The Geography Of App Bans In 2026: For The First Time, The Reversals Are Winning

Reviewed by Irfan Ahmad.

VPNRankIO analyzed 64 government actions involving major apps and platforms since January 2024. It found that more app bans were lifted than new bans were introduced in the first half of 2026.

App bans used to mostly move in one direction: governments imposed new bans, and few were lifted. That pattern is now changing. Governments are still imposing app bans, but they are also lifting more of them.

The numbers by half-year tell the story. In the first half of 2025, governments imposed seven new blocks on major platforms and rolled back two. In the second half, six new blocks and just one roll-back. Then the first half of 2026: six new blocks – and eight roll-backs. It is the first period in the dataset where the roll-backs outnumber the bans.

All roll-backs share one thing: very few of them were unconditional. Turkey restored Roblox in June 2026, after 680 days, once the platform introduced age verification and better moderation. Russia had banned the same game in December 2025 and rolled it back six months later, citing compliance with local law. Albania ended its year-long TikTok ban in February, once the company introduced new safety filters – and a month later the country's Constitutional Court ruled that the ban had breached freedom of speech anyway. Kuwait and Jordan both restored Roblox with in-game chat disabled. Nepal lifted its TikTok ban in 2024 once the company agreed to register locally. Even the biggest case of all follows the same template: the American saga of TikTok, which started with a divest-or-ban law in April 2024, included fourteen dark hours in January 2025, and ended in January 2026 once the app was transferred to a US joint venture. What seems to be the lesson learned by the platforms is that there is a price list for bans. Verification systems, local reps, disabled features – pay it, and access opens.

The exception to the rule is Russia, which used the same period to walk the other way: it banned Discord in October 2024, cut voice and video calls on WhatsApp and Telegram in August 2025, and in February 2026 banned WhatsApp altogether, and removed Meta's domains from the national DNS. By April, connectivity researchers estimated Telegram failure rate to be around 95% without a VPN. Iran went even further: authorities imposed a nationwide internet blackout in January 2026, followed by another near-total shutdown in late February. When international internet access began returning in late May, many users remained on a tiered, whitelist-based network where access to numerous foreign platforms continued to be restricted. As for China, it barely appears on the timeline for the simplest reason: nothing changed. Facebook has been blocked there since 2009, Instagram since 2014, every single language of Wikipedia since 2019. Stability, in that column of the map, is the whole point.

There is another, rapidly growing category between bans and roll-backs: the partial ban. The United Arab Emirates (UAE) and Qatar still allow WhatsApp messaging but block its voice and video calls at network level – the policy quietly ending in neighboring Oman in December 2024 once the calls simply started working. Egypt restricts the same calls intermittently at carrier level. Russian ban on calls on two platforms that it was allowing at the time also belongs in this family. So, perhaps, does the newest tool of all: the age gate. Australia implemented its under-16 minimum-age law for social media in December 2025, and Indonesia restricted Roblox for under-16s in March. Governments apparently feel less inclined to switch an app off for everyone when they can switch off a feature or an age group.

"Every date in this dataset shows up in our traffic," Matt, founder of vpnrank.io, the independent comparison site that compiled the timeline, says. "When a country blocks an app, searches for workarounds from that country spike within hours – and when a ban is lifted, they fall away just as fast. A map of app bans is, in practice, a map of where people are trying to route around their own network."

What should readers expect from the rest of 2026? The dataset suggests keeping an eye on three ongoing cases. Turkey's communications minister announced in early June that Discord now "meets our criteria," suggesting that a roll-back that has not happened yet by mid-July is about to happen soon. Kyrgyzstan's culture ministry officially proposed to roll back TikTok ban in February, but the proposal has not been passed yet. And Gabon, which banned TikTok, Facebook, Instagram, WhatsApp and YouTube in February as part of public sector strikes, has since signed a twelve-month compliance agreement with TikTok – again following the template of negotiations, not permanent bans. Wall builders are real, and Russia and Iran show how far down that road can go. But for the first time in this dataset, traffic on the other road is heavier.

