Why I don’t use social media

Silvino Rodrigues 32 min read Essays Social MediaPrivacyAutobiographic

People ask why I don’t use social media. Privacy is a good part of the answer, but that’s not all.

There is the time it takes; and the superficial and meaningless comparison with other people; … but the thing I keep coming back to, is this concept of friendship. Friendship is not a metric - neither in count, or in strength - and some arbitrary company has no right to who, what, or how, I interact with my true relationships; nor should they collect information about who I talk to or what we share. “Social platforms” are, ironically, less about socialising than they are about society’s addiction to convenience - and by doing so, destroy the the very essence of what builds societies.

Not that social media has necessarily made us less social. Rather, it has made socialising less demanding, stripping away precisely those obligations, compromises, and shared experiences through which societies develop cohesion. So-called “social platforms” may have optimised the mechanisms of social interaction while neglecting the conditions necessary for society itself. The consequences are difficult to ignore: increasingly superficial relationships, cyberbullying, social isolation, and, in some tragic cases, self-harm and loss of life.

I have listed eight reasons below as to why I stay away. Some have good research behind them. Others are about what I want from my own life. They don’t all carry the same weight - and on some others may disagree - but they’re strong enough for me to have a visceral distaste for social media.

1 The perpetual comparison is harmful

Psychologists studied comparison long before social media existed. When we compare ourselves with someone we see as better off, they call it “upward comparison.” Gerber, Wheeler and Suls reviewed more than sixty years of research and found that people generally preferred this kind of comparison. It often left them feeling worse and judging their own abilities more harshly. 1

McCarthy and Morina found a related pattern in research on depression and anxiety. Among people with mental health conditions, judging themselves unfavourably against others was associated with worse symptoms. That does not prove that comparison caused those symptoms, but it gives me reason to take the habit seriously. 2

Comparison can also inspire us. My objection is to making it a constant part of my day.

And social media is specifically designed for this! You do not even have to post anything. Scrolling through other people’s lives without talking to anyone is what researchers call “passive use.” Verduyn and colleagues found that this kind of Facebook use lowered wellbeing, with envy helping to explain the effect. 3

There is broader evidence too. When Facebook first opened, it reached US universities at different times. Braghieri, Levy and Makarin used that rollout to study changes in student mental health across 775 colleges. Mental health worsened after Facebook arrived, especially among students in circumstances that could make social comparison harder. 45

A later experiment paid 1,661 people to deactivate Facebook for four weeks before the 2018 US midterm elections. They gained about an hour a day and reported a small improvement in wellbeing. Political polarisation fell, but so did knowledge of the news. After the experiment, they used Facebook less and valued it less than before. 6

Hunt and colleagues found reductions in loneliness and depression when 143 undergraduates limited their use to ten minutes per platform each day for three weeks. 7

These findings don’t settle the question for everyone. Orben and Przybylski found a tiny association between screen use and adolescent wellbeing. 8 A review of twenty-five reviews found that most described the association as weak or inconsistent. 9 Odgers argues, in a Nature book review, that claims about social media causing the teenage mental-health crisis run ahead of the evidence. 10 That is her interpretation of the research. Publication in Nature does not make it the last word.

I don’t need social media to be bad for everyone before I decide it isn’t good for me. My concern is the comparison and the watching. Applying those findings to myself is my judgement, not a result from a study of autistic or ADHD adults. 11

Then there is the infamous fear of missing out, or FOMO. When your friends, news and invitations are all there, leaving costs you the convenience of keeping up without reaching out. But the effort matters to me. Making time for someone, and knowing they made time for me, is part of what gives a friendship value.

In the earlier Facebook experiment, people knew less about the news after a month away. They also valued Facebook less. Nine weeks later, one in twenty still had their accounts deactivated, without being paid to stay off. 6 In Hunt’s study, fear of missing out fell in both groups, even among people who weren’t asked to cut back. The authors suggested that monitoring their own use helped. 7

I can accept missing some updates. I would rather spend that time keeping in touch with the people I care about - quality, not quantity.

2 The risks are not worth it

Australia’s social media age restrictions took effect on 10 December 2025. Platforms covered by the law must take reasonable steps to prevent under-sixteens from holding accounts. The penalties fall on platforms, not children or parents. 1213

Similarly, the UK’s child protection rules took a different approach. Ofcom’s codes took effect on 25 July 2025, with requirements for age assurance and for filtering specified harmful material from children’s feeds. 14

This is two governments imposing enforceable rules on platforms to protect children. Compliance is not optional. Children are still developing the judgement needed to assess risks and consequences. And the companies cannot be trusted to volunteer what they know: in 2021, Frances Haugen testified that Facebook withheld information and misled the public about what its own research revealed about children’s safety. 15

As an adult, I get to make that judgement for myself. Being allowed to accept a risk does not make it disappear. 16 Nor does avoiding social media mean abandoning the internet. Australia’s rules distinguish social media from messaging, gaming, professional engagement, education and health support, which have exemptions. 17 I can use those services without spending my time in a social media feed.

