What can analogies about children’s digital safety tell us?

Not so long ago as I was listening to presentations and discussions at what was perhaps my 50^th event dedicated to online safety, it occurred to me that many of the analogies or metaphors people tend to reach for during discussions are remarkably stable. Regardless of how much technology and the way children interact with it has changed over time, the field keeps reaching for familiar shorthand.

Human mind thrives on analogies. They help to lower ambiguity of the situation or phenomenon at hand by transferring a familiar relational structure onto an unfamiliar one [i], reduce the amount of analytical or “computational” power required for deliberation by substituting an easier judgment for a harder one [ii], and can sometimes lower the level of contention that surrounds the topic to allow for an agreement to be achieved among parties with different views on a matter [iii]. Analogies also help to link new or unfamiliar issues to a domain of experience that feels familiar.

However, relying on analogies has a downside. They work by spotlighting certain features of an issue while hiding others, and those framing choices can shape how the public, political actors and other stakeholders think about the problem. Once an analogy becomes the default or a dominant way people talk about an issue, it hardens into a particular framework of meaning. That framework also has a potential to crowd out other, equally valid, ways of understanding that issue. Analogies we use about technology do not just describe the field. They have the potential to set research agenda and steer policy direction [iv], which is exactly why it’s important to scrutinise the specific language choices.

Research indicates that analogical reasoning isn't a failure to think rigorously so much as a structurally cheaper way to conduct causal assessment. Analogy offers a shortcut that lets regulators, platforms, civil society and researchers, who do not always agree on the underlying causal mechanism, still build a shared vocabulary and work towards a course of action.

However, the trade-off shows up in the fact that the analogy imports the source domain's regulatory toolkit and evidentiary confidence even if the underlying mechanism does not actually transfer. This is precisely why the discussions on child online protection tend to reach for the same handful of comparisons year after year even though technology and evidence base have moved on.

Below is the list of analogies I observed. These are deliberately not labeled as good or bad. Instead the focus in on potential consequences of each specific framing.

“Social media must have a health warning label like cigarettes”

This analogy surfaced particularly prominently and gained traction after being used by the US Surgeon General in 2024[iv] and later repeated by the Danish Prime Minister[v] and politicians in the UK[vi]. However, it can be traced at least back to 2018 when it was used by the Salesforce CEO[vii]. At its core it suggests that platforms are profit-driven industries that knowingly target vulnerable youth with a harmful product and therefore would benefit from a warning label similar to those now seen on most, if not all, tobacco products.

This is an attractive analogy for several reasons. Cigarette warnings are now uncontroversial, which makes resistance look bad by association. It also proposes something specific, visible, and produces a concrete legislative follow-through.

However, that is where the utility of this metaphor ends. Social media or any other digital product is not a substance one can ingest. While there is evidence that notifications and "likes" have an effect on the brain's reward system, especially in people who use social media more heavily [viii], it not comparable to the effects of nicotine. The brain releasing a little reward signal during any enjoyable activity (eating, talking, exercising) is normal, not a sign of harm. Critics have indeed argued that the comparison borrows tobacco's evidentiary base rhetorically without the underlying science to support it [ix].

Lastly, the analogy misinterprets what kind of warnings have been shown to work online. A cigarette label isn't a one-off notice as it's seen every time the pack is opened, so repetition at the point of use is part of its power. However, the digital measures with the best evidence aren't generic, always-on messages that only appear at the point of access. They tend to be targeted, interruptive prompts that ask to reconsider an offensive comment [x], do a fact-check before sharing [xi], or pause before sending an intimate image [xii]. So the cigarette comparison works as a general signal that platforms bear responsibility for harm, but as a literal blueprint it points to the weaker tool.

“We do not allow children to buy alcohol without checking their age”

Wielded most often by proponents of age verification or minimum-age access requirements, this analogy points to the fact that alcohol is almost universally treated as an age-restricted, harmful-if-misused product, and argues that social media warrants the same [xiii]^,. The alcohol comparison also recurs in the debates themselves – a supporter of the French minimum age bill argued the state must "protect our children, just as we protect our children from drinking alcohol." [xiv]

As with tobacco, the appeal is that age limits on known harmful products are broadly accepted across different legal and cultural systems, so the analogy travels easily and points to a concrete, familiar mechanism that includes an age verification and restriction of access. It also proposes something enforceable and administratively tidy.

