up boys. There’s still time for commonsense guardrails and effective education.” Education Next, 26(2), 27 May 2026.

The post Teen Boys Are Gambling. A Lot. appeared first on Education Next.

At FedMart, the discount chain Sol Price built in 1950s San Diego, you could buy a can of WD-40 in one size, the big one, and that was the end of the conversation. Anyone who wanted the small can went without. Price called it the intelligent loss of sales: carry one good version of a thing, refuse the other nine, and eat the customers you lose in exchange for the trouble you save everyone else.

Jim Sinegal, his mentee, carried that habit into Costco in 1983. A Costco warehouse stocks around 4,000 items; while a supermarket runs 30,000 or more, and Amazon runs into the millions. A Costco buyer looks after fewer than 200 products and spends the extra time that buys deciding which ones earn the floor space, killing the underperformers, and doubling down on the winners. By the time you push your trolley through the door, someone has already rejected almost everything that could have been there.

Most of the internet runs on the opposite instinct. Pile the shelf higher, add the SKU, take the margin, say yes to everything. And the people using it are worn out.

I'd bet the next decade runs the other way. People don't want infinite choice anymore; they want fewer decisions inside places where someone has already thrown out the worst options.

Call it the Costco theory of the internet.

For 20 years we built the internet around abundance: more products, more creators, more opinions, more newsletters, more podcasts, more apps, more tools, more marketplaces, more feeds. The founding promise was access: anything, from anyone, anywhere, instantly. No gatekeepers, no scarcity, no permission. The shelf went infinite.

For a while that felt like freedom.

And then it turned into drudgery...

Every ordinary decision now comes with a research burden. Buying a toaster means reading reviews, scanning Reddit, distrusting half the reviews, checking YouTube comparisons, searching for "best toaster no affiliate," then wondering whether the person recommending the toaster is paid, deluded, or defending the thing they already bought. Choosing project management software turns into a 6-week intellectual collapse involving Notion, Linear, ClickUp, Basecamp, Airtable, Todoist, Things, a whiteboard, a notebook, and some founder on X insisting that the wrong task app is why your company has no momentum...

The internet gave us access to anything, and then forced us to consume everything, and then made us responsible for sorting all of it.

The modern consumer has become a part-time procurement department. We audit quality, decode incentives, compare vendors, scan reviews, avoid scams, dodge subscriptions, read refund policies, assess creators, inspect screenshots, and attempt, against all odds, to tell actual expertise apart from people who bought a microphone.

This is considered normal behaviour now.

And it's deranged.

The sane response to all this is, I think, a form of bounded trust.

Costco never promised perfect quality or the best product in every category; and it isn't doesn't claim to be a temple of taste. It sells enormous muffins, bulk socks, patio furniture, protein shakes, car tyres, petrol, hearing aids, rotisserie chickens, appliances, and tubs of dip large enough to drown any and all sorrows.

But more than that: Costco sells a higher floor.

Their promise comes down to two things:

1. you probably won't get ripped off, and
2. you don't have to inspect 900 versions of the same item.

Costco doesn't necessarily take judgement away from you. But it does absorb enough of the evaluation that shopping feels sane again, limiting the shelf, buying with discipline, backing Kirkland Signature with its own name, keeping prices legible, and standing behind the lot with a return policy that assumes you're honest. You don't wander a marketplace full of fake brands, sponsored clutter, manipulated reviews, counterfeit risk, and algorithmic sewage.

Nobody walks into Costco believing every item is elite. They walk in trusting that the floor is higher than the open market, and they'll pay for that trust.

The internet doesn't need more curation in the precious boutique sense. It needs operators who cut fraud, noise, decision fatigue, and bullshit, and who clear the garbage off the floor before you arrive.

Amazon deploys abundance logic in soul-destroying reverse. It has everything, which by now means it has too much. You can still find good things there (or so I'm told), but you do the sorting, and it's very much a case of buyer beware - seriously, buyer fucking beware: parsing the brand names, the reviews, the images, the delivery dates, the sponsored placements, the counterfeit risk, and the chance that a product with 18,000 5-star reviews still singes off your eyebrows is down to you.

A world drowning in options will pay good money for someone else's refusal. Because refusal has become a premium service.

More results stop helping once the results are polluted. Reviews that are fake, incentivised, or written by people with no standards don't improve by multiplying. Creators performing expertise for an algorithm don't add up to expertise. Tools that keep making the same bullshit claim to replace every other tool cancel each other out. And more options stop being a gift the moment you have to become an amateur fraud analyst to choose between them.

The internet's problem has moved from access to trust.

We can find anything. We can't easily tell what deserves our belief, our money, our time, our attention, or our adoption. The old internet solved scarcity; the new internet has to solve filtration, and filtration and aggregation are vastly different jobs.

