AI Algorithms Shaped $229B in Holiday Sales Before EU AI Act Disclosures

AI recommendation algorithms shaped an estimated $229 billion in global online sales during the 2024 holiday season, according to Salesforce data, and until August 2, 2026, most of those interactions carried no disclosure that AI was involved. The EU AI Act’s Article 50 changes that, imposing transparency obligations on any provider whose AI outputs reach users in the EU, including US companies operating across borders.

How Recommendation Engines Work

These algorithms run on nearly every major digital platform, from streaming services to e-commerce. They analyse user data, past purchases, viewing habits, clicks, time on page, to build preference profiles, then apply two broad techniques: collaborative filtering, which surfaces items that similar users liked, and content-based filtering, which recommends items resembling what a user has already engaged with. The output is a personalised environment that shapes perception and purchase behaviour without requiring the user to ask for anything.

The persuasive effect is reinforced by what researchers call filter bubbles. An algorithm that identifies a preference, a product aesthetic, a news angle, a price range, will systematically prioritise matching content and deprioritise everything else. Serendipitous discovery falls away. The user’s digital experience narrows, but gradually enough that the narrowing is rarely noticed.

Direct Effect on Purchase Behaviour

Retailers deploy AI well beyond personalised product carousels. Dynamic pricing is one of the more visible applications: Walmart has discussed digital shelf tags that can adjust prices every 10 seconds in response to demand signals. AI also generates tailored ad copy and adjusts the order in which search results appear, all calibrated to the inferred intent of the individual user. The cumulative effect is a purchase path that feels frictionless, and that friction reduction is itself a commercial outcome the system is optimised to produce.

Consumer appetite for these tools is considerable. A 2024 IBM study found roughly three in five consumers expressed interest in using AI for shopping. A Bloomreach survey in March 2025 found 61% had already used general-purpose AI tools to help them shop online, with over half reporting their search habits had become more conversational in the preceding year.

The Disclosure Gap

The influence of these systems has remained largely invisible by design. Recommendations appear; they do not announce themselves. Algorithms adapt incrementally, so their suggestions feel intuitive rather than engineered. Without disclosure, users have no prompt to evaluate whether a recommendation reflects their actual interests or the system’s commercial objectives. That gap is precisely what Article 50 targets.

What Article 50 Requires

The EU AI Act’s Article 50 transparency rules require providers to design systems that inform users when they are interacting with AI, a chatbot, for instance, or are exposed to AI-generated content, including synthetic audio, images and text. For deepfakes and AI-generated material on matters of public interest, deployers must also disclose the artificial origin unless substantial human editorial review has occurred. The obligations extend beyond high-risk AI categories and apply to any provider whose outputs reach EU users, regardless of where the company is headquartered. The European Commission published guidance on July 20, 2026 to clarify compliance expectations. Non-compliance carries penalties of up to €15 million or 3% of global annual turnover.

For enterprises operating recommendation, personalisation or content-generation systems at scale, the practical question is whether current disclosure mechanisms, if they exist at all, meet the regulation’s specificity requirements, or whether architecture changes will be needed before the August deadline.