AI-Generated Images in Marketing: Credibility, Labeling, Risks

Why are AI images a question of credibility?

Image generators produce, in seconds, a photo of a smiling team in a bright workshop, a modern production hall or a satisfied customer shaking hands. For a small business without a photographer, that is tempting. Its own pictures are often dark, the workshop is not tidy, and the team does not like being in front of the camera. But an image on a business’s website is not just design. It tells the visitor: this is what it looks like here.

This article therefore answers the question: When do AI-generated images help a small business’s marketing, when do they cost credibility, and what do labeling obligations and copyright law require? We first clarify the terms and a model from sign theory that explains why photos work as evidence. Then we look at findings from experiments, assess the legal situation and weigh two views.

A photo says: this existed. An AI image says: this is what it could look like. In sales, that is a big difference.

Terms: AI image, editing, deepfake

Generative AI refers to software that creates new content: text, images, audio or video, as the article on AI tools in marketing notes. An AI-generated image is created entirely from a description, without a camera having recorded anything. Distinct from this is AI-assisted editing of a real photo, such as removing a cable, brightening or extending the edge of the image. A deepfake is a deceptively real, artificially generated or altered image, video or audio recording that shows or appears to show real people, places or events.

The thinking model: photo as imprint, AI image as likeness

The philosopher Charles Sanders Peirce distinguished signs by how they are connected to what they stand for. An icon resembles its object, such as a drawing. An index is physically connected to its object, like a footprint to the foot or smoke to the fire. A photo is both: it resembles what is shown, and it was created by light coming from that object. This second property makes it evidence.

An AI image has only the first property. It resembles a workshop, but no workshop left it behind. As long as the viewer does not know this, they read it like a photo, that is, as evidence. As soon as they recognize it or find out, the evidential value disappears, and with it often trust in the other images. The key question for every image follows: is it meant to prove or merely to illustrate?

What do experiments show about AI images and trust?

In the Journal of Theoretical and Applied Electronic Commerce Research, Li, Goh and Wu (2026) examined in three experiments how AI images of a travel destination work. Participants, adults from China, saw either an AI image or a photo taken by a tourist of the same place. In the first experiment, with 101 valid responses, the AI image reduced perceived authenticity and, through it, the intention to visit the place. In the second and third experiments, with 166 and 162 participants, additional information on the place’s cultural background mitigated the drawbacks, to different degrees depending on the type of information. The authors explain the effect by noting that AI images often lack the context real photos bring with them.

Other studies show a similar pattern with gradations. In an experiment with 521 adults in South Korea, Suh (2026) found that an AI endorser did not change purchase intention directly, but through perceived authenticity and the trust built on it. The more skeptical participants were of AI, however, the smaller the human endorser’s authenticity advantage became. With 474 participants, Kim (2026) showed that openly pointing out the weaknesses of an AI ad increased the brand’s perceived authenticity. The article on the EU AI Act cites a study by Feng and colleagues (2026) in which disclosing AI authorship indirectly worsened attitudes toward the brand because viewers assumed less effort behind the ad.

All studies come from consumer markets, from tourism, fashion and advertising, not from business-to-business markets. For a B2B business the situation is, if anything, sharper: a buyer evaluating a supplier uses images as evidence of capacity, order and quality. There, evidential value is the reason the image is on the page at all.

What do the AI Act, copyright law and competition law require?

The EU’s AI Act contains transparency obligations in Article 50 that have applied since August 2, 2026. Two of them matter for images. The provider of an image tool must mark its outputs in machine-readable form as artificially generated. For systems placed on the market before August 2, 2026, the transitional period for this runs until December 2, 2026. And as a deployer, that is, a business using the tool professionally, you must disclose that a deceptively real image of real people, places or events was artificially generated or altered. A recognizably artistic or schematic image does not trigger this obligation as currently understood. The timeline and other obligations are described in the article on the AI Act.

Under Section 2(2) of the German Copyright Act, copyright protects only personal intellectual creations. An image created from a short description without significant human creative input is therefore generally not protected. Competitors may then be able to copy it. Still open is how far training such tools with other people’s photos is permissible. In December 2025, the Higher Regional Court of Hamburg, in a case brought by a photographer against the association LAION, held that using a photo for a training dataset was permissible and granted leave to appeal to the Federal Court of Justice.

Finally, competition law prohibits misleading advertising. An AI image showing a hall the business does not have, a project that never existed or employees who do not exist can be misleading, whether or not it is labeled. For images of real people, personality rights also apply.

Two views: a tool like any other or a trust risk?

The first view sees AI images as a tool like photo editing or stock photos. It says: purchased stock photos do not show your own business either, and nobody takes them as evidence. AI images are cheaper, better tailored and free of licensing issues with picture agencies. In its favor is that many images on websites really only illustrate and no reader takes them as proof.

The second view sees a trust risk. It says: visitors are less and less able to recognize AI images and therefore read them as photos. If the deception is noticed, the business loses not only this image’s evidential value but trust in all the others. In its favor are the findings on authenticity and the logic of persuasion knowledge described in the article on persuasion without tricks: a trick that is recognized costs the reason the customer should believe you next time.

I consider both views accurate, but for different images. For explanatory and mood images, the first applies: there an AI image is a tool like a stock photo, and a note in the caption costs nothing. For evidence images, the second applies without qualification. The line runs not between technology and handwork but between illustrating and claiming.

The proof-or-picture test: a check before every publication

Model, findings and law yield a check that I call the proof-or-picture test. It consists of three questions answered before every image.

1. Does the image claim something about the business?

   Does it show people, rooms, machines, products or projects that the reader will take to be the business’s? Then it is an evidence image and must be real. Editing that changes nothing essential is allowed.

2. Could someone take it for a photo?

   Is an explanatory or mood image deceptively real, and does it show or appear to show real people, places or events? Then it is disclosed, ideally right in the caption.

3. Would I be uncomfortable if a customer asked?

   If so, the image is the wrong choice, even if it is legally permissible. This question corresponds to the honesty test: would the customer agree to the means if they knew how it is used?

For evidence images there is no way around your own photos. How a business gets usable photos of its own with little effort, and which image rights apply, is described in the article on images on the website. For reference projects, what the article on the references page notes also applies: logos and names only with the customer’s permission.

The bigger picture

AI images make visible what the effect of images in marketing has always rested on: the assumption that they show what exists. This assumption weakens the better image generators become. For businesses, it follows not that images lose value but that real images gain value when they are recognizable as real: with names in the caption, with place and date, with people you meet again in conversation. The same rule applies to customer testimonials, as the article on customer testimonials and references describes: verifiability creates trust.

The limits of this assessment are clear. The experiments come from tourism and consumer advertising in Asia, not from business-to-business markets in Germany. The law is in flux, the labeling obligation is being specified by guidance, and case law on copyright is not settled. The distinction between evidence and illustration is robust.

What remains and how to start

The question was
AI-generated images help a small business, when they cost credibility and what the law requires. They help with images that illustrate, such as explanatory graphics and mood images. They cost credibility with images meant to prove: team, rooms, products and projects. Experiments show that AI images lower perceived authenticity, and in business-to-business markets authenticity is the reason an image is on the page. Legally, deceptively real AI images of real people, places or events must be disclosed, purely AI-generated images are generally not protected by copyright, and misleading images remain impermissible. The proof-or-picture test turns this into a decision made in a minute before every image.

• Divide all images on the website into evidence images and illustrations.
• Replace AI images among the evidence images with real photos.
• Disclose deceptively real AI illustrations in the caption.
• Introduce a photo rule for the end of every project.
• Clarify with your image tool’s provider whether and how it marks its outputs.