AI images and e-commerce in Switzerland: obligations, EU rules and marketplaces

17 min readPackshotCreator
AI images and e-commerce in Switzerland: obligations, EU rules and marketplaces

A Swiss brand photographs a jacket, changes its colour with AI, then uses the same visual on its website, in a European campaign and on a marketplace. Does it have to state that the image was generated or retouched by AI? In the sources analysed as at 29 September 2026, the mere use of AI in a product image does not trigger a general disclosure obligation in Switzerland. That does not place the image beyond the reach of the law: it must remain faithful to the product shown and must not give rise to incorrect or misleading information.

Three questions are dealt with separately: does the image match the product sold, to which audience is it distributed, and what rules does the sales channel impose? Swiss law, the European AI Act where it applies, and platform rules have neither the same scope nor the same addressees.

For a Swiss e-commerce team, the first check is a practical one: customers must be able to rely on the colour, material, accessories and condition of the item they see on screen.

Key takeaways

  • The Swiss sources analysed as at 29 September 2026 show no general obligation to flag a product image merely because it was generated or retouched by AI. This does not exempt you from the rest of the applicable law.
  • The central Swiss issue is the fidelity of the image: the Federal Act on Unfair Competition (UCA) targets incorrect or misleading information about goods. A conventional photo can raise this problem; conversely, a generated image can be faithful.
  • As soon as an identifiable real person appears, data protection, personality rights and the permissions obtained come into play. The Federal Data Protection and Information Commissioner also considers that any manipulation of that person’s face or image must always be clearly indicated.
  • The AI Act may concern a Swiss company whose images are used in the Union, depending on its role and the way it distributes them. This is not automatic: the fact that a website can be viewed from Europe is not, on its own, enough.
  • A Google, Amazon or Zalando rule applies on the platform concerned; it becomes neither a Swiss legal obligation nor an AI Act obligation.

Verified on 29 September 2026; Swiss and European sources, Google and Zalando checked again on 6 October 2026. This article presents the rules, guidelines and platform policies available on that date. It is not legal advice. The header image is an AI-generated illustration.

Does Switzerland have an “AI Act” in 2026?

No. As at 29 September 2026, the sources analysed show no general, cross-cutting Swiss legislation equivalent to the European AI Act. It would nevertheless be wrong to conclude that “AI is not regulated in Switzerland”: rules apply depending on the context, in particular unfair competition, data protection, personality rights and sector-specific rules. The federal work announced for 2027 is presented at the end of this article.

Do you have to flag an AI-generated or AI-retouched image in Switzerland?

No general Swiss rule requiring a disclosure merely because AI was used appears in the sources analysed. That does not exempt you from analysing the image: the central Swiss question is what it leads people to believe about the product, followed, where relevant, by the people depicted and the context in which it is distributed.

Start by comparing the image with the product

For product photos, the most directly relevant text among those consulted is Article 3 paragraph 1 letter b of the Federal Act on Unfair Competition (UCA). It covers, in particular, incorrect or misleading information about goods or services. The risk therefore does not depend on the technology used, but on what the image leads people to believe about the product actually offered.

A few catalogue examples show where the problem arises:

  • the photo shows a jacket in a red that the brand does not sell;
  • the leather looks smoother, brighter or better finished than the item delivered;
  • the setting presents an accessory as if it were included;
  • a retouch erases a real defect on a second-hand item;
  • a generated perspective visually exaggerates the dimensions of a product.

These are practical applications of the principle of fidelity, not Swiss case law dealing with AI.

The most robust check is to compare the final file, before distribution, with the product, its technical data sheet and the commercial offer, or with the reference of the variant actually sold. A photo of the original colourway does not prove that a recoloured variant is faithful: where possible, compare it with the material or a sample of the variant. In fashion, check the drape and texture; in jewellery, the proportions and the rendering of the metal; for second-hand goods, the condition of the specific item actually sold.

Four renderings of the same bottle: the reference, then a warmer rendering, a cooler rendering and a modified liquid tint.
AI-generated illustration. One bottle, four renderings: the reference remains the product or variant actually sold.

Keep the original visual, a record of the transformations and the approval of the published version. This is particularly useful when several teams work on the image in succession (studio, retouching, agency, brand, marketplace): a colour can drift, a finish can be enhanced or an accessory added along the way, and the published image can move away from the product without any single step appearing problematic.