Methodology: vpnrank.io compiled 64 dated events – bans, roll-backs and feature-level restrictions – affecting TikTok, Facebook, Instagram, X, Telegram, Wikipedia, WhatsApp, Roblox and Discord between January 1, 2024 and mid-July 2026. Every entry is sourced to regulator statements or mainstream reporting (AP, Reuters, BBC, Al Jazeera and national outlets), and every current status was re-verified in the week of writing. Feature-level restrictions (such as voice call blocking) and age-based restrictions are counted separately from full bans. The underlying event list is available on request.

Author Bio: Matt is the founder of vpnrank.io, an independent VPN comparison site. The site publishes hands-on reviews, quarterly speed tests and daily-tracked price index, and monitors where major apps and platforms are banned worldwide.

Read next:

• What really happens to your data when you click ‘delete’?

• Gmail first, everything else later

• As AI reshapes newsrooms, leading media outlets are charting different paths for its use

by Guest Contributor via Digital Information World

AI chatbots need cultural awareness to earn trust

As AI becomes part of daily life, a UC researcher says one of its most unacknowledged blind spots is also one of the most human: culture.

Image: Katja Ano - Unsplash

In a new paper, ‘Culturally responsive AI chatbots: From framework to field evidence’, Te Whare Wānanga o Waitaha | University of Canterbury (UC) Professor Vik Naidoo and co-author Karman Kaur Chadha argue that AI systems designed around largely Western assumptions often perform poorly in different cultural settings. The result can be more awkward interactions, which can lead to lower trust, weaker engagement, and systems that fail the people they are meant to serve.

Their paper introduces the Culturally Responsive AI (Chatbot) Framework, or CRAIF-C, a practical model for building AI-powered chatbots that are designed with cultural diversity in mind from the outset rather than treated as an afterthought.

CRAIF-C is a four-part framework for building culturally responsive chatbots across the full AI lifecycle. It combines Enculturation, which embeds cultural norms, language, and context into data and design; Adaptive Interaction, which adjusts tone, pacing, and style in real time; Explainability and Transparency, which provides culturally appropriate forms of explanation; and Governance and Accountability, which embeds oversight, cultural risk assessment, and community-informed review. Overall, the framework treats cultural fit as a core design requirement rather than an afterthought, shaping everything from training data and interaction design to explanation and governance so chatbots are more natural, trustworthy, and appropriate across different cultural contexts.

“People often think cultural awareness is translating words, but it’s much more than language. It’s about context, social norms, communication styles, and how people interpret the world around them,” Professor Naidoo says.

Many AI systems reflect Western logic because they are trained mainly on Western data. “If it is trained mainly on Western data, then the outputs it produces will also reflect Western assumptions,” he says.

That becomes a problem when chatbots are deployed globally, especially with firms trading across borders. A system may appear efficient but still fails if users do not relate to it or trust it.

While working with an AI development company in Sydney, Professor Naidoo helped train engineers to think about cross-cultural communication at the beginning of the design process, rather than simply translating English-language outputs at the end.

The company had been deploying chatbot systems in countries including Indonesia, Thailand and Vietnam, but was not getting the user engagement they had expected.

“What they were finding was that consumers simply weren’t engaging with the chatbots, so we started asking what cultural nuances needed to be built into the models from the start.”

Sometimes the differences were subtle. In a Western setting, a chatbot might open with small talk about the weather. Whereas in Jakarta, Professor Naidoo says, a question about traffic could feel more natural and relevant.

The paper argues these differences should be considered across the full AI lifecycle: from training data and interaction design to explainability, transparency and governance.

Professor Naidoo says the issue matters not only for international organisations, but also for culturally diverse countries such as New Zealand and Australia.

“At the moment, the conversation around AI is heavily focused on cost-cutting and efficiency, but from a marketing and end-user perspective, the real question is: are we creating value?” he says. “If people don’t trust the technology or can’t engage with it, then it hasn’t solved the problem.”

As AI becomes more deeply embedded in everyday services, he says understanding culture will be essential to making chatbots useful, trustworthy and effective. For now, he hopes the paper helps move the conversation beyond the hype surrounding AI and toward better design practice.

This article is republished with permission from Te Whare Wānanga o Waitaha | University of Canterbury.

Reviewed by Irfan Ahmad.