Most people would avoid a dark alley in a shady part of town, even knowing they would probably get through unharmed. The small chance of something going badly wrong is enough to make them take another route. Yet with social media, we often set that caution aside. That is easier to do when the company selling you the service keeps the risks out of sight.

For me, social media doesn’t offer enough to justify the time and the risks. I can make that decision without waiting for a ban, or asking anyone else to make it with me.

3 I don’t trust platforms with my data

I used to say I didn’t post much (or at all). I find it less reassuring now, because posting is only a small part of what an account allows a platform to do.

For example, round 53 Australians installed an app called This Is Your Digital Life. Yet the Australian Information Commissioner’s case concerned information about more than 311,000 Australians. Almost all of them hadn’t installed the app. Their Facebook friends had, and that was enough to expose their information. Meta settled the proceedings for $50 million in December 2024. 1819

Another case concerned Facebook receiving browsing histories from other websites after users logged out. The US Court of Appeals reinstated the claims in 2020, and Facebook settled for US$90 million in 2022. 20

In 2025, a California jury found Meta liable for receiving reproductive health information sent by a period-tracking app. The information included cycle dates, sexual activity and pregnancy intentions. The class covered roughly 38 million people. These cases involved different laws, and settlements are not admissions of liability. 2122

Then theres LinkedIn’s connection experiments - and they bother me for a different reason. Between 2015 and 2019, the company changed the mix of connections it recommended to about twenty million members, then measured the effect on employment. Members weren’t told that they were in an experiment. LinkedIn said the work was consistent with its user agreement. 2324

The experiments produced useful research. They weren’t found unlawful. 25 But history is full of experiments that pursued useful knowledge while harming the people involved. Useful results do not excuse how researchers obtain them. Modern research ethics requires attention to the risks people bear and their freedom to choose whether to participate. 26

Having an account does not mean agreeing to become an experimental subject. And yes, there is the predictable answer: “You accepted the terms.” Even where those terms mention experiments, burying permission in a document people barely read does not establish informed consent. In a study of 543 people joining a fictional social network, 98% missed outrageous clauses, including one requiring their first-born child as payment. 27 A company can point to a clause without answering whether anyone understood what they were agreeing to. I signed up to find professional contacts. If you want to experiment with something that can affect my job prospects, explain what you are doing and give me a choice. Pointing me back to the small print does not excuse keeping me uninformed.

In these examples not posting wouldn’t have kept those members out of the experiments. Nor would it have protected the “friends” of the Australians who installed that app.

For me, I know I can’t control every way a company collects information about me. But, I can decide whether to keep an account with it. That is a more useful choice than deciding how often to post.

4 Social media isn’t about people anymore

In April 2025, Imperva reported that automated programs had overtaken people as the source of most web traffic in its measurements. Bots accounted for 51% of traffic during 2024, exceeding human activity for the first time in a decade. 28 Its next report estimated that the share reached 53% in 2025. 2930

There are bots designed to pass as people, though. Ferrara and colleagues described their methods and influence a decade ago. 31

Social media is supposed to connect people. If I am talking to software pretending to be a person, that defeats the purpose. Being shown an advert is not connection either. It is marketing. We recognise that distinction on television, where a commercial break interrupts what we came to watch. Ad-free streaming offers a way to avoid those interruptions. [^46] Putting an advert between friends’ posts does not turn it into human contact - nor does it offer a way to exclude being marketed to.

Imagine speaking in a dark room without knowing whether anyone is listening. Something answers, but you cannot tell whether it is another person, a recording or a sales pitch. How long would you keep talking? Yet on social media, those who post, post into that uncertainty and assume they have reached someone.

Me, I can choose which conversations to join and which to walk away from - I decide which ones I want no part of. A social media feed gives the platform a say in what reaches me. I can block an account after it shows me something ugly, but I have already seen it. I wanted contact with people I chose, yet I get whatever else the platform decides to put in front of me.

5 Finding something worth reading takes too much work

I am always looking for something new to learn. But on social media, more and more of that effort goes into separating useful knowledge from unsupported opinions and manufactured misinformation. I want to spend my time learning, not checking whether someone has simply made something up.