Eight common situations

Rather than asking whether a feature “uses AI”, start from the published result. For each case: what we advise you to do, then what Swiss law says and, if the image is used in the Union, the European framework. The European detail of each case is analysed in our article on the AI Act and product images.

1. Cut-out or light correction

Removing a studio background or correcting a colour cast can make the image easier to read without changing the product. Even so, compare the retouched file with the original photo: outlines, transparencies and significant details must not have been altered. In Switzerland, no general disclosure obligation was identified for these operations: the question remains the fidelity of the result.

In the Union, the Commission guidelines place certain preparation or minor correction operations under “standard editing”. Whether they apply depends on the actual operation and on the absence of substantial alteration. This is not a Swiss legal category, and it should not be transposed automatically.

2. Recolouring to a variant actually sold

The variant must exist and the rendering must match it. Link the image to a reference that is actually available and validate the colour against a sample or a photo of the variant. For textiles, check the shade, the texture and the way the material renders under light first: recolouring does not always reproduce these faithfully, as we explain on our fashion packshot page. In Switzerland, in our reading, if the generated colour matches a variant actually on sale, the main risk lies in how faithfully the image renders that variant.

In the Union, the texts consulted do not settle the precise classification of this case under Article 50.

3. Non-existent colour

Do not present a generated shade as a variant the customer can order. In Switzerland, in our reading, presenting a colour that does not exist in the range on offer may constitute incorrect or misleading information about the product.

In the Union, the Commission cites as a deep fake an image of a product, in an advertisement or on packaging, that makes it appear non-identical, more attractive or of better quality than it really is. How exactly this applies to a product page remains to be assessed case by case.

4. Defect or scratch removed

Distinguish dust caused by the shoot from the condition of the item sold. For a new, mass-produced product, a scratch present only on the unit photographed takes the image away from the product delivered, and erasing it brings the image closer to it. For a one-off piece, a second-hand item or a refurbished product, the condition photographed is what is being sold: smoothing out the signs of wear on a pre-owned watch changes the information given about the item. In Switzerland, in our reading, removing a real defect of the product may thus alter a relevant characteristic of the item shown.

In the Union, the guidelines distinguish minor corrections from alterations that make the product more attractive or of better quality than it really is. Between these two reference points, the texts consulted do not settle the removal of a scratch or of wear: it all depends on the image.

5. Enhanced material or finish

An enhanced texture can suggest a different quality from that of the product. Check the result on screen against the real product: relief, shine, transparency, stitching, reflections. The software used does not, on its own, determine whether the image is faithful. In Switzerland, in our reading, an artificially enhanced texture or an embellished finish may become problematic if the consumer perceives a quality that the product does not have.

Fictitious taupe grained-leather handbag in full view, with two close-ups of the same bag: the grain and the stitching on the flap, then the lining and the zip.
AI-generated illustration. Grain, stitching, metal and lining are characteristics to check against the reference actually sold; this image is not a photograph of a commercial product.

In the Union, the line between cosmetic correction and substantial alteration is assessed case by case.

6. Real product in a generated setting

A generated environment can help put a product in context. Check, however, that the scene does not attribute any additional characteristic to it: proportions, accessories, suggested use. In Switzerland, a generated setting is not, in itself, identified as triggering a general disclosure obligation. In our reading, it becomes problematic if it changes how the product, its dimensions, its accessories or its use are understood.

In the Union, the Commission’s examples distinguish an aesthetic setting from content likely to mislead about the product.

7. Fully generated or heavily transformed product

Check the image against the physical product or a documented commercial reference, not only against another generated image. A flattering appearance does not prove the real characteristics. In Switzerland, no general disclosure rule was identified; the depiction must nevertheless remain consistent with the product actually sold.

In the Union, the texts consulted do not settle certain classifications where a fully synthetic product is rigorously faithful. They are clearer where the transformation makes the product different, more attractive or of better quality than it really is.

8. Virtual model

In the workflow, distinguish the real garment or piece of jewellery, the synthetic person and the origin of any images used to create that person. In Switzerland, a fully synthetic model does not automatically become an identifiable real person; what remains are the fidelity of the product worn and the platform rules, some of which specifically target generated people.

In the Union, the texts consulted do not settle the systematic classification of a photorealistic virtual model in a still image. The guidelines provide related examples, not a rule covering all e-commerce virtual models.

When a person appears, check their provenance

A photograph of a real model or an identifiable influencer calls for a different analysis from the product alone. The Swiss sources consulted refer in particular to the Federal Act on Data Protection (FADP), to the positions of the Federal Data Protection and Information Commissioner (FDPIC) and to the rules on image and personality rights.