Read next: 

• AI Translation and Human Interpretation Differ in Handling Contextual Meaning, Study of UN Speeches Finds

• What really happens to your data when you click ‘delete’?
by External Contributor via Digital Information World

What really happens to your data when you click ‘delete’?

Deletion is a common part of modern life. We send files and folders to the recycle bin all the time, and often get rid of unwanted personal accounts. But do you know what really happens when you hit the “delete” button?

Image: Marija Zaric - Unsplash

The process of deletion is often poorly explained by tech providers. Lack of transparency around how requests are processed, and the absence of clear confirmation that data has been removed, are frequently highlighted in research.

These issues are not limited to operating systems. On social media platforms and a wide range of subscription services, deletion mechanisms are often just as unclear. Such misconceptions can have serious implications for the wellbeing of users, exposing them to security and privacy vulnerabilities.

A further complication is that many users fear losing data, files, photos and videos through accidental deletion – a condition dubbed “diagraphephobia”. Though this is not a recognised pathology, studies of digital hoarding behaviour show that some people get very anxious at the thought of losing or accidentally deleting personal information such as photos and music.

Such concerns extend the amount of personal data that sits untended but undeleted, ready for potential misappropriation.

Deletion vs erasure

A common misconception is thinking deletion means erasure – the idea that once a deleted item is no longer visible, it has been completely erased with no means of retrieving it. This is largely wrong.

When users delete files – moving them to the trash bin, then deleting permanently – the data isn’t actually erased. Rather, the system marks storage space as reusable and updates metadata to unlink the file. But the underlying content often remains recoverable with the right tools, especially when data is duplicated across multiple systems or devices.

More robust forms of deletion include physical destruction of the storage medium, overwriting the memory blocks with new data to bury the original, or encrypting the data and then destroying the key.

However, such methods can be expensive and may render the storage device unusable (for example, when overwriting data on magnetic media). The characteristics of cloud infrastructure also pose challenges to secure data deletion.

Similarly, content on public social networks may not really be erased after deletion due to replies, comments and internet archives which can all store the posts in some fashion. The meaning of a deleted tweet can, for example, be recreated based on replies and mentions.

Zombie accounts

Another misconception is that deleting an app from a device automatically deletes the associated account. Users often leave mobile app accounts undeleted because they are unaware of their existence, having deleted the related app.

These are called zombie accounts – abandoned profiles for a wide array of services, from shopping and storage to dating, finance and streaming. Millions of users possess zombie accounts, with personal data that is vulnerable to cyber-attacks and data breaches.

The exponential growth of AI raises a further question: can data really be deleted once it has been used to train AI models? Machine learning modules easily memorise the data they have been trained on, but “unlearning” is difficult.

In theory, people in Europe and many other countries around the world have a legal “right to be forgotten”. This means they can request the full erasure of any personal data a platform or a company may hold. AI developers are considered data controllers under the EU’s General Data Protection Regulation (GDPR) laws, for example, and are subject to this obligation.

However, there are no clear guidelines on how erasure should be enforced within AI systems. Regulators may request deletion, but AI companies can argue that compliance is infeasible on account of technical constraints.

Explainable deletion

To counter the risks posed by incomplete deletion, I believe there is a pressing need for provision of concise, accessible and clear information about how deletion really works.

One potential approach is “explainable deletion” – a protocol developed by Marvin Ramokapane at the UK’s National Research Centre on Privacy, Harm Reduction and Adversarial Influence Online (Rephrain), based at the University of Bristol.

The intention is to make deletion processes more transparent and understandable without overwhelming the user with too much information in one go.

Explainable deletion breaks down information into six categories: what, how, when, who, where and why (see diagram). Each offers users bite-size information about that part of the process.

Explainable deletion’s six categories:

Marvin Ramokapane and Dana Lungu, CC BY-SA

Explainable deletion is designed to give users control over their actions, the autonomy to choose the deletion type that is right for them, and the assurance that their desired actions have been fulfilled. It may also tackle the anxiety of accidentally losing data by giving options for data recovery.

While explainable deletion has not yet been adopted in practice, service providers – both platforms and developers – could heighten user trust by adopting such protocols. This framework can also be proof of compliance with GDPR regulations and the right to be forgotten.

