Review sponsored AI chatbot answers under section 2 subsection 4(2) UWG, disclose the commercial purpose clearly and document answers, instructions and versions.
, Mag. Bernhard Brandauer, Rechtsanwalt
An answer in an AI chatbot may look like a neutral recommendation even though a company paid for the mention or influenced the selection of the answer. Users then need to be able to recognise the commercial purpose before relying on the answer for a decision.
Section 2 subsection 4(2) of the Austrian UWG covers commercial practices that conceal material information, provide it unclearly or fail to identify the commercial purpose when it does not follow directly from the circumstances. Article 7(2) of Directive 2005/29/EC follows the same principle. The fact that a language model produced the wording does not change that assessment.
This article distinguishes sponsored chatbot answers from organic answers, paid search placement and AI-generated advertising claims. It focuses on commercial influence, the disclosure in the interface, the roles of the parties and the preservation of evidence.
This short review path distinguishes preparation, an answer that is already visible and a received objection. You can then gather the relevant material in an orderly way.
Payment, instructions, presentation and logs shape the initial review.
For an initial review, payment, a commercial agreement, prompt instructions, ranking rules and the visible disclosure matter.
Identify the advertiser, provider, compensation or other benefit, prompt and ranking instructions, planned answers and the position of the disclosure. Test realistic questions and save the approved version.
Preserve the chat, interface, disclosure, timestamp, model and prompt version, and the question that produced the answer. Review alternative wording and mobile presentation before changing the output.
Preserve the complete letter, challenged answer, screenshots or exports, commercial agreements and internal approvals. Only then decide on disabling, correcting or responding to the output.
A commercial purpose is likely where an answer is intended to promote a product, service or business economically. The specific commercial practice remains decisive. An answer can therefore have a commercial character even if it is not a conventional advertisement and does not contain a visible buy button.
Several facts should be considered together: who initiated the answer, who benefits from the mention, what consideration is provided and what instructions apply to selection, order or wording. The stronger the connection between these factors and the output, the less the answer resembles a purely editorial or neutral recommendation.
A lack of disclosure is particularly problematic where the chatbot creates the impression of an independent selection. Users may assess an answer differently once they know that a provider paid for the mention or influenced it in its own interest. The specific suitability to influence a commercial decision must still be assessed in context.
A direct payment is a strong indicator, but it is not the only possible form of influence. A preferred position under a distribution agreement, a commission for referred contacts, free data or instructions favouring specific products may also support a commercial classification.
The technical implementation may operate at several levels. A provider may prefer a product database, an advertiser may define answer components or a platform may insert a paid recommendation into the dialogue. The legal review must therefore be able to trace the path from the agreement to the specific output.
A mere mention of a product is not automatically a sponsored answer. Without economic influence, a factual product description may be assessed differently. Conversely, calling something a recommendation does not conceal a payment or fixed instruction.
The disclosure must be understandable when the answer is read and assessed. A clear designation such as “Sponsored answer” or “Advertising” can be suitable if it appears directly with the affected answer and identifies the relationship to the named provider.
The design must reflect the dialogue format. A notice shown only on a general help page, in terms of use or after several clicks may reach the user too late. If a sponsored answer card contains a product and a further action button, the commercial character should appear within the same visible unit.
Vague expressions such as “partner content”, “preferred” or an unexplained symbol may conceal the economic meaning. Colour, font size, position and mobile presentation form part of the overall impression. If a sponsored answer is copied into another channel, check whether the disclosure remains attached to it.
An organic answer is generated under the system’s factual or technical rules without a company paying for the specific mention or agreeing a preferred treatment. A sponsored answer is based on economic influence that users must be able to recognise.
A product-related answer issued by a company about its own offer is a different case. The surrounding context may make the commercial purpose immediately clear. Whether an additional disclosure is needed therefore depends on the interface, sender, presentation and the expectations of the people addressed.
The article on paid placement in search results concerns disclosure of a financially influenced ranking. A chatbot also requires review of how the dialogue frames the recommendation and whether it appears independent. Native advertising and hidden advertising provide a related comparison, but they do not replace review of the specific chatbot interface.
A sponsored answer may involve the advertiser, chatbot provider, platform, agency and data supplier. The roles should be described economically as well as technically. The key questions are who initiated the output, who influenced the selection and who controls the visible disclosure.
The advertiser should provide its products, permitted claims and disclosure requirements in full. The chatbot provider must implement the agreed placement in the actual interface. An agency should document approvals, changes and tests so that the delivered answer can be linked to a specific version.
Contracts can distribute tasks, but they do not remove the need for joint control. The party controlling the user interface should test the disclosure before release. The party supplying product data or answer instructions should verify that they are accurate and do not create an impermissible exaggeration.
A later review needs more than the name of the model used. Preserve the agreement or campaign instruction, product data, relevant prompt, answer, visible disclosure and time of output together.
For dynamic systems, also record the model version, retrieval or data state, selection rules, target group and any A/B tests. Preserve the interface in which the answer appeared. A text export often does not show whether the notice was above, beside or only after the answer.
The records should use unambiguous versioning. When a notice, answer rule or product database changes, it must remain clear which version was visible before and which was delivered afterwards. This is important for internal approval and for responding to an objection.
A single approval test is often insufficient for a dynamic chatbot. Answers can change because of new data, model updates, prompt changes or a different user question. The disclosure must therefore remain visible and understandable across variants.
Define test questions that represent typical user intentions. Check the answer, product order, notice, link and presentation on smaller screens. If answers are shortened or summarised automatically, the commercial classification must not disappear from the dialogue.
If a system displays a paid answer only in certain countries, periods or target groups, those rules must be included in the monitoring. The article on AI-generated advertising claims shows in a different context why model state, approval and actual delivery should be preserved together.
After an objection, first preserve the specific output without changing it. This includes the user question, answer, notice, visible surroundings, time, model and data state, and the commercial agreement behind the placement.
Then assess whether the notice was missing, appeared too late, failed to identify the provider or was practically unclear because of the interface. A correction may be necessary, but it should not replace preservation of the earlier version. If the output continues, also consider temporary suspension or adjustment.
The appropriate response depends on the letter, the legal position asserted and the evidence. An UWG warning from a competitor or association should be preserved in full and reviewed legally. The technical explanation that a model generated the answer does not settle the issue.
No. The key questions are whether there is a commercial purpose and whether it follows directly from the circumstances. Payment, commission, preferred placement or binding instructions may support a sponsored classification. The specific interface and overall impression remain decisive.
That cannot be assumed in every case. The notice must be understandable where the specific answer is perceived. A general notice elsewhere may come too late or fail to explain the connection between the answer and the commercial influence.
A paid placement is not unlawful solely because money is involved. The lack of clear disclosure can become problematic where the answer appears to be an independent recommendation and can influence a commercial decision.
Useful records include the agreement, compensation model, product data, answer and prompt instructions, disclosure, test questions, approvals, model and data state, and screenshots or exports of the actual interface. Dynamic systems also require variants and change logs.
Review ranking, consideration and disclosure in search results separately.
Classify commercial communication and visible disclosure in the presentation.
Preserve wording, evidence, approval and actual delivery.
Organise the objection, cease-and-desist demand and evidence.
Topics
Give us a concise outline of the situation. We will clarify which documents are needed and which next step is appropriate.
Office
BRANDAUER Rechtsanwälte GmbH Giselakai 51 5020 Salzburg Österreich
Phone
+43 662 6280000