The Counterfeit Web: How Synthetic Content Broke Online Trust

May 27, 2026

The first sign that something was wrong, for a Manchester woman who had spent a fortnight choosing a residential care home for her father, was that all the reviews sounded the same. Not similar in sentiment, identical in cadence. Five-star write-ups praising “compassionate, attentive staff” and “a warm, family atmosphere” appeared on three different aggregator sites for three different homes, separated only by a swap of proper nouns. When she telephoned the regional CQC inspector and asked plainly whether any of these reviews could be believed, the inspector did not hesitate. “Increasingly,” she was told, “we tell families to come and see for themselves. The websites are not what they were.”

That sentence, mundane on its surface, contains the whole problem. The websites are not what they were. The shared informational substrate that ordinary British and European life has come to depend on, the latticework of star ratings, customer write-ups, expert round-ups, patient testimonials and tradesperson endorsements, has been quietly replaced by something else. Something that looks the same and reads the same, but is not the same. Something that, in many cases, was never written by anyone at all.

In April 2026, the technology publication Silicon Canals ran an analysis arguing, citing aggregated industry estimates, that as much as ninety per cent of online content will be AI-generated by the end of the year, and that the detection tools commercial platforms rely on already fail to identify synthetic material more than half the time. The piece, “The AI content flood isn't just an information problem, it's a trust problem,” landed the same week Yelp published its 2025 Trust and Safety Report, disclosing it had filtered out almost five hundred thousand suspected AI-generated reviews and shut more than 1.3 million user accounts in twelve months, a 138 per cent year-on-year jump. Weeks earlier, in Frontiers in Psychology, a team led by Zhixuan Gong of Hunan University had shown that when readers saw AI disclosure labels they did not become more discerning. They became more avoidant. The labels triggered cognitive dissonance and readers simply looked away.

Read together, those three documents describe an inflection point: the moment at which the everyday infrastructure of trust began to give way under its own weight. The synthetic web is here. The question is what can be done, and who carries the responsibility for the damage already done.

The scenes the statistics describe

Statistics flatten things. Five hundred thousand fake reviews sounds like a number on a slide. To see what it means, walk around any British high street.

The plumber who took two days to come to a flooded kitchen in Walthamstow last winter had a profile on a national tradesperson aggregator with 167 reviews, an average rating of 4.9, and a write-up praising his “calm, methodical approach to even the most chaotic emergencies.” After the kitchen flooded again forty-eight hours later, the homeowner read the reviews carefully and noticed thirty-one used the phrase “calm, methodical approach.” Twelve mentioned “chaos.” Eight used “lifesaver” in the closing line. The Competition and Markets Authority, which gained new enforcement powers in April 2025 under the Digital Markets, Competition and Consumers Act 2024, would describe what she found as a textbook unfair commercial practice. The homeowner described it as “a casino dressed up as a directory.”

The same pattern, with different stakes, attaches to medical products. A 2025 sweep by Greater London trading standards officers working with the Chartered Trading Standards Institute found dozens of listings for over-the-counter sleep aids and joint supplements supported almost entirely by reviews bearing the structural hallmarks of large language model output: balanced clauses, even pacing, an absence of the random small grievances real customers cannot help producing. One product, a magnesium spray marketed at people with restless leg syndrome, had two hundred and forty reviews, of which the trading standards team estimated, conservatively, that fewer than thirty had been written by humans.

And then there are the care homes. The CQC, which inspects social care providers in England, has for years quietly cautioned families against weighting online reviews heavily. Internally, inspectors will tell you the gap between a home's online reputation and what they find on a visit has grown wider every year since 2023. By 2026, according to one senior inspector who would speak only on condition that her employer not be named, “the correlation has effectively broken.” A home rated 4.8 stars on a popular aggregator can be in special measures. A home with a thin online presence and three reviews can be exemplary. The signal has decoupled from the substance.