Most of us use systems that collect and store our data on a daily basis. If these systems clearly explained what they store and what deletion really means, it could help us to spot the accounts and data that pose a real risk – and to take back control of our digital footprints.

Dana Lungu, Research Associate, National Research Centre on Privacy, Harm Reduction and Adversarial Influence Online, Rephrain, University of Bristol

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Reviewed by Irfan Ahmad.

Read next: 

• Gmail, YouTube, and Facebook dominate web usage as researchers examine where Americans spend online time

• Influencers increasingly drive U.S. purchases, with YouTube links and TikTok videos leading product discovery paths

by External Contributor via Digital Information World

The Influence of Influencers on Product Choice

By Katharina Buchholz, Statista. Reviewed by Irfan Ahmad.

The number of Americans who have been coaxed by an influencer or a celebrity endorsement to buy a product is rising. In a survey by Statista Consumer Insights of more than 1,000 U.S. respondents, the most popular way to carry out such purchases was via a link in a YouTube video description – a tried and tested way for creators to peddle their own or sponsored products. Almost as popular is buying a product after seeing a TikTok video. The social media platform in 2023 launched¹ its own in-app shop, making it easier for influencers and brands to list curated products.

On Instagram, products advertised via Stories are a little more popular than those presented in regular posts on the feed, but the difference is small. For TikTok, the stories feature hasn't caught on as much as a sales tool, with only 17 percent of consumers mentioning purchases via TikTok stories or TikTok Now as opposed to 28 percent saying the same about regular TikTok videos. Also popular is the Amazon Storefront, which provides a free and customizable landing page for influencers and other sellers on Amazon to present products.

Americans were most likely to say that influencers had made them try a new restaurant or food – not always a direct purchase from the creator as unpaid and paid restaurant recommendations as well as product sales and recipe tips are all popular among food influencers. A product was purchased upon recommendation from a creator, however, for 29 percent of respondents in the survey. 19 percent said they tried out a new routine because of an influencer – again, not always a purchase, but likely an effective tool to turn viewers into customers further down the line. While 15 percent said they had booked a travel experience due to creator, 14 percent mentioned attending an event.

Notes: Originally published by Statista. Republished under the Creative Commons CC BY-ND License. Added a TikTok Shop reference link. Formatting corrections made only where necessary.

[1] https://ift.tt/EY6W1bf

Read next: What Tools Would You Build If You Didn't Need A Developer?
by External Contributor via Digital Information World

Research Finds Social Media Makes People Overestimate How Often Rare Experiences Occur

By Jacob Levin

Image: Swello - Unsplash

People are naturally attracted to excitement.

According to new Virginia Tech research, the same is true on social media, where feeds are overrun with international travels, skydiving, and other experiences that may seem out of the ordinary.

When social media users constantly are shown rare and unique events, they overestimate how often those occur and seek to share their own. This unbalanced content stream creates a “rareness bias diffusion.” This new concept by Alice Jang, assistant professor, and Viswanath Venkatesh, the Verizon Chair of Business Information Technology, both in the Pamplin College of Business, defines that this bias can spread rare information widely, leaving everyday social media users so exposed to it that they perceive it as more common than it truly is.

“They think everyone else is traveling to Paris or doing something exciting, while they are just sitting at home doing mundane things,” Jang said. “It makes people feel like their lives are falling short, when in reality, everyone is mostly doing mundane things — they just only post the exciting parts.”

These studies revealed that human interaction with social media content is a compounding factor in how inaccurate information spreads online. According to Jang, the research challenges the current understanding of social media content.

The tendency to overlook the ordinary happens because of social media users’ perception bias and their motivation to seek variety in the content they see and share, according to the research published in MIS Quarterly.

The result is a skewed picture of how the world actually is, where the unusual looks common, and the common disappears. This suggests social media users should be cautious — and not for the usual reasons. The distortion on social media is not totally driven by the algorithm. Instead, it comes from ordinary human biases that social media compounds into a pattern where rare information gets shared more often.

During the research, Jang and Venkatesh conducted six experiments to understand how people choose what to share online and their reaction to seeing rare content. She presented research participants with fictional scenarios of a dystopian city under attack by various monsters – something unique that people would probably share online.