That apparent tidiness is the first thing worth questioning, because the consensus the analogy relies on is less clear-cut than it seems. Interestingly, that is the aspect the debates seem to overlook despite it being remarkably relevant to the issue at hand. Societies broadly agree that alcohol should be age-controlled, but not on much beyond that. The threshold ranges from 16 in parts of Europe to 21 in the United States and elsewhere. Many countries set different ages for different categories of alcohol, and most laws regulate purchase and public consumption rather than consumption itself. "Restrict it like alcohol" therefore invokes not a single rule but a loose family of them, distinguishing buying from using, public from private, and one drink from another. The same unresolved questions sit beneath most proposals for age-based restrictions on social media. Does it apply to public feeds, or private messaging too? Are all platforms or only some covered? Do we draw the line at 13, 15,16 or 18? And how much of it comes down to the role of parents? In this sense the analogy may be more apt than it first appears but it is rarely used that way.

Applied as it commonly is, the comparison also overlooks that alcohol regulation is not just an age line. In most countries it relies on an accountability chain which includes licensed sellers who are liable for serving minors, advertising restrictions, and retailer penalties. In many instances it also includes bystander awareness backed by the near-universal acceptance of the existing age limit .

So while the analogy usefully establishes that age-gating a harmful product is legitimate, what it understates is how much harder that gate is to build and enforce when the "point of sale" is every connected device at once, and that the digital version becomes reliable only by constructing an identity layer whose costs the alcohol comparison never has to reckon with.

“We require seat belts in cars nowadays so we need an equivalent for the internet”

Where the tobacco and alcohol analogies argue about the inherent harmfulness of a product, the seat-belt analogy sidesteps that fight entirely. Instead it claims that a broadly beneficial technology was released faster than the safety infrastructure around it, and that the answer is to build safeguards without removing the underlying activity or altering the nature of the product.

What this commonly points to is a concrete set of measures such as limits on autoplay, push notifications and infinite scroll, as well safer defaults for younger users. Commentators frequently invoke a "seatbelt moment," recalling that mandatory belts on vehicles were fiercely resisted before becoming unquestioned, to imply that today's objections will look similarly quaint after a period of time.

The appeal of this analogy is that it does not require proving social media is intrinsically harmful, only that it carries avoidable risks a responsible designer could mitigate. This is an easier claim to defend, and one the belt's now-uncontroversial status reinforces by making resistance look reactionary.

However, it is important to remember that it is not just the belt on its own that makes driving safer. It is part of the "safe systems" approach to road safety, which accepts that people make mistakes and designs roads, vehicles and rules to accommodate that error rather than punish it [xv]. Even behind the seatbelt itself sits a system with detailed safety standards, crash-testing against agreed measures of harm, independent approval, and laws requiring people to buckle up with penalties imposed for non-compliance. The seatbelt analogy borrows the reassuring image of the device itself while dropping the system that makes it work. Current safety-by-design standards and guidelines, on the other hand, tend to be principle-based. With rare exception they name the end goal such as "the best interests of the child a primary consideration" or "high privacy by default" rather than specific testable features that can be verified in a lab. While some parts of the codes, such as switching on geo-location or making an account private by default, are specific and reviewable, the dialogues that invoke the seatbelt analogy do not use this degree of specificity. As a result they do not offer an engineer anything concrete to build or a regulator anything firm to test. The analogy lends the authority of a precise safety standard to what is more of a statement of intent.

Even more crucial for the this analogy and its relevance is the fact that car-safety testing depends on two things a social media feed cannot offer. The first is a clear measure of harm. A crash test works because injuries can be measured against an agreed threshold, but there is no agreed way to measure whether and how a particular feed may harm someone (the underlying question is still disputed and is not part of this overview). The second is availability of relatively static conditions to test. It is possible crash one representative car and approve the model, whereas a feed is personalised and constantly changing, so there is no single version to examine. It would not be feasible to copy the car-testing model, because it has neither a clear measure of harm nor a stable object to test.