Social media companies are actively encouraging more generated content. In September 2025, Meta introduced Vibes, a feed of AI-generated videos within its Meta AI app. It lets users create and remix videos, then share them to Instagram and Facebook Stories and Reels. 32 The company is making it easier to produce more material for its platforms. I still have to do the work of finding something worth my time. And more slop makes that harder.

AI assistance doesn’t tell me whether a piece has anything to say. A person can use it to edit an account of something they actually did, fix grammar, paraphrase, etc. But generating another article from material already online doesn’t, by itself, give us another person’s experience - it just adds to the noise.

Shumailov and colleagues found that models trained repeatedly on generated data can lose the rare and unusual parts of the original data. 33 That is a finding about training models, not proof that a social media feed is doing the same thing. It does help explain my concern about relying on recycled material.

The more immediate problem is time. Herbert Simon was already writing about the shortage of attention caused by too much information in 1971. 34 Then came the web, putting libraries’ worth of information from around the world within reach. Social media added feeds that keep delivering more, without waiting for us to finish what is already there. Now AI lets people generate another batch in seconds. Each stage adds more material and speeds up its arrival. But there is only so much I can read, however cheap it becomes for others to produce more.

Qiu and colleagues modelled a network where people had limited attention and memory. As the volume of information increased, the system became worse at separating good material from bad. Under realistic conditions in their model, quality offered little advantage in determining what spread. 3536

A study of product-review searches on Google, Bing and DuckDuckGo found that pages earning commissions from sales appeared much more often in the results than their share of reviews on the web suggested. All three engines were vulnerable to link spam. 3736 The study examined search results. The parallel I see with social media is that content can be very good at getting in front of me without having much to offer once it does.

I want to learn something, and I resent how much time goes into sorting through rubbish before I can. I’d rather spend that time reading a book or a worthwhile article than keep scrolling in the hope that something useful turns up.

6 Social media pushes people into opposing camps

The familiar explanation is that social media traps us in echo chambers. I don’t think that explains the whole problem.

The proverbial “algorithm” is designed to keep us engaged. Facebook describes ranking posts by predicting what people will interact with, using their previous behaviour to inform those predictions. 38 What I readily accept is useful to that process: show me more of what I already agree with, and I have an easy reason to keep reading.

But that selection also determines what I miss. Cinelli and colleagues found that users on Facebook and Twitter clustered with like-minded people, with information spreading mainly within those groups. 39 Facebook’s own researchers found that feed ranking further reduced exposure to opposing news, although users’ choices contributed more to that reduction. 40 An alternative view does not have to be banned to become harder to find. I cannot judge an issue properly if important parts of it never reach me.

The effects are not as simple as “more agreement means more polarisation.” Researchers reduced like-minded content by about a third for 23,377 Facebook users during the 2020 US election. They found no detectable change in eight measures, including polarisation and belief in false claims. 4142

Nor does disagreement automatically produce better thinking. Bail’s experiment exposed participants to opposing political views through a Twitter bot. Republican participants subsequently reported more conservative opinions. 43 That does not tell me they failed to think. Considering an argument does not oblige anyone to accept it. The question is whether the way disagreement reaches us encourages consideration or simply another reaction.

Agreement is not the only way to hold attention. A study published in 2025 found that Twitter’s engagement-based ranking amplified emotionally charged posts and hostility towards political opponents compared with a chronological feed. Users did not prefer the political posts the algorithm selected. 44 Törnberg’s model offers another explanation for division: digital media can sort people into political camps through interaction with opponents, rather than isolation from them. 45

That is why I see “the algorithm” as hostile to critical thinking. I need to question familiar conclusions, consider other explanations and decide whether the evidence holds up. A feed that keeps rewarding agreement and provoking arguments encourages me to react before I do that work.

The algorithm is not about expanding thought. It is about locking attention, and it does that by curbing thought: narrowing what reaches me, reinforcing what I already accept and serving disagreement as something to fight. I want to examine an idea on its merits. I don’t want my thinking reduced to choosing a camp and defending it.

7 LinkedIn doesn’t get an exception

People usually assume I must at least use LinkedIn. I spent thirty years in senior roles making hiring decisions. That experience is part of why I don’t.

Start with the jobs themselves. A professional-looking advert doesn’t establish that a vacancy exists, or that the recruiter represents the company. The FTC warns about fake recruiters on LinkedIn who seek money and personal details that can be used for identity theft. 46 Recruitment can also provide cover for gathering sensitive information. In June 2026, the Five Eyes intelligence agencies warned about Chinese intelligence officers using fake companies and recruitment on professional networking sites. 47

A fake vacancy could also reveal which employees at a competitor are looking to leave. That is a possible abuse, rather than something I can establish from an advert. The applicant reveals information before knowing whether anyone actually intends to hire them.