Of the various monsters and attacks, participants were shown some scenarios more than others. In the end, those participants were more likely to share the scenario that they had seen the least. This reinforces the idea that people are more likely to share content that is unique, rare, or out of the ordinary. Then, she created a simulated mock network of thousands of social media users and unique social media content to analyze these human biases on a larger scale.

“There is a bias that we simply cannot fix. It's impossible to fix,” said Jang. “A lot of the prior literature just disregards this fact and assumes that people are rational”.

Venkatesh added, “Yes, it's a bias but when people know this, they could perhaps know that what we are seeing on social media is others' highlight reel and not the day-to-day life. This may reduce harmful negative impacts on oneself."

This innate human bias forces social media users to wonder: Is what I am seeing online representing what is actually happening? While people may actively edit their timelines and curate unique content to post online, the research reveals that the corresponding problem runs much deeper than just a network’s algorithm.

There is a baseline human bias that the platform cannot fully fix. Instead, the responsibility falls back on the user.

Ultimately, the research found, social media users need to understand that their feeds are overrun with exciting and out of the ordinary content because of what they engage with. Having this knowledge allows users to better protect themselves from this false reality on social media.

Reviewed by Irfan Ahmad.

This article was originally published by Virginia Tech News and is republished with permission.

Read next: 

• Study Finds Low-Quality AI Videos Dominate TikTok Feeds for New Users and Children

• Investing podcasts are super popular. But do they help people learn the market?

by External Contributor via Digital Information World

Investing podcasts are super popular. But do they help people learn the market?

By J. Merritt Melancon, University of Georgia

New study suggests individual traders benefit from the information provided by experts hosted on podcasts.

Image: Cody Board - Unsplash

From game highlights to cold cases to political pontificating, podcasts provide the stage for public conversations in 2026. But are we learning anything from all that chatter?

When it comes to investing and playing the markets, the answer seems to be yes.

New research from the University of Georgia Terry College of Business suggests that people who trade individual stocks, also known as retail investors, do better because of the information they glean from financial markets podcasts.

“More and more Americans are listening to podcasts each day, and investing podcasts represent a growing segment of this industry,” said Braiden Coleman, co-author of the study and an assistant professor in UGA’s Terry College of Business. “We wanted to see if the podcasts actually influenced investor trading activity and whether that helps or hurts capital markets from an information asymmetry standpoint.”

They found that in the three days after an investing podcast discussed a company’s earnings, trading activity significantly increased, especially among retail investors.

But what they found especially exciting was that these surges in extra trading didn’t seem based solely on name recognition or social media buzz but on well-considered information.

“One of the main takeaways of the paper is that podcasts don’t seem to lead to purely speculative-based trading,” Coleman said. “We don’t see any return reversals or big price corrections following podcast coverage.”

Podcast discussions reduce information asymmetry by 22%

Usually, when earnings news is announced, stock prices bounce around a little as investors digest the new information.

In cases of meme stocks or other hyped-up stocks, prices will often surge as investors rush to buy the trendy stock and then fall when everyone comes to grips with that company’s fundamentals and starts to sell.

Information asymmetry, when one party in an economic transaction has more or better information than the other, is what causes this rush and retreat. It’s the enemy of retail stock traders who may not have as much time to evaluate firm announcements or the background to process the information accurately.

The research team found that podcast discussions help retail investors reduce information asymmetry by 22%.

Coleman believes podcasts can bridge the information divide because they increase the amount of long-form discussion the average investor can consume.

“There’s more information out there than ever, but with a lot of these platforms you have to be actively consuming — looking at your phone, reading your computer screen,” Coleman said. “With podcasts, you can listen to them while you are driving to work, while you’re exercising or while you’re cleaning the house.

“There’s a bigger reach because you can consume the content while you’re performing other activities.”

Expert commentary provides multiple viewpoints on investment strategies

The other benefit is the number of experts invited to speak on podcasts. Most segments are structured as conversations between hosts and industry or sector experts. The research team found that podcasts including multiple viewpoints further reduced information asymmetry in the markets.