Aggregation scales because it dodges responsibility. Open the gates, index the world, invite the vendors, let the users sort, take a cut. That became the dominant model because it suited the economics of software: more supply made more surface area, more surface area made more searching, more searching made more money.

Every open system becomes a target for the people gaming it. SEO gaming, review gaming, marketplace gaming, social gaming, recommendation gaming, affiliate gaming, attention gaming. The larger the platform, the stronger the pull to manipulate it. Eventually the user starts paying the tax, spending more time verifying, comparing, doubting, checking, and defending themselves against the system.

Costco-style trust starts when the operator takes - at least - some of that tax back.

A trusted operator narrows the field first, making the choices in advance and accepting the cost of everything it leaves out. Then it absorbs the complexity, doing the dull part before you get there: testing, comparing, rejecting, negotiating, standardising. Then it holds the floor. It doesn't have to make every item extraordinary, it only has to clear the obvious junk and keep a baseline you can feel the moment you walk in.

Most internet businesses miss this. You build trust by making the customer feel less exposed. Announcing your own excellence does nothing.

A marketplace makes you inspect everything. A trusted operator lets you relax, and in some categories that relaxation is the entire product.

Think about the felt difference between buying from a chaotic marketplace and buying from a retailer you trust. In the first, you're on guard the whole time, because every image might mislead you, every review might be bought, every brand might be a shell, every discount might be bait, every result might have paid its way to the top. You'll probably still get what you need. You'll get it defensively.

In the second, you still choose, but you choose inside a zone of lowered suspicion, because the retailer has put skin in the game. Sell you something bad and its reputation pays. Price something absurdly and the relationship cools. Make the returns hostile and the trust drains out. You might never put any of this into words. You feel all of it.

The internet is starved for that feeling.

And it goes well beyond retail...

A Costco-shaped media company wouldn't publish 200 takes a day. It would publish fewer pieces with a higher floor. Readers would show up because it spares them the feed, and it would earn its keep through what it refuses to run.

A Costco-shaped software company wouldn't sell a platform with 70 use cases, 11 pricing tiers, and a thousand features. It would make a clear promise to a clear user. It would end the internal debate. It would say: for this kind of team, doing this kind of work, this is the system. Use it and // or move on.

A Costco-shaped agency wouldn't offer every service that can technically be billed. It would define its shelf. It would turn down bad-fit clients, weak briefs, vanity deliverables, pointless retainers, and work that makes the operator richer while leaving the client more confused. Its standards would be part of the offer.

A Costco-shaped community wouldn't confuse growth with health. It would moderate hard, keep its standards visible, and guard the useful conversation from people who treat every room as a stage, because the health of a community depends on who it removes as much as who it lets in.

A Costco-shaped creator wouldn't post every half-formed thought chasing reach. They'd become a reliable filter. Their audience would trust their judgement because they show restraint, and in a world of constant output restraint becomes a signal.

The internet trained all of us to fear leaving something out. More pages mean more search traffic, more products mean more revenue, more posts mean more shots at virality, more features mean more markets, more services mean more deal flow.

Saying yes has become cheap. Yes to more inventory, more formats, more creators, more sponsors, more categories, more features, more partnerships, more slop, as long as it performs.

The next premium goes to whoever can say no and survive the revenue they walk away from.

Bullshit pays, at least in the short term. Low-quality suppliers pay, bad-fit clients pay, sponsored placements pay, mediocre content pulls clicks, extra features close deals, fake urgency lifts conversion, confusing pricing pulls more money out of people, dark patterns move the metrics. A growth team can always find a way to monetise confusion - and plenty of internet businesses start to rot the moment they work out that confusion is profitable.

The Costco theory says: sell relief, instead. Make people feel that someone competent is handling the market for them.

This is why the membership model works as well as it does. Costco runs as a relationship with an institution, and the annual fee puts the trust down in a contract, in black and white. You hand over money, habit, attention, and your default preference, and in return Costco has to keep the thing worth renewing every year.

It's a different game from the open web's casual opportunism. The mass internet wants traffic and optimises for clicks. The Costco internet wants repeat belief and optimises for "I'll just get it there."

"I'll just get it there" means the customer has taken you out of the comparison set. You've stopped fighting transaction by transaction. You've become infrastructure in someone's life - AKA, the default answer before the question is even formed.

Every founder says they want loyalty, but (time and time again) they build the machine that kills it. They overcomplicate the product, dilute the brand, chase adjacent customers, bolt on tiers nobody understands, publish filler, wave bad actors into the marketplace, swap human judgement for engagement metrics, and reach for pricing tricks, urgency tricks, retention tricks, interface tricks...