On the other side of the process, a profile tells me what someone wants an employer to see. Even an entirely truthful work history leaves important questions unanswered. What did you contribute? What went wrong? What did you do about it?

Six thousand connections do not mean six thousand satisfied colleagues, managers or clients. The count doesn’t distinguish people who know your work from strangers accepting invitations for reach. It doesn’t tell me whether invitations were personal or automated. I need to know what those people can actually say about your work.

Written recommendations have the same weaknesses as ordinary references. Candidates choose people willing to recommend them. Those people might be friends, or colleagues who like them but know little about their work. Aamodt’s review describes references that are overwhelmingly favourable and often disagree with one another. 4849 LinkedIn makes the praise public. I don’t see why an audience, including the candidate, would encourage greater candour.

Nor does a busy feed establish expertise. Marketing spiel and regurgitated AI slop can keep a name in circulation without telling me whether its owner can do the job. If I am hiring, I need evidence of work, not evidence that someone keeps posting.

AI adds another problem. Applicants can generate tailored applications and send them at a scale that makes individual assessment harder. Greenhouse reports that applications per job more than doubled between 2022 and 2025, while vacancies took longer to fill. 5051 More applications arrive, but that doesn’t mean employers learn more about the people applying.

Automated filtering is supposed to manage the flood. But when screening depends on matching keywords and rigid requirements, relevant skills can disappear because someone describes them differently. Harvard’s Hidden Workers report describes hiring processes that systematically exclude qualified people. 5253 AI-written applications can also gain an advantage unrelated to ability. In a controlled experiment, AI models favoured résumés they generated themselves over otherwise equivalent human-written versions. 5455 The wording changes. The candidate’s ability doesn’t.

That is where digital recruitment loses its value for me. More adverts, more applications and more polished claims are no improvement if employers find it harder to identify who can actually do the work. Relevant experience gets buried in the flood, while an application written to satisfy the filter gets through. I see a process becoming worse at its purpose: finding the right person for the job.

The same confusion between presentation and ability appears in the problem I discussed in my writing about neurodiversity and employment. 56 An autistic candidate can have stronger skills and still lose out because their social manner feels less familiar. Whelpley and May found that non-autistic candidates received better ratings when evaluators watched mock interviews. When evaluators read the transcripts instead, the autistic candidates received better ratings. The answers were the same. The social presentation changed the judgement. 57

That is a serious failure in how we select people. Someone can be awkward in conversation and excellent at the work. Someone else can be charming and contribute very little. Communication matters where the job requires it, but judging the relevant skill takes more than deciding whether I like the person. Hiring for social acceptance can exclude the candidate who would add more value.

LinkedIn can introduce people to employers, and the earlier experiments found that connections helped people move between jobs. 2358 But I think its reputation now exceeds its value. Being buried in a flood of profiles and applications does nothing for a capable candidate. It does nothing for an employer struggling to find that candidate either. The promise was useful professional connections. When making those connections requires sorting through this much noise, much of that value disappears. I don’t use LinkedIn because I no longer see enough professional benefit, or validity, to justify the effort of being there.

8 A “friend” count doesn’t tell me who I can rely on

Dunbar asked two UK samples about their Facebook friendships. Average friend counts were about 155 and 183. People considered roughly a quarter of those genuine friends. They could turn to about four in a crisis and fourteen for sympathy. Those smaller circles resembled the ones found offline. 59

Those numbers make me question what social media actually adds to friendship. Hundreds of names, but still only a handful of people to turn to when things go wrong. Keeping in touch has value. But adding someone to a list doesn’t create the relationship that makes them willing to help. That takes time together. Hall’s research describes the hours people spend together as they become closer friends. 60 Clicking “Add friend” doesn’t do that work.

Turkle writes about the trade between more contact and less connection. 61 For the friendships I value, I don’t see what a public feed adds. Knowing someone well, trusting them and being there when they need me takes attention and effort. The platform doesn’t supply any of that.

What social media does make easy is an exchange: react to my post and I’ll react to yours; share my work and I’ll promote yours. Researchers distinguish relationships built around repayment from relationships where people respond to one another’s needs. 62 If a connection depends on what each person gets back, it can fall apart when that exchange changes. Stop providing attention, access or favours, and the reason for keeping you around can disappear. I want friendships that survive the times when one of us has nothing to offer.