This is what a counterfeit web looks like in practice: not the obvious deepfake of a politician, but the slow replacement of the small textual artefacts that hold ordinary commerce together.

The numbers, sourced

The Silicon Canals figure of ninety per cent is, the publication concedes, an aggregated industry estimate rather than a peer-reviewed result. Some consider it an overstatement; others think the proportion of AI-touched content (as distinct from purely AI-generated) is already higher. What is not in dispute is the direction of travel. Studies cited by the publication, including work from the University of Mainz, found participants rated AI-generated and human-generated text as similarly credible, and in some conditions perceived AI prose as clearer and more engaging. The arms race between generators and detectors is being won by the generators.

The detection failure rate is more empirically tractable. Independent benchmarking through 2025 and 2026 has consistently shown no widely deployed detector exceeds roughly eighty-five per cent accuracy across generation models, that even leading detectors miss fifteen to thirty per cent of synthetic content, and that false-positive rates for human prose run three to twelve per cent, with non-native English speakers and technical writers disproportionately misclassified. For a platform processing hundreds of millions of reviews, the maths is grim. Every percentage point of false negative is hundreds of thousands of synthetic items waved through. Every percentage point of false positive is real customers, often the ones with the least linguistic privilege, accused of fakery.

Yelp's report is the cleanest empirical window onto what this looks like at scale. Its trust and safety team filtered nearly half a million suspected AI-generated reviews in 2025, removed over 193,700 reviews flagged by the community (a quarter of which lacked any firsthand experience), and closed roughly 1.3 million accounts for terms-of-service violations, including 889,800 tied to fake airline customer-support scams. The platform reported a 49 per cent rise in accounts linked to “review exchange rings,” a 29 per cent rise in lead-generator business pages, and an 80,000-strong wave of removals tied to viral review brigading. Yelp filed over 1,020 cross-platform reports to Instagram, Facebook, X, LinkedIn, Reddit, TikTok and Craigslist; sixty per cent resulted in third-party action, a 62 per cent improvement on 2024.

The numbers tell a coherent story. The platforms are working harder; the volume is rising faster; the surface beneath everyone's feet is moving.

Why disclosure labels are making it worse

The intuitive policy response to a synthetic-content crisis is to label the synthetic content. The European Commission's Code of Practice on marking and labelling AI-generated content, first drafted on 17 December 2025 and expected to be finalised in May or June 2026 ahead of Article 50 of the EU AI Act coming into force in August, takes precisely this approach. It proposes a common visual marker (a two-letter “AI” icon) alongside machine-readable metadata, allowing users to identify, at a glance, whether content has been generated or substantially manipulated by artificial intelligence.

The trouble, suggested by the Frontiers in Psychology study by Gong, Peng, Cui and Lv, is that disclosure does not behave the way policymakers think. Across two experiments with 760 participants on simulated Bilibili and TikTok-style interfaces, the researchers tested three conditions: clear AI labels (e.g. “content generated by AI”), ambiguous labels (e.g. “suspected AI, please verify”), and no label. The headline finding was uncomfortable. Ambiguous labels significantly increased information avoidance compared to clear labels or no labels, with a Cohen's d effect size of 0.57 versus the no-label condition in the first study, replicated at d = 0.88 in the second. The mediating mechanism was cognitive dissonance: the conflicting signal of “we don't know if this is real” produced enough psychological discomfort that readers disengaged rather than evaluated. They did not weigh the content more carefully. They closed the tab.

The implication is structural. Where a platform cannot distinguish synthetic from authentic with confidence, and so relies on probabilistic, hedging warnings, the labels do not restore trust; they corrode it further. Readers learn quickly that the label is a tax on attention without an information dividend, and stop paying it. The authors propose moving from probabilistic warnings to high-threshold binary classification, leaning on provenance-based authentication rather than detection-based labelling. That maps onto an emerging architecture the standards community has been quietly building for years.

