Sweden's consumer watchdog is cracking down on misleading retail marketing

What KO is doing

The Swedish Consumer Ombudsman (Konsumentombudsmannen, KO) has stepped up enforcement against misleading marketing in retail ^[1]. The regulator has initiated dozens of supervisory cases, issued around thirty injunctions and started a review of the sector's quality labelling.

The focus is on claims that mislead: false statements about products, unclear pricing, misleading environmental claims and marketing that gives consumers the wrong impression about what they are buying ^[1].

KO emphasises the importance of consumers not being misled. That sounds obvious, but it matters because the retailer is responsible for everything that appears on its platform, including claims made by third-party sellers and, increasingly, by chatbots.

Why chatbots make it harder

Retailers have long used automated systems to answer questions, suggest products and guide shoppers through a purchase. When those systems are powered by AI, the risk changes.

Think of it like this: a static FAQ page says the same thing to everyone, and the retailer controls every word. A chatbot improvises. It generates answers on the fly, and those answers can vary depending on how the question is asked. If a chatbot promises something the terms don't allow, or contradicts the product page, that is still the retailer's marketing.

Under Swedish and EU law, a retailer is responsible for the marketing that appears on its site ^[2]. That includes what a chatbot says. If the chatbot misleads, it is the retailer who has to answer for it.

The evidence gap

When a consumer complains about what a chatbot promised, the evidence question is simple: what did it actually say?

Often, the customer has a screenshot. That becomes the record of the conversation. If the retailer has nothing to show in return—no log of the session, or a log it could have edited—the screenshot is the only evidence that matters.

That is how the Air Canada case played out in 2024. A customer asked the airline's chatbot about bereavement fares, got the wrong answer and kept a screenshot. In a tribunal in British Columbia, the airline argued it could not be held liable for information from the chatbot, as if the bot were a separate legal entity. The tribunal did not accept that: the airline was responsible for everything on its website, whether it came from a static page or a chatbot ^[3].

The customer had the screenshot, the airline offered no evidence to the contrary, and the tribunal ordered compensation. The point for anyone running a customer-facing AI is straightforward: when something is disputed, the side without its own record starts on the back foot.

What the AI Act requires

Since 2 August 2026, the AI Act has required that people be told when they are interacting with an AI system, unless it is obvious from the context ^[4]. There are narrow exceptions, for example when an AI is used to detect or prevent fraud, but those do not generally apply to retail.

That transparency rule is about making it clear who—or what—the customer is talking to. It does not, on its own, deal with the evidence question. Knowing you spoke to an AI is one thing. Proving what it said is another.

Sealing what the chatbot said

This is where a digital witness fits.

The thing to seal is each answer, or each conversation when it ends: the question, the chatbot's response, the time and the notice that the customer was talking to an AI. Sealing takes a fingerprint of the record, not the content, so the conversation itself never leaves your systems. If it is challenged later, anyone you share the record with can check that it is the one generated at the time, unchanged since.

Your system sends the fingerprint to WitniumChain through the REST API, the SDK or the MCP server, and gets a witness ID back. To check, you—or the customer, or an ombudsman—calculate the fingerprint again and compare it with the sealed one. If they match, the record hasn't changed. Change a single character and it won't match. The first 2,000 witnesses a month are free. Read the docs.

A sealed record doesn't show that the chatbot's answer was correct, or settle whether the retailer is bound by it. It shows that the record is the one the chatbot generated, and that nobody has quietly rewritten it since.

How it would work

1. The chatbot generates an answer and your system seals it with a digital witness at that moment.
2. If the customer later disputes what was said, you have your own record with a fingerprint anyone can check.
3. The customer can verify the record themselves, so it is not a matter of trusting the retailer's word.

What this means for you

If you run a site with a chatbot:

• You are responsible for what it says, just as you are for every other piece of marketing.
• Since August 2026, you must tell customers when they are talking to an AI, unless it is already obvious.
• When a customer questions what the chatbot promised, your own checkable record of the conversation is the strongest position to be in.
• A screenshot is evidence. If that is all there is, it is the evidence that counts.

If KO or a court asks what your chatbot said on a particular day, the question is whether you kept a record, and whether anyone can trust it. That is a question to answer before the first complaint, not after.

What to watch

KO's work on misleading marketing in retail is ongoing. The review of quality labelling schemes will clarify what retailers can and cannot claim about their products.

For AI systems, the next area to watch is automated decision-making about people—credit, recruitment, access to services. Those systems will fall under the high-risk rules from 2 December 2027 for systems in Annex III, with logging requirements under Article 12 ^[5]. Even before that date, the question of who keeps the record, and whether it can be checked, applies just as much as it does to chatbots today.