ChatGPT Shopping, Google AI Mode, Perplexity Shopping: these engines no longer present a list of links. They synthesize, compare, and directly recommend a product. The user asks a question — "what is the best stand mixer for under 300 euros?" — and gets an answer built by AI from multiple sources. According to a Visa study (3,700 respondents), 73% of consumers are ready to delegate product discovery to an AI assistant. On the infrastructure side, two agentic commerce protocols are already being deployed: OpenAI/Stripe's Agentic Commerce Protocol (ACP) and Google's Universal Commerce Protocol (UCP) with Shopify and Walmart.
We are no longer in the SEO era. We are entering the era of GEO — Generative Engine Optimization. And with it, a trust crisis that few e-commerce players have anticipated.
AI Cannot Tell Real from Fake
The paradox is simple. Producing a convincing fake product description costs virtually nothing today. Fake reviews generated by LLMs are indistinguishable from real ones — human judges perform no better than chance at detecting them (50.8% accuracy). And LLMs themselves suffer from a veracity bias: they tend to accept any well-formulated content as reliable.
Anyone can produce credible product content at scale, and generative engines lack the tools to sort it out. This is the lemon market problem described by economist Akerlof: without reliable quality signals, markets deteriorate.
The Trust Signal Gap
At Keyban, we conducted systematic research by synthesizing 38 studies covering GEO, platform economics, and decentralized identity. The central finding is what we call the trust signal gap — a structural divide between what AI rewards and what resists manipulation.
The signals AI rewards most are the most vulnerable. Citations, numerical data, and expert references generate up to +115% visibility in generative engines. But fake statistics or invented citations can be manufactured at near-zero cost.
The most credible signals are invisible to AI. Traditional certifications (ISO, organic labels) exist as logos — a format that an AI agent can neither read nor verify. No study has demonstrated that they influence selection by a generative engine.
Platform badges work but remain captive. An "Overall Pick" badge on Amazon increases selection by AI agents from 20 to 43%. But it has no value for ChatGPT Shopping or Google AI Mode. A real advantage, but entirely locked in.
Added to this is a concentration effect: ChatGPT draws more than 92% of its citations in consumer electronics from third-party editorial content, and high-traffic sites receive roughly three times more citations. For challenger brands and SMEs, the strategic question becomes: how do you prove product quality to an AI agent without depending on media visibility?
Verifiable Cryptographic Attestations as the Answer
This question led us to examine Verifiable Credentials (VCs), which became a W3C recommendation in May 2025.
The principle is intuitive. A laboratory certifies that your product meets a standard. Today, this certification exists as a PDF or logo — a dishonest competitor can copy it in seconds. With a Verifiable Credential, the same certification becomes a cryptographically signed attestation, linked to a unique digital identifier of the laboratory, verifiable by anyone without having to contact them.
VCs combine three properties that no other signal offers simultaneously: native machine readability (JSON-LD), cross-platform portability, and resistance to falsification through asymmetric cryptography.
Let's be transparent: no controlled experiment has yet proven that generative engines treat products accompanied by cryptographic attestations differently. This is the number one priority of our research. But the architecture is solid — and regulation is making it inevitable.
The DPP as an Accelerator
The European Digital Product Passport, effective from 2027 for the first product categories, will require machine-readable and verifiable product data. The format prescribed by the UNTP recommendation relies precisely on W3C Verifiable Credentials. In parallel, eIDAS 2.0 establishes the infrastructure for European Digital Identity Wallets for credential issuance and verification across the single market.
The opportunity is clear: since VC infrastructure will need to be set up for the DPP anyway, it makes sense to leverage it immediately as a trust signal in generative engines. The marginal cost is near zero — the infrastructure is the same. The first movers will transform a regulatory constraint into an AI visibility advantage, while others will simply check a compliance box.
Key Takeaways
The shift from SEO to GEO raises the fundamental question of online commerce: why trust this product over another? AI needs signals it can read, verify, and compare — and that resist falsification. This is exactly the gap that verifiable cryptographic attestations are designed to fill.
The players investing now in these trust signals are not just making a technology bet. They are building a durable competitive advantage for a world where automated recommendation is the norm — not the exception.
This article is based on research published by Keyban: "SoK: Trust Signals in AI-Mediated Commerce — How Generative Engines Select and Rank Products" (2026), a systematization of 38 studies covering GEO, platform economics, and decentralized identity.