The technical layer: provenance over detection

The Coalition for Content Provenance and Authenticity, known as C2PA, is one of the more interesting institutions to have grown up around the synthetic-content problem. Founded in 2021 as an alliance between Adobe, Arm, Intel, Microsoft and Truepic, hosted under the Linux Foundation's Joint Development Foundation, and now claiming, as of January 2026, more than six thousand member organisations including Google, Meta, OpenAI, Sony, Nikon and Leica, C2PA's premise is that detection is the wrong end of the stick. Instead of asking “is this image AI-generated?” after the fact, the standard asks “what is the cryptographically signed history of this content from the moment of capture or creation?”. Cameras, editing tools and AI generators that implement Content Credentials embed signed metadata describing the origin and edit history of a file. A viewer can inspect the chain of custody.

It is, in principle, the right architecture. Provenance scales where detection cannot, because it does not have to outrun the generators; it sidesteps the race entirely. In practice, however, C2PA has run into uncomfortable empirical realities. As the World Privacy Forum's 2024 technical review noted, very little internet content currently carries C2PA credentials. Worse, the credentials usually do not survive social media sharing, because the major platforms recompress and reformat uploaded images in ways that strip the metadata. An ecosystem-wide rollout still depends on coordinated decisions by platforms, generators, capture-device manufacturers and browser vendors, none of whom have a strong commercial incentive to move first.

The EU AI Act may force the issue. From August 2026, providers of AI systems must ensure machine-readable marking and detectability of AI-generated or manipulated content; deployers must disclose when AI is used to create realistic synthetic media. The draft Code of Practice for transparency leans heavily on the C2PA framework as the de facto reference architecture. Whether the Code, when finalised, manages to push provenance onto the platforms in a form that survives recompression, is, as one Brussels-based standards engineer put it, “the whole game.”

Watermarking, the closely related technique of statistically marking AI-generated outputs at the moment of generation, is making slower progress. OpenAI, Google and Meta have published research on text watermarking, but academic work has consistently shown that watermarks can be removed by light paraphrasing, that they degrade rapidly under translation, and that detection requires access to the model's likelihood functions. None of the major chatbot providers has yet made watermarking the default for free-tier text output. The asymmetry is brutal. A determined adversary needs five seconds of paraphrasing to defeat a watermark; a reviewer who wants to verify it needs cooperation from the model provider and a working detector.

The regulatory layer: three jurisdictions, three theories

Geography matters in this fight, because different jurisdictions have arrived at different theories about what the synthetic-content problem actually is.

In the European Union, the prevailing theory is that the problem is a transparency failure. The EU AI Act, finalised in 2024 with provisions phasing in across 2025 and 2026, treats AI-generated content principally as something that must be labelled and made detectable. Article 50 imposes transparency obligations on both providers (the model makers) and deployers (the platforms and users who run the models in production). Deepfakes must be disclosed unless used for law enforcement or evidently artistic purposes; published AI-generated text on matters of public interest must be flagged; machine-readable provenance must be embedded by providers. Penalties scale, as elsewhere in EU tech regulation, with global revenue. The architecture is the GDPR and Digital Services Act paradigm applied to the substance of content, with the European Commission, working through the AI Office, as the central rule-maker.

In the United Kingdom, the theory is more piecemeal but, in places, more aggressive on consumer-facing harms. The Digital Markets, Competition and Consumers Act 2024 came into force in stages from April 2025, and Schedule 20, the fake-reviews provisions, is the most immediately relevant. The Act bans the commissioning and publishing of fake consumer reviews, defines “fake” expansively to include reviews that purport to be but are not based on a person's genuine experience (which captures AI-generated reviews even where the underlying business does not realise it has commissioned them), and requires platforms to take “reasonable and proportionate” steps to verify authenticity. The Competition and Markets Authority, which acquired direct enforcement powers including the ability to impose fines of up to ten per cent of global turnover, published its CMA208 fake-reviews guidance in 2025 and began enforcement action against several large aggregators that year. Separately, Ofcom, working under the Online Safety Act 2023 and a February 2026 government clarification that closes the loophole around large language model chatbots, can fine platforms the higher of £18 million or ten per cent of global turnover for failure to address illegal content, including AI-generated illegal content carried on user-to-user services.