“These guest appearances really appear to drive value,” Coleman said. “If you have multiple voices on a podcast and they’re sharing more unique insights about the company at play, then our results are stronger for those podcasts. Listening to podcasts that include guests and expose you to new ways of thinking, fresh insights and fresh perspectives can really help investors process information better.”

Podcasts offer fast way to get a lot of information on the market

The researchers turned to a podcast metadata library to gather the names of all investing podcasts listed between 2008 and 2022. They used artificial intelligence to transcribe and sort those podcasts, identifying those episodes that included discussions about publicly traded companies.

That resulted in a sample of 1,782 shows and 49,772 individual podcast episodes. The number of episodes per year grew over time, with approximately 2,000 in 2016 to more than 11,000 in 2022.

They mapped several trends in coverage over the years.

First, retail stocks, those with less of their shares owned by institutional investors, are covered more by investing podcasts than other stocks. Publicly traded stocks usually saw a pickup in coverage around each firm’s earnings announcements.

Coleman’s team also found podcasts that spent more time discussing the fundamentals of a company’s performance improved investor performance. Podcasts with tighter delivery (discussing more information in a shorter timeframe) had the same effect.

The overarching message is that the more high-quality viewpoints or data points an investor can get before making a trade, the better off they are. Podcasts, Coleman said, seem to be a great way for people with limited time to take in information.

Published by the Review of Accounting Studies, the publication was co-authored by Texas A&M University accounting professor Brady Twedt, Terry accounting doctoral student Matt Hall and Terry doctoral graduate and current Texas Christian University professor Karson Fronk.

Reviewed by Irfan Ahmad.

This article was originally published by the University of Georgia and has been republished with permission.

Read next: Study Finds Low-Quality AI Videos Dominate TikTok Feeds for New Users and Children
by External Contributor via Digital Information World

Study Finds Low-Quality AI Videos Dominate TikTok Feeds for New Users and Children

By Liam Curtis, Kapwing

Research from Kapwing reveals that nearly 60% of TikToks served to new users and children are AI slop. But which categories and tags are the worst affected — and what does this landscape look like to kids?

TikTok has a slop problem.

Some 59% of videos served to a new TikTok account’s “For You” page are AI slop, according to Kapwing’s research.

That’s three times as much slop as a new YouTube user encounters. And a similar share (57.4%) of all TikTok videos aimed at children are AI slop, too.

Image: Kapwing In Partnership With Neomam Studios. CC BY-SA 

Back in 2025, TikTok announced a new tool to help users control the level of AI-generated content (AIGC) in their feeds, declaring that “many people enjoy content made with AI tools, from digital art to science explainers, and we want to give people the power to see more or less of that.”

But when it comes to slop, not only is it bad for kids but many adult users would rather see far less on social media. As the BBC’s Joe Tidy notes, often “the number of likes for the AI backlash comments far exceeds the original [AI-generated] post.”

To understand the depth of the problem, Kapwing analyzed thousands of videos across TikTok’s top categories and hashtags to measure the prevalence of AI slop. This was defined in Kapwing's previous YouTube AI Slop Report as “careless, low-quality content generated using automatic computer applications and distributed to farm views and subscriptions or sway political opinion.”

Image: Kapwing In Partnership With Neomam Studios. CC BY-SA

One in Three TikToks a New User Sees Is AI Slop

TikTok adjusts its video feed for users based on “signals” including follows and likes, category preferences, and previous scrolling activity. So, to get a better idea of the ‘raw’ TikTok experience, we established a new account and recorded which of the first 500 videos were human-made and which were AI slop.

When you sign up for a new account, TikTok feeds you “popular content appropriate for a broad audience” and “content influenced by your location and language settings” until it figures out what you like. In our test run, the very first TikTok served to Kapwing’s dummy account was a low-quality AI-generated video that appears to have since been deleted.

Overall, 294 or 59% of the first 500 were AI slop.

Image: Kapwing In Partnership With Neomam Studios. CC BY-SA

TikTok delivers nearly three times as many AI slop videos as we found when running the same experiment with a fresh YouTube account, where 104 (21%) of the first 500 videos on the YouTube Shorts feed were AI brainrot.