Meanwhile, the feed competes for the attention those friendships need. Someone’s news arrives mixed with adverts, arguments and irrelevant posts. Keeping up with that stream takes time without necessarily bringing me closer to anyone. I don’t want to mistake watching people’s updates for taking an interest in their lives. I would rather call a friend, listen properly and make time for them.

What I willingly give up

Leaving social media costs me access to people and communities. For autistic adults, that can be a serious loss.

Van Driel and colleagues describe what autistic adults value in these spaces: text instead of faces, time to compose an answer, and contact with people like them who aren’t available nearby. 63 I recognise those needs. They are reasons social media could work for me too. But for me, the cons far outweigh the pros.

I can’t wave that away by telling people to find a better place. A Discord chat or a forum only helps if the people you need are there. For some, the alternative to the platform is being alone.

What I want from those spaces is the conversation and the belonging. Research on autistic community connectedness finds value in particular friendships, shared understanding and shared political purpose. 64 Those things can also exist in smaller groups, without a public feed. Smaller spaces have their own problems, and I am not claiming that they are automatically safe. 65

That is the compromise I prefer. Other people will make a different one, especially if leaving means losing their community.

I write here instead. The pieces are long, and fewer people are likely to find them than if I put them in a feed. But anyone can read them without an account. I don’t have to maintain a profile or keep up with a stream of other people’s updates to publish something.

I still want to write, and I still want people in my life. But I also don’t want social media enough to accept what comes with it.

⁂

  1. Gerber, J. P., Wheeler, L., & Suls, J. (2018). A social comparison theory meta-analysis 60+ years on. Psychological Bulletin, 144(2), 177–197. DOI: 10.1037/bul0000127 ↩

  2. McCarthy, P. A., & Morina, N. (2020). Exploring the association of social comparison with depression and anxiety: A systematic review and meta-analysis. Clinical Psychology & Psychotherapy, 27(5), 640–671. DOI: 10.1002/cpp.2452 ↩

  3. Verduyn, P., Lee, D. S., Park, J., Shablack, H., Orvell, A., Bayer, J., Ybarra, O., Jonides, J., & Kross, E. (2015). Passive Facebook usage undermines affective well-being: Experimental and longitudinal evidence. Journal of Experimental Psychology: General, 144(2), 480–488. DOI: 10.1037/xge0000057

    The longitudinal precursor is Kross, E., Verduyn, P., Demiralp, E., Park, J., Lee, D. S., Lin, N., Shablack, H., Jonides, J., & Ybarra, O. (2013). Facebook use predicts declines in subjective well-being in young adults. PLoS ONE, 8(8), e69841. DOI: 10.1371/journal.pone.0069841 ↩

  4. Braghieri, L., Levy, R., & Makarin, A. (2022). Social media and mental health. American Economic Review, 112(11), 3660–3693. DOI: 10.1257/aer.20211218 ↩

  5. The study uses the timing of Facebook’s rollout rather than random assignment. Its conclusion depends on that timing being unrelated to other mental health trends. The authors defend this assumption in the paper. Their data covers US college students in 2004–2006, so it does not describe every population or today’s Facebook. ↩

  6. Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). The welfare effects of social media. American Economic Review, 110(3), 629–676. DOI: 10.1257/aer.20190658 ↩ ↩2

  7. Hunt, M. G., Marx, R., Lipson, C., & Young, J. (2018). No more FOMO: Limiting social media decreases loneliness and depression. Journal of Social and Clinical Psychology, 37(10), 751–768. DOI: 10.1521/jscp.2018.37.10.751 ↩ ↩2

  8. Orben, A., & Przybylski, A. K. (2019). The association between adolescent well-being and digital technology use. Nature Human Behaviour, 3(2), 173–182. DOI: 10.1038/s41562-018-0506-1 ↩

  9. Valkenburg, P. M., Meier, A., & Beyens, I. (2022). Social media use and its impact on adolescent mental health: An umbrella review of the evidence. Current Opinion in Psychology, 44, 58–68. DOI: 10.1016/j.copsyc.2021.08.017 ↩

  10. Odgers, C. L. (2024). The great rewiring: Is social media really behind an epidemic of teenage mental illness? Nature, 628(8006), 29–30. DOI: 10.1038/d41586-024-00902-2 ↩

  11. The studies identify passive use and unfavourable social comparison as concerns. Applying those findings to myself is an inference. None of the studies cited here reports a separate effect for autistic or ADHD adults. ↩