In the United States, the theory is that the problem is fraud, and the response is consumer-protection enforcement under existing statutes. The FTC's Final Rule on Fake Reviews and Testimonials, finalised in August 2024 and in force from October, prohibits creating, buying or distributing fake or AI-generated reviews, carrying civil penalties of up to $53,088 per violation. On 22 December 2025 the FTC sent warning letters to ten unidentified companies, its first enforcement step. The American architecture is less centralised than the EU model, more reactive, more dependent on case-by-case enforcement, and for now more limited in its leverage over generative AI providers as opposed to the businesses deploying their outputs.

The three regimes share a problem. None was designed for a world in which the cost of generating a plausible review is approaching zero and the cost of verifying it remains, by the testimony of the platforms themselves, stubbornly high.

Brandolini's law, scaled

In 2013, the Italian software developer Alberto Brandolini coined Brandolini's law, the Bullshit Asymmetry Principle: the energy required to refute bullshit is an order of magnitude greater than the energy to produce it. He coined it watching a televised political interview; it has since been applied to anti-vaccination campaigns and cryptocurrency promotion alike. The synthetic-content economy is Brandolini's law expressed in code.

Generating a thousand-word, plausible, contextually appropriate restaurant review with current tooling costs less than half a penny in compute and takes under a second. Verifying that review (by contacting the named diner, cross-referencing the booking system, checking the device fingerprint, examining the IP path, comparing stylometrically against the same author's prior reviews, and adjudicating the result) can take a trust and safety team several minutes of human attention plus several pence of automated compute per item. The asymmetry is not five-to-one or ten-to-one. It is, on the platforms' own internal numbers, several orders of magnitude. Yelp's filtering of half a million suspected AI reviews in 2025 was the visible top of an unknown but likely much larger underwater mass.

The political theorists Bobby Chesney and Danielle Citron, in the California Law Review in 2019, anticipated a related dynamic they called the “liar's dividend.” As the public becomes aware that audio, video and text can be convincingly fabricated, the dividend accrues to liars, who can dismiss authentic embarrassments as deepfakes. Pre-registered experiments with more than fifteen thousand American respondents in the American Political Science Review found the dividend operating reliably against text-based reporting, though largely ineffective against video. The synthetic-content economy generalises this. It is not only liars who benefit, but anyone whose interests are served by the listener being unable to tell the difference, a much larger population.

The wisdom of crowds, considered as a casualty

The aggregator economy was built on a 2004 idea, popularised by James Surowiecki, that under the right conditions large numbers of independent, diverse opinions converge on accurate judgements. The wisdom of crowds is the foundational logic of every star rating you have read. It has always been imperfect: selection bias is significant, manipulation has always been possible. But the basic premise, that a large enough sample of independent human experience reveals something real, has anchored consumer behaviour for two decades.

Synthetic content breaks that premise at the root. The crowd is no longer independent, because one actor can generate a thousand voices. It is no longer diverse, because the underlying language model has its own statistical fingerprint and draws from a narrower distribution than human writers. And it is no longer made of human experience, because the experience never happened. What looks like a wisdom-of-crowds signal is the output of a very small number of decisions amplified to look like consensus.

This is the substantive sense in which the synthetic web is not a degraded version of the old web but a categorically different thing. It does not produce noisier signals. It produces non-signals dressed in the visual grammar of signals. Onora O'Neill, the British philosopher who delivered the BBC Reith Lectures on trust in 2002 and whose work has shaped how regulators and ethicists think about institutional confidence, has long argued that trust requires what she calls “intelligent accountability”: the capacity, in principle, to interrogate the source of a claim, examine the reasoning, and verify the chain. Synthetic content is engineered to resist that capacity. It looks accountable while being structurally unaccountable.