On TikTok, the prevalence of slop on a fresh feed may be a matter of sheer saturation — by November 2025, the company had already labeled a staggering 1.3 billion videos as AI-generated.

On the other hand, AI companies train the models that generate AI slop on existing footage, and their output represents a flattening or aggregation of the patterns found in human-made content (rather than in “reality”) — an effect that scholar Roland Meyer has labeled “platform realism.”

Filling a new user’s experience with slop videos optimizes the feed as a familiar space and acclimatizes them to the aesthetic and thematic norms of TikTok culture.

Kids, Science and Education, and Health TikTok Categories Are Most Clogged With AI Slop

Next, we checked a sample of 10,742 TikTok videos across the most popular tags in 20 categories, noting the number of AI slop and non-AI slop videos. The category with the highest slop density by far was Kids (57.4%) — more on that below.

Science and Education (35.0%), Health (33.8%), and History (33.5%) are the nearest contenders. In the top nine categories, more than one in ten videos were AI slop. But videos in the Fitness (1.6%), Music (1.5%), and Fashion (1.3%) categories are almost entirely human-made.

Image: Kapwing In Partnership With Neomam Studios. CC BY-SA

Scientific facts and concepts lend themselves to visual illustration through animation. But when such videos are generated with haste and sensationalism as their guiding principles, the potential for deeper value is compromised.

Researchers have warned that slop educational material clutters the platform and competes with more authoritative sources.

TikToks such as the baffling Eating Lemon video below rattle along at the turbo-TikTok pace the models seem to have learned and feature the misspelled words and misshapen lettering that have come to be associated with AI imagery.

@vitalverse1 Science in action. What happens while Eating Lemon in the human body?😱🤮😱#humanbody #anatomy #3danimation #sciencetok #viral ♬ original sound - user99611117491

Meanwhile, creators such as Jonathan Laramy, the man behind Chloe VS History, see AI as a chance to bring educational topics to life, exploiting the technology rather than the viewer.

“Yes, there are people out there that obviously don't care about history,” he told Sky News, regarding his competitors, “because there are mistakes left, right, and center.”

Laramy’s content is hosted by an ultra-realistic AI presenter called Chloe, who is easily mistaken for a real person due to the “human emotion and non-verbal cues” facilitated by the latest video models.

However, the scripts are AI-generated, too, and viewers have criticized the channel for its historical inaccuracies.

Brendan Gillis, Director of Teaching and Learning at the American Historical Association, warns that AI-generated history is only as reliable as the source material it draws from. Because AI fills gaps using patterns from its training data, it can introduce inaccuracies, biases, and stereotypes.

97% of #CartoonKids TikToks Are AI Slop

Of the 2,000 featured videos we analyzed in TikTok’s Kids category, some 1,147 (57.4%) were AI slop.

The worst-affected tag, #cartoonkids, was almost entirely made up of slop, with only three of the 100 videos we checked being human-made. Around one-third or more of the videos for nearly all of the tags we checked were AI slop, and even #babytok — which had significantly less slop than other tags — featured one slop video out of every ten on its tag page. 

Image: Kapwing In Partnership With Neomam Studios. CC BY-SA

The dangers of automated content for children are not new, but the scale at which children are being targeted with slop — ranging from nonsense ‘brainrot’ to dangerously inaccurate songs and lessons — is staggering.

“I think of this as toddler AI misinformation at an industrial scale. It’s very risky for the developing brain,” says Dr. Dana Suskind, a professor of pediatrics at the University of Chicago, speaking to Mother Jones.

“Every experience is building a million new neural connections. You will be unintentionally wiring the brain in incorrect ways.”

One of the first videos to pop up under #preschoollearning appropriates Sesame Street characters to deliver a lesson about counting cookies. Not only do the numbers not match the cookies, but the animation is careless, and the voices are borderline terrifying. The comments section is stuffed with short, nonsensical sequences of letters that boost the video’s visibility.

