In May 2026, Beijing Daily published a systematic exposé of the flood of AI-generated "buyer shows" — buyer-posted photos meant to show products in real use — across e-commerce platforms. This is not an isolated case of technological abuse, but a microcosm of how AI is striking at the trust ecosystem of consumption: when a convincingly lifelike "buyer photo" can be mass-produced in a matter of seconds, the very foundation of the entire e-commerce review system is being quietly eroded.
From "Seeing Is Believing" to "Seeing No Longer Proves"
The review ecosystem consumers have long depended on — buyer-show photos serving as a trust anchor somewhere between official marketing and consumer decision-making — is being wrecked by AI-generated fake imagery. This is more than a problem of technical fraud; it is the collapse of a piece of trust infrastructure. In the past, what consumers had to sort out was "order brushing" — padded sales and planted reviews, where at least the review text was written by a real person and the loyalty points came from real transactions. Now, the very figure of the buyer itself is fabricated.
On some e-commerce platforms, large numbers of AI-generated buyer-show photos have appeared. AI image generation has grown ever more convincing, and there are tutorial posts and dedicated tools besides: enter a product image and a few selling points, and the system can generate model photos with a single click. Without clear labeling, ordinary consumers can easily fall into the trap.
On the technical side, the evolution of current AI image-generation models (such as GPT Image 2 and HappyHorse) has made producing a convincingly lifelike buyer-show photo virtually barrier-free. A user need only upload a product image and type in a few selling-point keywords; within seconds, multiple "model display photos" come out. In lighting, skin tone, and scene consistency, these images have reached a level that the naked eye can barely distinguish from real photographs.
The Platform Logic of Bad Money Driving Out Good
The reporting reveals a deeper structural problem: the platforms themselves play an ambiguous role in this fraud.
The sheer volume of transactions already makes policing difficult; and going a step further, for merchants this low-barrier marketing pays off enormously — while for the platform, isn't a screen full of beautiful images itself a kind of "storefront dressing" and "traffic formula"? By comparison, merchants who trade honestly and use "raw, unretouched" product displays end up looking rather less impressive. Bad money drives out good — and who can stay outside of that?
The key insight of this passage is that it identifies the platform's dual identity: it is simultaneously the rule-maker and rule-enforcer, and a beneficiary of the traffic. Polished AI-generated buyer-show photos raise the overall display quality of the platform and deliver better conversion rates — at least in the short term. This interest structure leaves platforms inherently short of motivation to police the practice.
From the perspective of institutional economics, a platform's willingness to regulate follows a simple cost–benefit model:
- Short-term gains: AI buyer-show photos improve display quality → higher click-to-conversion rates → more GMV and commission revenue
- Short-term costs: investment in detection technology and human review teams → higher operating expenses
- Long-term risks: collapse of consumer trust → failure of the review system → depreciation of platform credibility
When short-term gains are concrete and long-term risks highly uncertain, platforms naturally lean toward inaction or minimal action — and this constitutes the first structural dilemma of trust governance in the AI era.
The Multi-Front Spread of AI Fabrication
The reporting further notes that AI fabrication is no longer confined to buyer-show photos; it has spread across multiple links in the e-commerce chain:
- Livestream sellers: AI-generated clip videos, in which fully algorithm-synthesized sales talk replaces live human hosts
- Electronic customer service: automated programs talking to themselves, with consumers steered by disguised "human support" into fraudulent pages
- Store displays: mass-produced "photo-deceptions," with systematic gaps between virtual scenes and the physical goods
This means that from traffic acquisition, to conversion, to after-sales service, the entire e-commerce process can be blanketed by AI-fabricated content. Consumers are trapped in a dilemma of "nothing the eye sees can be trusted." When every link in the chain can be faked with AI, consumers' trust expectations slide from "partial doubt" to "wholesale skepticism" — and the value of trust itself, as the lubricant of transactions, is systematically dissipated.
The Regulatory Dilemma and Platform Responsibility
Relevant regulations explicitly prohibit the practice, and most e-commerce platforms require in their own rules that review content and images posted by buyers must genuinely reflect the purchased product, forbidding the fabrication of user reviews. Yet the sheer volume of transactions inherently raises the difficulty of enforcement.