Sissela Bok, the Swedish-American philosopher whose 1978 book Lying and subsequent work at Harvard's Kennedy School made her one of the most cited scholars on the ethics of deception, makes a similar point about the social cost of routine, low-stakes lying. Each individual lie may do limited damage. The cumulative effect of large numbers of lies, normalised, is to deplete the public stock of trust on which all communication depends. The synthetic-content economy is the industrial-scale version of that depletion.

Whose responsibility is it, really

The accountability question is the one regulators, platforms and ethicists are circling. Four candidate answers compete.

The platforms argue they are running moderation at scales no previous information regime has managed, that volume is rising faster than headcount can be added, and that they are investing in detection, cross-platform coordination and rule changes. Yelp's report, in this reading, is not an admission of failure but an account of the work needed to keep the system from collapsing. The platforms are the people inspecting the bridge. They did not build the river.

The model providers argue that they have built guardrails, watermarking research, terms-of-service prohibitions on reputation manipulation and provenance metadata at the point of generation, and that misuse by determined bad actors is at most a partial responsibility, comparable to a typewriter manufacturer's liability for a forged document. They point to the EU AI Act's provider-side obligations as the appropriate institutional response. They made the typewriter. The crime is not theirs.

The regulators argue that statutes have been passed, rules have been finalised, and the task now is to enforce them. The CMA, the FTC, Ofcom and the European Commission have all taken concrete enforcement steps in the past eighteen months. The pace of enforcement is the pace of due process, and due process is slow.

And the users, finally, are told they bear some residual responsibility, on the rationale that any consumer should know better than to trust online reviews unconditionally. This is, by some distance, the weakest of the four arguments. It privatises a cost imposed on people without their consent. The Manchester woman did not cause the synthetic-content economy to exist. She inherited it.

The honest accounting is that all four parties carry responsibility, but unequally. The model providers have prioritised capability over containment, releasing systems whose ability to generate plausible review-style prose vastly outstrips verification infrastructure. The platforms have until recently treated trust and safety as a cost centre. Regulators in all three jurisdictions moved slowly through the 2020s and only since 2024 begun to apply rules with serious teeth. Users have done what users do: use the tools they have been given.

What restoration could look like

A realistic programme for restoring trustworthy informational infrastructure would draw on at least four threads, none of which alone is sufficient.

First, provenance would have to win. Not as a niche feature for professional photographers and journalists but as a baseline expectation, embedded in capture devices, generation tools and platform pipelines, surviving recompression, visible by default. The C2PA standard exists; the EU AI Act may force its adoption in the European market; the open question is whether the United States and the United Kingdom follow, and whether the major social-media platforms can be persuaded or compelled to preserve credentials through their image and video processing pipelines. This is a multi-year project at a minimum.

Second, the reputation economy would have to develop alternatives to pure aggregator-driven review systems. Several promising approaches exist in narrow domains: verified-purchase reviews tied to receipts; closed networks within professional bodies (the Royal Institute of British Architects, the Federation of Master Builders, the CQC's own ratings) that carry the weight of an institution behind each rating; personal-network recommendation systems within messaging platforms, where the trust is borrowed from existing relationships rather than synthesised from strangers. None of these scales as cheaply or as universally as the aggregator model. All are more resistant to synthetic capture, because they tie reputation to something other than easily generated text.

Third, regulation would have to focus less on labelling and more on liability. The Frontiers in Psychology finding suggests disclosure regimes alone are insufficient; the experimental evidence is that ambiguous labels make readers disengage rather than evaluate. A more durable approach, hinted at in the EU's provider-and-deployer architecture and in the FTC's finalised rule, is to assign clear legal liability to the platforms that publish synthetic reviews and to the businesses that benefit from them, regardless of who pressed the generate button. The CMA's powers under the DMCCA, in particular the ability to fine ten per cent of global turnover, are the kind of incentive that can change platform behaviour quickly when applied. The question is whether enforcement will keep pace with generation.