  12. Australian Government Department of Infrastructure, Transport, Regional Development, Communications, Sport and the Arts. Social media minimum age. https://www.infrastructure.gov.au/media-communications/internet/online-safety/social-media-minimum-age - the obligation was inserted into the Online Safety Act 2021 (Cth) by amendment in November 2024. Government material rather than peer-reviewed research. ↩

  13. eSafety Commissioner. Social media age restrictions. https://www.esafety.gov.au/about-us/industry-regulation/social-media-age-restrictions. On the commencement date and the initial list of age-restricted services, see also University of Sydney. (2025, December 5). What is Australia’s under-16 social media ban? The world-first law explained. https://www.sydney.edu.au/news-opinion/news/2025/12/05/what-is-australias-under-16-social-media-ban-the-world-first-law-explained.html ↩

  14. Ofcom. New rules for a safer generation of children online. https://www.ofcom.org.uk/online-safety/protecting-children/new-rules-for-a-safer-generation-of-children-online - the Protection of Children Codes under the Online Safety Act 2023 (UK) took effect on 25 July 2025, following a children’s risk assessment deadline of 24 July 2025. ↩

  15. Haugen, F. (2021, October 4). Statement of Frances Haugen. Written testimony for the US Senate Committee on Commerce, Science, and Transportation’s October 5 hearing. On page 2, Haugen states that Facebook withheld information and misled the public about its research on children’s safety. Written testimony. ↩

  16. An age restriction does not mean that an activity becomes safe above the cutoff. It also does not prove that everyone above it is harmed. Being allowed to use a platform leaves me free to decide that its risks are not worth taking. ↩

  17. The eSafety Commissioner maintains the list of restricted platforms. The exemptions for messaging, gaming, professional networking, education and health support appear in the 2025 Rules. ↩

  18. Office of the Australian Information Commissioner. (2024, December 17). Landmark settlement of $50m from Meta for Australian users impacted by Cambridge Analytica incident. https://www.oaic.gov.au/news/media-centre/landmark-settlement-of-$50m-from-meta-for-australian-users-impacted-by-cambridge-analytica-incident ↩

  19. UNSW Sydney. (2024, December). Tech giant Meta will pay Australians $50 million for enabling the Cambridge Analytica scandal. https://www.unsw.edu.au/newsroom/news/2024/12/meta-pay-australians-50million-cambridge-analytica - the figures of 53 installers and approximately 311,074 affected friends are as reported there and in the Commissioner’s proceedings. ↩

  20. In re Facebook, Inc. Internet Tracking Litigation, 956 F.3d 589 (9th Cir. 2020), decided 9 April 2020. Opinion at https://cdn.ca9.uscourts.gov/datastore/opinions/2020/04/09/17-17486.pdf. The case settled for US$90 million in 2022. ↩

  21. Frasco v. Flo Health, Inc., N.D. Cal. Jury verdict of 1 August 2025 finding Meta liable under the California Invasion of Privacy Act; the court declined to decertify the class or set aside the verdict on post-trial motions in September 2025. ↩

  22. These proceedings involve different Australian and US laws. The Australian case and the browsing-history case ended in settlements, which are not admissions of liability. The Flo case involved a jury finding. They do not establish that any particular reader’s information was misused. ↩

  23. Rajkumar, K., Saint-Jacques, G., Bojinov, I., Brynjolfsson, E., & Aral, S. (2022). A causal test of the strength of weak ties. Science, 377(6612), 1304–1310. DOI: 10.1126/science.abl4476 ↩ ↩2

  24. The scale of the experiments, and the fact that participants were not informed, were reported in The New York Times. (2022, September 24). LinkedIn ran social experiments on 20 million users over five years, syndicated at https://www.forbesindia.com/article/news/linkedin-ran-social-experiments-on-20-million-users-over-5-years/80083/1. LinkedIn said it had acted consistently with its user agreement; its privacy policy did not state that it conducted experiments on members. Catherine Flick of De Montfort University described the exercise as closer to corporate marketing than to research, and Michael Zimmer of Marquette University raised the question of long-term consequences for people whose job prospects had been varied. Press reporting rather than peer-reviewed research. ↩

  25. The LinkedIn experiments were not found unlawful. My objection concerns participation without separate notice, rather than an allegation of illegal conduct. ↩

  26. National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research. (1979). The Belmont Report. Discusses historical research abuses, the assessment of risks and benefits, and informed consent through information, comprehension and voluntary participation. Report. ↩

  27. Obar, J. A., & Oeldorf-Hirsch, A. (2020; first published online in 2018). The biggest lie on the Internet: Ignoring the privacy policies and terms of service policies of social networking services. Information, Communication & Society, 23(1), 128–147. Experimental study of 543 participants joining a fictional service. Authors’ abstract DOI: 10.2139/ssrn.2757465. ↩