An AI transformation does not automatically remove the personal nature of the data. If AI transforms a model’s photo, ask yourself whether the person remains recognisable, which images were used to generate the new visual and which uses were envisaged at the time of the shoot. In our reading, if the person remains identifiable, or if the workflow reuses their images to create a new depiction, the legal basis for the processing and the scope of the permissions obtained must be checked. The sources consulted do not allow us to consider that consent given for a photo shoot necessarily covers all future generative transformations: check this against the contract, the context and the use concerned.

The FDPIC goes further for this case: according to its page on AI and data protection, the use of programs that enable the manipulation of faces, images or voice messages of identifiable persons must always be clearly indicated, where such use is not quite simply unlawful under criminal law. This is the position of the authority responsible for supervising the application of the FADP; it concerns identifiable persons, not every product image generated or retouched by AI.

Three portraits of fictitious women suggesting a studio shoot, a transformation of the same person and a fully synthetic person.
Illustration entirely generated by AI, with fictitious adult characters, suggesting photography, transformation and synthesis; the visual rendering alone cannot establish a person’s provenance.

A fully synthetic model, with no identifiable real person, calls for a different analysis. Product fidelity then remains central: cut, length, volume, drape, colour or the apparent size of a piece of jewellery must not be depicted in a misleading way. Also check the origin of the workflow: the sources consulted do not allow all synthetic people derived from real images to be treated in the same way, and a character created from photographs of people is not necessarily equivalent to a fictitious character created without any identifiable human source.

When the AI Act may concern a Swiss company

For an image, Article 50 of the AI Act provides for two distinct obligations. The provider, which develops an AI system or has it developed, and places it on the market or puts it into service under its own name or trademark (Article 3, point (3)), must, in the cases covered, mark generated or manipulated images in a machine-readable format. The deployer, which uses the system under its authority (Article 3, point (4)), must add a visible disclosure where the image is a “deep fake”: an image that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful.

A Swiss company is not outside the scope of the AI Act as a matter of principle. Article 2(1)(c) provides for situations in which providers or deployers established or located in a third country may fall within its scope, where the output produced by the AI system is used in the Union. Article 2(1)(a) also covers providers placing an AI system on the market or putting it into service in the Union, irrespective of whether they are established in the Union or in a third country.

The Commission guidelines clarify these two cases. A provider established outside the Union is covered if the output of its system is used in the Union; incidental, unforeseeable or unauthorised downstream use is not enough, on its own, where the system is neither placed on the market nor put into service in the Union (point 10). A deployer established outside the Union is covered where it itself intends the outputs to be distributed and used in the Union, by directing or authorising their distribution in the Union, including by publishing deep fakes on the globally accessible internet; it is not covered where the content reaches an audience in the Union through unforeseeable channels beyond its control (point 13).

Situation to examineQuestion to ask
Swiss brand publishing for the Swiss marketWhich uses and which territories were intended? Does the company direct or authorise distribution in the Union?
Swiss campaign also aimed at the UnionDid the company direct or authorise the distribution of the outputs in the Union?
Swiss studio producing for a European brandWho uses the system under its authority, that is, who decides to use it and how to use it?
Image sent to a European marketplaceWhat is each party’s role under the AI Act and, separately, what are the platform’s rules?
AI tool from a vendor inside or outside the UnionThe vendor’s obligations and those of the Swiss user remain separate: who is the provider, who is the deployer?

For the studio, the Commission specifies that a company remains the deployer where service providers operate the system on its behalf, under its responsibility and control, and that a company that merely commissions an advertisement, without deciding whether or how the agency uses AI, is not a deployer (points 12 and 14). These roles are not assigned solely on the basis of the commercial contract. The Commission also cites a company established in a third country that uses an AI system to generate a deep fake of a celebrity in an advertisement distributed in the Union: that company is a deployer within the scope of the Regulation (point 14). Applying this to a product image is a matter of interpretation, which depends on the image and its use.

The mere global accessibility of a website therefore does not, on its own, support the conclusion that a Swiss company is subject to Article 50 as soon as an image is visible in Europe. In our reading, a situation remains purely Swiss as long as the company neither directs nor authorises the distribution of the images in the Union; intermediate situations are assessed case by case. The actual role of the parties, the system used, the nature of the content and the intended use of the images must all be assessed.