Fourth, social mechanisms would have to be rebuilt at a level beneath the platforms. Local newspapers that survive (and several British regional titles, against the run of play, are growing again under philanthropic and cooperative ownership) carry weight precisely because they are accountable to a defined audience. Community Facebook groups, message-board collectives, neighbourhood WhatsApp networks, and the older mechanism of word of mouth are all forms of trust harder to counterfeit at scale, because they tie reputation to identifiable people in identifiable places. The synthetic-content economy is, in some senses, encouraging a return to these older forms by destroying the credibility of the algorithmic layer above them.

What the inspector said next

The CQC inspector who told the Manchester woman to come and see for herself was disclaiming the digital signal and substituting an institutional one. Implicit in her advice was the older, slower architecture of trust: an inspectorate, accountable to a statutory regulator, whose ratings are produced by named human beings who have walked through the building and spoken to the residents. That signal is expensive to generate (a single CQC inspection takes days of trained-inspector time and is published with a named lead and a methodology); for that reason, it is also very expensive to fake. Authenticity is costly to produce and cheap to verify, because the inspectorate tells you who did the work.

The synthetic-content economy has inverted that asymmetry across most of the consumer web. Restoration, if it is possible, requires inverting it back: making authentic content cheap to verify, by way of cryptographic provenance, and making synthetic content expensive to deploy, by way of liability. Neither half of that programme is technically impossible. Both are politically and commercially difficult, because they impose costs on actors who have, until now, externalised them.

The Manchester woman, in the end, picked her father's care home by visiting four of them in person, talking to staff, talking to residents, reading the most recent CQC report cover to cover, and ignoring the aggregator scores entirely. She found a home with an unfashionable website, a dog that lived on the premises, and a manager who returned phone calls. Her father has been there for six months. She does not know how she would have made the decision if her father had been further away or her time more constrained, and she does not pretend that what worked for her would scale to a country.

That is the awkward truth at the centre of this story. The everyday trust that ordinary consumer decisions depend on was always a public good, sustained by institutions and norms and a basic shared assumption that other people existed. The synthetic-content economy has begun to erode each of those pillars at once. The mechanisms that could restore them, technical, regulatory and social, exist but are partial, contested and slow. The damage in the meantime is being borne by the people least equipped to verify what they are reading: the elderly choosing care, the renters choosing landlords, the patients choosing treatments, the small businesses being review-bombed by competitors with access to a chatbot.

Whether the next decade looks more like a restoration or more like a managed decline depends on whether the institutions still capable of generating trustworthy signals (the regulators, the inspectorates, the standards bodies, the surviving local press, the professional registries) can be given the resources, the legal teeth and the cultural authority to fill the gap left by the failing aggregators. It also depends on whether the platforms and model providers can be made to internalise costs they have, until very recently, been allowed to externalise. There is nothing inevitable about either outcome. The one thing that is certain is that the websites are not what they were, and pretending otherwise is no longer a tenable position.

The counterfeit web is here. The question is what we build alongside it.