  28. McKeever, G. (2025, April 15). 2025 Imperva Bad Bot Report: How AI is Supercharging the Bot Threat. Reports that automated traffic accounted for 51% of web traffic in 2024, surpassing human activity for the first time in a decade. Imperva. ↩

  29. Imperva. (2026, April 29). Bad Bot Report 2026: Bots in the agentic age. https://www.imperva.com/blog/bad-bot-report-2026-bots-agentic-age/ - automated traffic at 53% of web traffic in 2025, against 51% in 2024. ↩

  30. Imperva measures traffic across its own customers. It is a commercial security vendor, and this is not peer-reviewed research. Web traffic includes activity outside social media. The 53% figure cannot establish the proportion of bots in a feed. ↩

  31. Ferrara, E., Varol, O., Davis, C., Menczer, F., & Flammini, A. (2016). The rise of social bots. Communications of the ACM, 59(7), 96–104. DOI: 10.1145/2818717 ↩

  32. Meta. (2025, September 25). Introducing Vibes: A New Way to Discover and Create AI Videos. Announces the AI video feed and tools for sharing videos to Instagram and Facebook Stories and Reels. Announcement. ↩

  33. Shumailov, I., Shumaylov, Z., Zhao, Y., Papernot, N., Anderson, R., & Gal, Y. (2024). AI models collapse when trained on recursively generated data. Nature, 631(8022), 755–759. DOI: 10.1038/s41586-024-07566-y ↩

  34. Simon, H. A. (1971). Designing organizations for an information-rich world. In M. Greenberger (Ed.), Computers, Communications, and the Public Interest (pp. 38–71). The Johns Hopkins Press. The passage quoted is at pp. 40–41: “a wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it.” ↩

  35. Qiu, X., Oliveira, D. F. M., Sahami Shirazi, A., Flammini, A., & Menczer, F. (2017). Limited individual attention and online virality of low-quality information. Nature Human Behaviour, 1(7), 0132. DOI: 10.1038/s41562-017-0132 ↩

  36. Qiu and colleagues studied a simulation calibrated against platform data. Bevendorff and colleagues measured product-review searches. Neither study measures the quality of a particular social media feed. ↩ ↩2

  37. Bevendorff, J., Wiegmann, M., Potthast, M., & Stein, B. (2024). Is Google getting worse? A longitudinal investigation of SEO spam in search engines. In Advances in Information Retrieval (Lecture Notes in Computer Science, pp. 56–71). Springer. DOI: 10.1007/978-3-031-56063-7_4 ↩

  38. Lada, A., Wang, M., & Yan, T. (2021, January 26). How Does News Feed Predict What You Want to See? Facebook describes its ranking predictions, including engagement and users’ previous interactions, alongside survey measures of perceived value. Facebook’s explanation. ↩

  39. Cinelli, M., De Francisci Morales, G., Galeazzi, A., Quattrociocchi, W., & Starnini, M. (2021). The echo chamber effect on social media. Proceedings of the National Academy of Sciences, 118(9), e2023301118. DOI: 10.1073/pnas.2023301118 ↩

  40. Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. The authors worked at Facebook. DOI: 10.1126/science.aaa1160. ↩

  41. Nyhan, B., Settle, J., Thorson, E., Wojcieszak, M., Barberá, P., Chen, A. Y., Allcott, H., et al. (2023). Like-minded sources on Facebook are prevalent but not polarizing. Nature, 620(7972), 137–144. DOI: 10.1038/s41586-023-06297-w

    The companion algorithm experiment is Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., et al. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 381(6656), 398–404. DOI: 10.1126/science.abp9364 ↩

  42. The Meta-partnered election studies used the company’s cooperation and data. That access enabled the research but limits independent scrutiny. Three months during one US election also cannot settle questions about effects over years. ↩

  43. Bail, C. A., Argyle, L. P., Brown, T. W., Bumpus, J. P., Chen, H., Hunzaker, M. B. F., Lee, J., Mann, M., Merhout, F., & Volfovsky, A. (2018). Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37), 9216–9221. DOI: 10.1073/pnas.1804840115 ↩

  44. Milli, S., Carroll, M., Wang, Y., Pandey, S., Zhao, S., & Dragan, A. D. (2025). Engagement, user satisfaction, and the amplification of divisive content on social media. PNAS Nexus, 4(3), pgaf062. DOI: 10.1093/pnasnexus/pgaf062. ↩

