When a marketing message looks like an invoice, *who answers for it*?

When the category misleads

Konsumentverket, Sweden's consumer ombudsman, has criticised insurance broker Söderberg & Partners for the way its marketing appears in Kivra, the digital mailbox service.^[1] The marketing message was placed under the category 'Payments' and looked like an invoice, even though it was an offer. That, the ombudsman says, creates a misleading impression.

The case is about an insurance broker and a digital envelope, but the underlying question will matter more as AI writes more messages: when the envelope looks official, who is accountable for how the message is presented?

It's like putting a flyer for a sale inside an envelope marked 'Final notice'. The flyer itself might be fine, but the envelope changes how you read it.

What the ombudsman said

Konsumentverket's criticism is directed at how the marketing was categorised and presented, not just what it said. Under Swedish marketing law, the Marketing Practices Act (marknadsföringslagen), marketing must not mislead.^[2] An offer that looks like a bill crosses that line.

The ombudsman has not issued a formal injunction, but the criticism is public and puts the practice under scrutiny. Söderberg has not been fined, and the case has not reached a court. The point, though, is made: presentation matters.

The AI Act asks who is speaking, not who is sorting

Since 2 August 2026, Article 50 of the AI Act has required that people be told when they are interacting with an AI system.^[3] If an AI writes the message, the recipient must know.

But the AI Act does not address how messages are categorised or presented by the platforms that deliver them. Article 50 is about disclosure, not about how a message lands in someone's inbox or under which heading.

That gap is starting to show. An AI can write a perfectly compliant message, and a platform can then put it under a heading that makes it look like something else entirely. The question is where accountability sits.

The envelope and the letter

In the Söderberg case, the message itself was marketing, and the platform put it under 'Payments'. The broker chose the channel and the platform chose the category. Under the Marketing Practices Act, the trader is responsible for how the marketing is presented.^[2] That includes, the ombudsman suggests, the category in which it appears.

In practice, that means a trader using a third-party platform needs to know how its messages will be categorised and whether the presentation is likely to mislead. The trader does not control the platform, but it is still answerable for the impression the message creates.

The same logic applies when an AI writes the message. The system that generates the content does not decide where the message appears, but the organisation that deploys the system is responsible for both.

What this means for you

If your organisation uses AI to generate messages, or if you send marketing through third-party platforms, here are three questions worth asking:

1. Do you know how your messages are categorised when they reach the recipient, and does that categorisation match what the message is?
2. If a customer disputes what a message said or how it was presented, do you have your own record of what was sent and when?
3. If a platform changes how it sorts messages, do you have a way of knowing before a complaint lands?

The last question is the one that matters for evidence. A customer who says 'this looked like a bill' can take a screenshot. If you do not have your own record of what was sent and how it appeared, the customer's version becomes the only version.

Seal each message as it is sent

This is a case where a digital witness fits well. The thing to seal is each message as it is sent: the content, the recipient reference, the category under which it was placed (if that information is available) and the timestamp. Sealing takes a fingerprint of the message, not the message itself, so the content stays with you.

If the message is questioned later, you calculate the fingerprint again and check it against the sealed one. If it matches, the message has not changed since it was sent. Change a single character and it will not match.

What it shows is that the message you have on file is the one sent at that time. What it does not show is whether the platform presented it correctly, or whether the message was compliant with marketing law. Those are separate questions.

Anyone you share the record with can check it: the customer, an ombudsman or, if it comes to it, a court. Your system sends the fingerprint to WitniumChain through the REST API, the SDK or the MCP server, and gets a witness ID back. The first 2,000 witnesses a month are free.

How it would work

1. Your system sends a fingerprint of each message as it goes out.
2. The fingerprint is sealed, and you store the witness ID with the message.
3. If a dispute arises, you produce the message, recalculate the fingerprint and check it against the sealed one. If it matches, the message has not been altered.

What to watch

The AI Act's transparency requirements are already in force, and more guidance is expected from the AI Office on how Article 50 applies in practice.^[3] The Swedish Consumer Agency publishes guidance on the Marketing Practices Act, and cases like this one often lead to clearer rules about what counts as misleading presentation.^[2]

For organisations using AI to generate customer-facing content, the lesson is simpler: the message and the envelope are both yours to answer for. Keep your own record of both.