References & Sources

Brennan, J. (2026). “The AI content flood isn't just an information problem, it's a trust problem.” Silicon Canals, April 2026. https://siliconcanals.com/m-the-ai-content-flood-isnt-just-an-information-problem-its-a-trust-problem/
Yelp. (2026). 2025 Trust & Safety Report. Yelp Official Blog, 25 February 2026. https://blog.yelp.com/news/2025-trust-and-safety-report/
Yelp Inc. (2026). “Yelp Releases 2025 Trust & Safety Report.” Yelp Investor Relations, 25 February 2026. https://www.yelp-ir.com/news/press-releases/news-release-details/2026/Yelp-Releases-2025-Trust--Safety-Report/default.aspx
Gong, Z., Peng, D., Cui, J., and Lv, Z. (2026). “The paradox of AI content labeling: how clarity influences information avoidance via cognitive dissonance on social platforms.” Frontiers in Psychology, Vol. 17, published 10 March 2026. DOI: 10.3389/fpsyg.2026.1751670. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2026.1751670/abstract
Coalition for Content Provenance and Authenticity (C2PA). https://c2pa.org/
Content Authenticity Initiative. “How it works.” https://contentauthenticity.org/how-it-works
World Privacy Forum. (2024). “Privacy, Identity and Trust in C2PA.” https://worldprivacyforum.org/posts/privacy-identity-and-trust-in-c2pa/
European Commission. “Code of Practice on marking and labelling of AI-generated content.” https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content
EU Artificial Intelligence Act. Article 50: Transparency Obligations. https://artificialintelligenceact.eu/article/50/
Jones Day. (2026). “European Commission Publishes Draft Code of Practice on AI Labelling and Transparency.” January 2026. https://www.jonesday.com/en/insights/2026/01/european-commission-publishes-draft-code-of-practice-on-ai-labelling-and-transparency
Herbert Smith Freehills Kramer. (2026). “Transparency obligations for AI-generated content under the EU AI Act.” https://www.hsfkramer.com/notes/ip/2026-03/transparency-obligations-for-ai-generated-content-under-the-eu-ai-act-from-principle-to-practice
UK Government. Digital Markets, Competition and Consumers Act 2024, Schedule 20 (fake reviews). https://www.legislation.gov.uk/ukpga/2024/13/schedule/20
Competition and Markets Authority. (2025). “Fake reviews guidance” (CMA208). https://assets.publishing.service.gov.uk/media/67eeb64fe9c76fa33048c790/CMA208_-_Fake_reviews_guidance.pdf
Ofcom. “Ofcom's strategic approach to AI.” https://www.ofcom.org.uk/about-ofcom/annual-reports-and-plans/ofcoms-strategic-approach-to-ai
Ofcom. “AI chatbots and online regulation.” https://www.ofcom.org.uk/online-safety/illegal-and-harmful-content/ai-chatbots-and-online-regulation-what-you-need-to-know
Federal Trade Commission. (2024). “Final Rule Banning Fake Reviews and Testimonials.” Press release, 14 August 2024. https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials
DLA Piper. (2025). “FTC releases warning letters for fake consumer reviews and AI.” December 2025. https://www.dlapiper.com/en-us/insights/publications/2025/12/ftc-warning-letters-ai-consumer-reviews
Chesney, R. and Citron, D. (2019). “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security.” California Law Review, 107. https://www.californialawreview.org/print/deep-fakes-a-looming-challenge-for-privacy-democracy-and-national-security/
Schiff, K. and Schiff, D. (2024). “The Liar's Dividend: Can Politicians Claim Misinformation to Evade Accountability?” American Political Science Review. https://www.cambridge.org/core/journals/american-political-science-review/article/liars-dividend-can-politicians-claim-misinformation-to-evade-accountability/687FEE54DBD7ED0C96D72B26606AA073
Brandolini, A. (2013). The Bullshit Asymmetry Principle. https://en.wikipedia.org/wiki/Brandolini%27s_law
O'Neill, O. (2002). A Question of Trust: The BBC Reith Lectures 2002. Cambridge University Press. https://philpapers.org/rec/ONEAQO
Bok, S. (1978). Lying: Moral Choice in Public and Private Life. Pantheon Books.
Hill Dickinson. (2025). “Digital Markets, Competition and Consumers Act 2024: new consumer law protections now in force.” https://www.hilldickinson.com/our-view/articles/digital-markets-competition-and-consumers-act-2024-new-consumer-law-protections-now-in-force/
Burges Salmon. (2026). “Ofcom and the Online Safety Act in 2026.” https://www.burges-salmon.com/articles/102mi1g/ofcom-and-the-online-safety-act-in-2026/
UCLA HumTech. “The Imperfection of AI Detection Tools.” https://humtech.ucla.edu/technology/the-imperfection-of-ai-detection-tools/

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Discuss...