  45. Törnberg, P. (2022). How digital media drive affective polarization through partisan sorting. Proceedings of the National Academy of Sciences, 119(42), e2207159119. DOI: 10.1073/pnas.2207159119 ↩

  46. Federal Trade Commission. (2023, August 8). Scammers impersonate well-known companies, recruit for fake jobs on LinkedIn and other job platforms. Warns about fake recruiters seeking money and personal information that can be used for identity theft. FTC warning. ↩

  47. ASIO, CSIS, FBI, MI5, & NZSIS. (2026, June 3). Safeguarding Our Secrets. Joint bulletin warning about Chinese intelligence officers using fake companies and recruitment on professional networking sites to target people with access to sensitive information. Five Eyes bulletin. ↩

  48. Aamodt, M. G. (2006, February). Validity of recommendations and references. Assessment Council News, 4–6. https://maamodt.asp.radford.edu/Research%20-%20IO/2006-Feb-References.pdf ↩

  49. This source is a professional association’s technical newsletter, not a peer-reviewed journal. It summarises Aamodt’s work on leniency and agreement between referees. His citations identify the underlying research. ↩

  50. Greenhouse. (2026, March). The hire standard. https://www.greenhouse.com/recruiting-benchmarks ↩

  51. Greenhouse reports that applications per job increased from 116 in 2022 to 244 in 2025, while time to fill increased from approximately 44 to 60 days. These figures describe its customer data and do not establish that AI alone caused the changes. ↩

  52. Fuller, J. B., Raman, M., Sage-Gavin, E., & Hines, K. (2021). Hidden workers: Untapped talent. Harvard Business School & Accenture. https://www.hbs.edu/managing-the-future-of-work/research/hidden-workers-untapped-talent ↩

  53. The report describes how hiring processes and automated screening exclude qualified workers whose backgrounds do not match employers’ narrowly defined requirements. It predates the recent growth of generative AI and is not specifically about LinkedIn. ↩

  54. Xu, J., Li, G., & Jiang, J. Y. (2025). AI self-preferencing in algorithmic hiring: Empirical evidence and insights. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(3), 2757–2758. https://doi.org/10.1609/aies.v8i3.36755 ↩

  55. In a controlled résumé experiment, AI models favoured their own generated versions over otherwise equivalent human-written or other models’ versions. The experiment tests model behaviour rather than measuring actual hiring outcomes. ↩

  56. Rodrigues, S. (2025). Modernising HR for variant minds: Autism - The monotropic mind. Variant Minds Publications. ↩

  57. Whelpley, C. E., & May, C. P. (2023; first published online in 2022). Seeing is Disliking: Evidence of Bias Against Individuals with Autism Spectrum Disorder in Traditional Job Interviews. Journal of Autism and Developmental Disorders, 53(4), 1363–1374. Evaluators rated non-autistic candidates more favourably when watching mock interviews, but autistic candidates more favourably when reading the transcripts. DOI: 10.1007/s10803-022-05432-2. ↩

  58. Two authors worked for LinkedIn, and the data is proprietary. Independent researchers cannot reproduce the study from publicly available data. I accept its finding about job mobility while keeping those limits in view. ↩

  59. Dunbar, R. I. M. (2016). Do online social media cut through the constraints that limit the size of offline social networks? Royal Society Open Science, 3(1), 150292. DOI: 10.1098/rsos.150292 ↩

  60. Hall, J. A. (2019). How many hours does it take to make a friend? Journal of Social and Personal Relationships, 36(4), 1278–1296. DOI: 10.1177/0265407518761225 ↩

  61. Turkle, S. (2012). Alone together: Why we expect more from technology and less from each other. Basic Books. ISBN 978-0-465-03146-7 ↩

  62. Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal relationships. Journal of Personality and Social Psychology, 37(1), 12–24. https://doi.org/10.1037/0022-3514.37.1.12 ↩

  63. van Driel, M., Vines, J., Barros Pena, B., & Koteyko, N. (2023). Understanding autistic adults’ use of social media. Proceedings of the ACM on Human-Computer Interaction, 7(CSCW2), Article 257. DOI: 10.1145/3610048 ↩

  64. Botha, M., Dibb, B., & Frost, D. M. (2022). ‘It’s being a part of a grand tradition, a grand counter-culture which involves communities’: A qualitative investigation of autistic community connectedness. Autism, 26(8), 2151–2164. DOI: 10.1177/13623613221080248 ↩

  65. Small groups can also expose people to harassment or isolation. Choosing a space without a public feed does not guarantee safety or belonging. ↩