Lastly, it is worth remembering that the seat belt does not protect everyone equally. Belted women face significantly higher odds of serious injury than belted men in similar crashes, largely because car safety was designed and tested around a male body [xvi]^,[xvii]. Similar patterns are already visible in the digital safeguards that are being adopted. Facial age-estimation, for example, has been found to be less accurate for non-white and female-presenting users near the age indicated in the policies [xviii], and other measures will likewise work better for some groups than others. This does not defeat the analogy, but it completes it. If we are to adopt the seat belt as a model, we should adopt it fully. This includes the lesson that coma with it and the fact that a safeguard which works "on average" can systematically fail whoever was not the reference case.

The key value of the seatbelt analogy as well as its main appeal is the central question of "does this need guardrails?" rather than "is the product bad?". However, its persuasive power borrows from something measurable, tested and independently checked to justify a system that currently has very few of those things. Like the alcohol comparison, it points to a familiar endpoint and leaves out what makes it work.

“When defects in products are found it leads to recall or liability. With an overwhelming evidence of harm, platforms or some of their features should be labelled as defective”

The idea behind this analogy is simple and familiar. When a manufactured product, a faulty break or an exploding battery, is found to be dangerous the law can force a recall, often followed by a compensation to victims. Similarly, if a platform, or a particular feature, can be shown to cause harm, it should be treated the same way. Of all the analogies listed here, this one has travelled furthest into the legal domain.

The EU's revised Product Liability Directive redefines a "product" to include “operating systems, firmware, computer programs, applications or AI systems” [xix]. In the United States the same argument is being fought case by case in court. Earlier this year jury found Meta and YouTube found liable for negligently designing products and features that harmed children [xx] [xxi] [xxii].

The appeal of this analogy is that it usually targets specific, nameable features rather than the vague question of whether social media as a whole is "bad." Once again autoplay, infinite scroll, disappearing messages, anonymous accounts are regularly be pointed to [xxiii], and each has a plausible safer alternative. It also aligns with how products work. Since software is updated continuously, the law can impose an ongoing duty to fix or warn as evidence of harm accumulates, rather than treating the product as finished at the point of sale. Snap's awareness that disappearing messages were being used for sextortion is a classic case of a known danger shipped without a warning [xxiv].

What physical product recalls share, though, is something a digital feed lacks – a demonstrable, measurable harm tied to the product by an established causal link. A contaminant in baby formula can be isolated, an adverse effect in a drug identified, and a faulty airbag re-engineered. Causation works differently in case of social media and digital product more broadly. A feature reaches any one young person alongside everything else in their life, including home, friends, and circumstances, so establishing that it was a substantial factor in a particular child's harm depends on multiple pieces of evidence often supported by expert testimony, as recent court cases show.

Of all the analogies listed, this one may be the most compelling and strengthened by the legal precedent. That is exactly why it is worth looking closely at what the word "defect" is communicating. In its original setting the term describes a settled, technical finding that a part has failed or a standard was breached. Transferred to digital products, the word makes unsettled questions about measuring and finding direct causes of harm sound settled.

That is the wider caution running through this whole list. Each comparison borrows the certainty of the thing it names and lends it to a question that remain rife with uncertainty. What's important is next time anyone reaches for an analogy this of a reason why that particular one applies and whether it truly serves the argument or simply makes it easier for the audience.

 _________________________________________________________________________________________________

References:

[i] Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive science, 7(2), 155-170.

[ii] Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases: Biases in judgments reveal some heuristics of thinking under uncertainty. science, 185(4157), 1124-1131.

[iii] Sunstein, C. R. (1992). On analogical reasoning. Harv. L. Rev., 106, 741.

[iv] Fink, K., Dorning, C. & Kelly, M.L. (2024). U.S. surgeon general calls for tobacco-style warning labels for social media. NPR. https://www.npr.org/2024/06/17/nx-s1-5008816/u-s-surgeon-general-calls-for-tobacco-style-warning-labels-for-social-media

[v] Mathur, P. (2026). ‘Would rather have kids smoke than stay on social media’: Danish PM’s remark goes viral, sparking debate. WION. https://www.wionews.com/trending/would-rather-kids-smoking-than-stay-on-social-media-danish-pm-s-remarks-go-viral-sparking-debate-1780311508274

[vi] Stacey, K., Milmo, D., Booth, R. & Campbell, D. (2026). ‘Like tobacco’: Wes Streeting calls for partial social media ban for under-16s. The Guardian. https://www.theguardian.com/uk-news/2026/may/25/social-media-should-be-treated-like-tobacco-streeting-calls-for-under-16s-ban-on-certain-platforms