People commit when commitment lowers their anxiety. They pay when the payment buys them standards, accept fewer options when the survivors are safer, tolerate constraint when it comes from competence, and come back when the operator has proven that trust beats another night of searching.

A brand is a pattern of kept promises. Over time, people learn what you allow, what you reject, what you repeat, what you protect, what you punish, and on what and where you refuse to compromise.

Digital brands (and particularly the current era of influencer founded DTC companies) run this backwards, blurring their standards over time. They start with a point of view and end as a marketplace, or they start with taste and end as inventory, or they start with a community and end as a growth channel, or they start with a product and end as a bundle of loosely related monetisation experiments.

At no point do they stop to answer:

What do you let in? Who do you let near it? What do you push? What do you kill? What do you refund? What do you ignore? What earns the no?

AI means we can now produce content, software, images, video, music, analysis, pitch decks, landing pages, sales emails, reports, strategies, and whole micro-products at near-zero marginal cost - and so the shelf expands again, and the flood rises. The average unit of internet output gets cheaper, faster, and less trustworthy at the same time.

When production turns abundant, selection turns scarce. Raw output stops being the scarce thing - because the scarce thing is someone willing to tell you which output deserves your attention, which vendor is real, which product works, which argument holds, which plan makes sense, which tool is worth adopting, which document to read and which to delete before it costs you another minute.

The winners here will be operators with both A) taste and B) the power to enforce it. Taste without enforcement turns into slop. Enforcement without taste turns into bureaucracy.

The shallow version of this will be boutiques, directories, newsletters, AI wrappers, and "handpicked" marketplaces that wrap a tasteful interface around ordinary affiliate arbitrage; and of course, it won't last. People can smell fake standards. They know when a list exists to help them and when it exists to monetise their confusion. They know when the operator has actually turned down good money to protect the shelf.

The better version will be companies and people who make trust operational. They'll publish their criteria, keep the offering narrow, explain the tradeoffs in plain language, cut the items that underperform, and refuse to become a dumping ground, so that every interaction leaves the customer's life a little simpler.

The test: does dealing with you lower the load in someone's head, or add to it? If it adds, you're part of the noise.

A Costco-shaped business sells:

1. relief from evaluation
2. the feeling that someone competent has gone ahead with a machete and cleared the path
3. a smaller world that works better than the larger one.

The internet's current default setting is actively hostile to sustained attention. Everything asks for a decision, wants a preference, requests a subscription, a rating, a login, a notification permission, a plan, a personalised feed, an upsell, a dashboard, a profile, a follow, a like, a reaction, a review. The strongest customer experience on offer might soon come down to three words: we handled it.

It can be whole business model, if you let it, but only when it's true.

You can't fake the Costco theory with branding. You can't write your way into trust while the shelf is garbage, can't design your way out of weak standards, can't pose as a filter while the market dumps trash through your side door. It takes operational severity, the real kind.

The operator has to disappoint suppliers, partners, clients, contributors, sometimes the customers themselves. They have to pick the long-term trust account over the short-term revenue hit, and accept that every low-quality thing they wave through taxes the whole system. One bad product makes you inspect the next ten. One lazy essay makes you doubt the whole publication. One weak hire makes the team start to lower the bar. One incoherent feature makes the user wonder who's steering.

A low tolerance for bullshit has to run as an operating system, built it into how the place works.

In practice that means you remove things, you simplify, you say no earlier than feels comfortable. You define what good actually means and then make the standard explicit enough that people can test you against it. You refund when you fail, cut the features that confuse the product, stop publishing when you've nothing worth saying, turn down the clients who'd drag the standard down, and refuse any form of scale that lowers the bar.

Anyone can launch a store, start a newsletter, build a course, spin up a community, publish a directory, open a marketplace, wrap an AI model in a UI. Creation stopped being the test a while ago.

Literally, anyone.

The test now is whether people can trust you to exclude.

The Costco theory of the internet is simple. People are tired of sorting. Tired of comparison, fake reviews, infinite tabs, marketplaces that play like casinos, creators who recommend everything, software that needs a consultant to explain its pricing page, experts with hidden incentives, brands that treat their attention as something to strip-mine.

Most of the time they'd take a safer set of things over the theoretically best one. They want fewer decisions and a higher floor. They want someone with the reputation and buying power to bin the obvious garbage before they walk in. They want the kind of constraint that protects them.

None of this kills abundance. The infinite shelf stays. Some people will always want to browse, research, compare, optimise, hunt for the edge. But across a lot of categories the centre of gravity is moving.

The new edition of Permissionless is now free

I’m releasing a new edition of my book Permissionless today. I've rewritten it from the ground up: clearer, sharper, and rebuilt around the ideas I think matter most in this moment. And it’s entirely free.
You can read it here!