Bottle in the centre of three panels representing Swiss law, the European framework and platform rules.
AI-generated illustration. Swiss law, the possible application of the European framework and platform rules are three distinct layers: one cannot automatically be inferred from another.

For details of the transparency obligations, machine-readable marking, deep fakes and exceptions, see our article “AI Act 2026 and e-commerce product images: what brands actually have to disclose”.

Google, Amazon and Zalando: a third set of rules

A platform policy is neither Swiss law nor, in itself, an AI Act obligation. It applies to what you send to the platform, under its own terms.

The same fictitious bag shown three times: on a white background, in a staged scene with a stone and foliage, and in a close-up of its clasp.
AI-generated illustration. The same fictitious item can be turned into visuals intended for different distribution contexts; the policies actually in force remain specific to each platform and must be checked at the time of publication.

According to the policies analysed on 29 September 2026 (6 October 2026 for Google and Zalando):

  • Google Merchant Center requires images created with generative AI to contain IPTC metadata indicating the digital source of the content (DigitalSourceType family), and asks that this metadata not be removed when it is present;
  • Amazon asks, for media containing a photorealistic person fully generated by AI, for XMP tagging with dc:subject and the value contains-synthetic-performer. This tagging does not apply to images without people, to non-photorealistic people or to real people, even if modified by AI;
  • Zalando focuses on product fidelity (textures, shapes, proportions, finishes, shadows, reflections, realism of the setting). Its guidelines, updated on 1 October 2026, reject any visible disclosure on the image, such as overlaid text or a visible watermark, and ask partners not to supply content that would require such a disclosure by law, such as deep fakes. For invisible marking, the Zalando guidelines updated on 1 October 2026 carry the heading “required by December 2026”: they “strongly recommend” it for all AI-generated content, then ask for identifying data to be embedded in the metadata or as an invisible watermark, without naming a standard. As the page combines a recommendation and a requirement, reread the clause in force before any submission.

These requirements concern the way each platform receives images. Do not turn this dated comparison into a universal file rule: before each campaign or export, recheck the platform’s current instructions and the exact type of content concerned.

Are C2PA and IPTC mandatory in Switzerland?

No: no general Swiss obligation to use C2PA or IPTC was identified in the sources analysed for product images. Article 50 of the AI Act provides, for the provider of the tool and in the cases covered, for machine-readable marking, without imposing any particular technology. The technical requirements come from the platforms: IPTC metadata at Google Merchant Center for images created with generative AI, an XMP keyword at Amazon for fully generated photorealistic people, and invisible marking announced by Zalando for December 2026, with no standard named. C2PA remains a content provenance standard, not a Swiss legal obligation. How to prepare these files and check that the metadata is preserved is covered in detail in our article on metadata for AI-generated product images.

A validation method for your team

  1. Keep the original photograph or a physical reference of the product.
  2. Identify what the AI has changed: cut-out, colour, texture, defect, background, proportions, accessories, person or the product itself.
  3. For each variant, compare colour, material, dimensions, accessories and condition with the product actually offered.
  4. Identify any identifiable real person, the human sources used and the corresponding permissions.
  5. Specify where the image is going: Swiss market, international website, European campaign, marketplace or product feed.
  6. If the Union is concerned, assess the company’s role and whether the AI Act may apply, instead of applying European rules automatically; check the platform’s technical requirements separately.
  7. Keep the final approved version and check the file actually delivered, including after conversion or compression.

This method separates three validations that are often confused: the product, the applicable law and the platform. It is a recommended production method, not a legal register that some “Swiss AI image law” would impose. Correctly completed metadata does not make a misleading depiction legally acceptable.

What may change in 2027

Switzerland has signed the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law. According to the SME Portal of the State Secretariat for Economic Affairs (SECO), in a page dated 23 September 2026, a draft for consultation is to be drawn up in spring 2027; the title of the same page announces it “by early 2027”. The Federal Chancellery’s undated “Regulation of AI” page, for its part, mentions a draft to be drawn up “by the end of 2026”. This is an announced work schedule, not an obligation applicable to e-commerce catalogues: no content of this future text should be presumed today.

Sources

Sources consulted on 29 September 2026; Swiss and European sources, Google and Zalando consulted again on 6 October 2026. Swiss sources are linked in their English versions; the English translations of federal acts published on Fedlex are provided for information only and have no legal force.

Swiss law

Official Swiss guidance

European framework

Platform policies

Technical standards

Back to blog

Discover our solutions

Automate your photo production and calculate your return on investment.

Related articles

CallRequest a demo