[vii] Peterson, B. (2018). Salesforce CEO Marc Benioff: Facebook should be regulated like the cigarette industry. Business Insider. https://www.businessinsider.com/salesforce-ceo-marc-benioff-said-at-davos-that-facebook-should-be-regulated-like-cigarettes-2018-1 

[viii] Dores, A. R., Peixoto, M., Fernandes, C., Marques, A., & Barbosa, F. (2025). The Effects of Social Feedback Through the “Like” Feature on Brain Activity: A Systematic Review. Healthcare, 13(1), 89. https://doi.org/10.3390/healthcare13010089 

[ix] Cerny, C. (2026). Why comparing social media and tobacco misses the point. Action on Smoking and Health (ASH). https://ash.org.uk/media-centre/news/blog/why-comparing-social-media-and-tobacco-misses-the-point

[x] Van Royen, K., Poels, K., Vandebosch, H., & Adam, P. (2017). “Thinking before posting?” Reducing cyber harassment on social networking sites through a reflective message. Computers in Human Behavior, 66, 345–352. https://doi.org/10.1016/j.chb.2016.09.040

[xi] Ji, J., Qin, X., & Calabrese, C. (2026). Fact-checking misinformation on Chinese Social Media: Impact of corrections, awareness prompts, and legal warnings on endorsement. New Media & Society, 28(5), 2217–2236. https://doi.org/10.1177/14614448251334764

[xii] Prichard, J., Scanlan, J., Krone, T., Spiranovic, C., Watters, P., & Wortley, R. (2022). Warning messages to prevent illegal sharing of sexual images: Results of a randomised controlled experiment. Trends and Issues in Crime and Criminal Justice, (647), 1–15. https://doi.org/10.52922/ti78559

[xiii] Armstrong, C. (2025). Prime minister confident social media ban will work and will empower parents. ABC. https://www.abc.net.au/news/2025-11-06/prime-minister-says-social-media-ban-will-work/105978424

[xiv] Associated Press. (2026). French lawmakers approve a sweeping social media ban for children under 15. NPR. https://www.npr.org/2026/07/22/g-s1-134875/france-social-media-ban

[xv] World Health Organization. (n.d.) Making roads safer. https://www.who.int/europe/activities/preventing-road-traffic-injuries

[xvi] Criado-Perez, C. (2021). Invisible women : data bias in a world designed for men. Abrams Press.

[xvii] Bose, D., Segui-Gomez, ScD, M., & Crandall, J. R. (2011). Vulnerability of female drivers involved in motor vehicle crashes: an analysis of US population at risk. American journal of public health, 101(12), 2368-2373. 

[xviii] Age Assurance Technology Trial. (2025). PART D. Age Estimation. https://ageassurance.com.au/wp-content/uploads/2025/08/AATT_Part_D_DIGITAL.pdf

[xix] European-Parliament, and Council-of-the-European-Union. (2024). Directive (EU) 2024/2853 of the European Parliament and of the Council of 23 October 2024 on Liability for Defective Products and Repealing Council Directive 85/374/EEC. Off. J. Eur. Union. https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=OJ:L_202402853

[xx] Hays, K., Saad, N. &. Morris, R. (2026). Campaigners welcome Meta and YouTube's defeat in landmark social media addiction trial. BBC. https://www.bbc.com/news/articles/c747x7gz249o

[xxi] Hays, K. & McMAhon, L. (2026). Meta fined $567m in largest child safety ruling against social media giant. BCC. https://www.bbc.com/news/articles/cd7lz3wr2rlo

[xxii] Hays, K. (2026). Meta told to pay $375m for misleading users over child safety. BBC. https://www.bbc.com/news/articles/cql75dn07n2o

[xxiii] 5Rights Foundation. (2021). Pathways: How digital design puts children at risk. https://5rightsfoundation.com/wp-content/uploads/2021/09/Pathways-how-digital-design-puts-children-at-risk.pdf

[xxiv] Horwitz, J. (2024). Snap Failed to Warn Users About Sextortion Risks, State Lawsuit Alleges. The Wall Street Journal. https://www.wsj.com/tech/snap-failed-to-warn-users-about-sextortion-risks-state-lawsuit-alleges-0b170fc7