The crux of regulation is this: the rules exist, but enforcement is costly. Detection technology for AI-generated images is admittedly advancing, but its race against generation technology is essentially a cat-and-mouse game — once detection models learn to identify certain signatures of the current generation of AI, generation models fix those signatures in their next iteration. This cycle dooms the technical side to a passive position.
The remedy proposed in the reporting is to "strengthen governance at the source, raise the cost of violations, and make the price of fraud exceed that of a bad review" — but from the standpoint of institutional economics, the incentive to commit fraud is only suppressed when the probability of being caught and punished is high enough and the penalties heavy enough. At present, neither condition is met.
One overlooked dimension is the absence of a functioning labeling regime for AI-generated content. Although China's cyberspace administration already required in 2025 that AI-generated content be clearly labeled (see "The Triple Gates of AI Regulation"), in e-commerce scenarios like buyer-show photos, the labeling obligation effectively falls on merchants and the users of generation tools — and fraudsters will naturally not label their output voluntarily. That means the effectiveness of the labeling regime depends heavily on front-end detection and after-the-fact accountability, both of which are currently weak.
An Analytical Framework: The Five Stages of Trust Inflation
From this episode, one can extract an analytical framework for "trust inflation" in the AI era — much like monetary inflation, the devaluation of trust is not a one-off event but a structural process that transmits stage by stage:
- The trust-accumulation stage: platforms build review systems, and consumers gradually develop trust in buyer-show photos — a virtuous cycle of review → exposure → purchase
- The technological inflection point: AI generation reaches a convincingly lifelike level and the barrier to fraud collapses — from "order brushing that requires heavy investment" to "one-click generation"
- The platform-ambivalence stage: platforms become aware of the fraud but handle it cautiously because of short-term commercial interests — enforcement stays at the level of "rules exist, but execution is limited"
- The trust-collapse stage: consumers realize reviews cannot be trusted and begin doubting all reviews — information costs rise and transaction efficiency falls
- The institutional-reconstruction stage: new trust mechanisms must be introduced (AI detection, provenance labeling, third-party certification) — a shift from "trusting user-generated content" to "trusting verified content"
China's e-commerce ecosystem is currently in the transition between stages 3 and 4. Some consumers have already noticed the flood of fake buyer-show photos, but a generalized crisis of trust has not yet formed. This window is precisely the critical moment for institutional reconstruction — once full trust collapse at stage 4 sets in, the cost of repair will far exceed the cost of preventive investment today.
AI-generated buyer-show photos are not mere marketing fraud; they are the structural erosion of e-commerce's trust infrastructure. When consumers cannot distinguish "real human reviews" from "AI reviews," the entire review system loses its informational value. Repairing this trust requires more than technical detection tools — it demands a deep adjustment of the platforms' incentive structure, from "pursuing attractive content displays" to "maintaining a trustworthy information ecosystem."
Structural Implications: Beyond E-Commerce
Buyer-show fraud is only the tip of the iceberg of the AI trust crisis. The same logic is spreading into other trust-driven domains:
- Medical consultation platforms: AI-generated "patient testimonials" are diluting the credibility of online consultations
- Education and training: AI-simulated "learning-outcome showcases" strip course reviews of their reference value
- Travel and accommodation: AI-generated "real hotel photos" have become a latent risk on OTA platforms
- Social platforms: AI-generated "ordinary-user recommendations" are dissolving the trust base of recommendation-driven communities
This framework reveals a general law: wherever economic activity creates incentives to fabricate, AI will accelerate the corrosion of its trust infrastructure. This is not the governance problem of any single platform, but a structural risk of the digital economy — trust, the core mechanism for minimizing transaction costs, is being deconstructed from within by technology.
Over the next five to ten years, as AI generation continues to evolve, "authenticity verification" itself will become an independent industry — the combination of third-party certification, blockchain provenance, and biometric binding may eventually rebuild the trust foundation that AI has eroded. But until those new mechanisms mature, consumers will have to live with a consumption environment of default suspicion.