@spongebob.uk1 Counting Cookies for Kids 🍪 Fun Learning Numbers with Cookie Monster Learn numbers in a fun way with Cookie Monster while counting delicious cookies! 🍪 Perfect for toddlers and preschool kids to practice counting, early math skills, and number recognition through a playful learning adventure. Great educational video for kids who love Sesame Street characters and fun learning! #countingcookies, #cookiemonster, #learnnumbers, #countingforkids, #kidslearning, #preschoollearning, #toddlerlearning, #numbersforkids, #learncounting, #educationalvideo, #kidseducation, #learningisfun, #funlearning, #earlylearning, #preschoolactivities, #toddleractivities, #kidsvideos, #learningnumbers, #countinggame, #numbersong, #kidsmath, #mathforkids, #kidsfunlearning, #educationalforkids, #kidsyoutube, #kidscontent, #learningvideos, #kidsteaching, #homeschoolkids, #preschoolkids, #learnwithfun, #smartkids, #educationchannel, #learningtime, #kidsstudy, #numberspractice, #countingtime, #learn123, #numbers123, #kidsnumberlearning, #funforkids, #kidslearningvideos, #playandlearn, #kidsbrain, #learningathome, #kidseducationvideos, #childeducation, #kidsedutainment, #learningnumbers123, #kidsactivityvideo, #educationfun, #learningforkids, #countingactivity, #kidsmathfun, #preschoolmath, #toddlermath, #educationalcontent, #funeducation, #kidsknowledge, #learningadventure,#animationviral #spongetherapy #underwateradventure #oceanfun #cartoonvideo #animationshorts #viralshorts #funnyanimation #cartoonforkids #seaadventure #oceanworld #cartoonfun #spongeasmr #viralvideo #youtubeviral #kidsentertainment #animatedshort #cartoonstory #oceanlife #kidsyoutube ♬ original sound - English

A recent survey from the Family Online Safety Institute found that only 51% of parents use parental controls on tablets and 47% on smartphones.

Meanwhile, over 70% of babies and under-twos use screens, and one in 10 babies regularly falls asleep with a screen, according to the UK’s 1001 Critical Days Foundation — so named because the 1,001 days from pregnancy to age two are critical for brain development, with up to one million neural connections forming every second.

The TikTok AI Slop Backlash

The people at TikTok are well aware of the AI slop backlash. In May, the company began to rein in its new AI-generated summaries, which users had noticed regularly produced bizarre mistakes, such as erroneously identifying dancer Charli D'Amelio as a “collection of various blueberries with different toppings.”

Alongside allowing users to reduce AI content in their feed, the company announced a $2 million educational fund for experts to develop content around AI literacy and safety.

But the stats reinforce the feeling that AI slop-clogged feeds are already the new normal. Meta’s Mark Zuckerberg has tagged this era of generative AI content the “third phase” of social media — following personal and then creator-based content.

This third AI phase itself has gone through different eras, from the comedic novelty of Will Smith spaghetti videos to the emotionally nuanced but factually inaccurate Chloe VS History.

But slop is likely to continue being slop, and children are sure to encounter damaging AI material as long as humans outsource the hard work of video production to robots without putting in the time and oversight to ensure higher standards are met.

This is because the training data contains mistakes. And most of all, it is because video is a mode of human communication, which is nuanced, dynamic and — well — human.

Methodology

We manually analyzed 10,742 TikTok videos across 20 categories. We began by building a seed list of 20 popular TikTok categories (e.g., Food, Travel, Fitness, etc.) and at least three of the most popular tags for each category (e.g., #traveltok and #foodie).

Next, we manually analyzed the featured videos displayed on each tag's page (e.g., https://ift.tt/K8nrFN3), recording the count of AI slop and non-AI slop videos. This allowed us to calculate the percentage of AI slop videos for each tag, which we then aggregated for each category.

To find the proportion of AI slop that is served to new users, we established a brand new TikTok account and recorded the appearance of AI slop videos in the “For You” section while scrolling the first 500 TikToks.

AI slop videos were defined as those with obvious use of AI-generated visuals, as well as low-quality clip/compilation-style videos with clearly AI-generated scripts and voiceovers.

Reviewed by Irfan Ahmad.

This article was originally published on Kapwing and republished here with permission.

Read next: 

• Why LinkedIn Top Voices Are Outperforming Famous CEOs (And What It Means for You)

• AI Translation and Human Interpretation Differ in Handling Contextual Meaning, Study of UN Speeches Finds
by External Contributor via Digital Information World