AI Made Your Visuals. Will Your Customers Still Trust You?

Your customers can't tell. Until they can.

Here's the uncomfortable part. In a 2024 series of experiments, people could not tell AI-made brand content from human-made content. They only turned on it once they were told what it was.

So the problem isn't the pixels. It's the moment of discovery.

Every brand now has an AI decision to make, whether it admits it or not. Moodboards, product shots, social posts, ad variants: generative tools can make all of them faster and cheaper than any studio. The question is not "can we?" It's "what happens to trust when people find out?"

The core question of our research is, how generative AI in visual communication affects consumer trust on social media. This post pulls together what that research and the wider literature say, three real brands that learned it the hard way, and the rules we work by at Boulder.

What the research says

The clearest evidence comes from Jasper David Brüns and Martin Meißner, published in the Journal of Retailing and Consumer Services in 2024. Across three experiments, they found that using generative AI for social media content lowers how authentic followers think a brand is. That drop in authenticity then drives negative reactions toward the brand.

Three details matter for anyone running a brand:

• Disclosure is the trigger. The negative effect can be set off simply by telling people AI was used. The Technical University of Munich's summary of the work notes that participants couldn't distinguish AI content from human content, yet judged it negatively once its origin was revealed.
• Assist, don't replace. The penalty shrinks when AI is positioned as helping humans rather than doing the work instead of them.
• Hiding it is no longer an option. The study exists because social media platforms started requiring disclosure of AI content.

Regulation is closing the same door. Since 2 August 2026, Article 50 of the EU AI Act requires anyone publishing AI-generated deepfake content to clearly disclose it, with a lighter regime for evidently artistic or fictional work. Georgia is not in the EU, but any Georgian brand selling into Europe, or advertising to European audiences, now works under these rules.

The direction is clear: disclosure is becoming the default. So the real question is not whether to disclose. It's how to use AI in a way that survives disclosure.

Three brands, three ways to lose trust

1. Hide it, deny it, get caught: Wizards of the Coast

In January 2024, the publisher of Magic: The Gathering posted a promotional image for a new card set. Fans quickly spotted the tells, including warped numbers on a pressure gauge in the background.

The company's first answer was a flat denial: the art, it said, was made by humans, not AI. Three days later it reversed, admitting AI components had come in through a third-party vendor and tools like Photoshop.

The damage went beyond one post. Wizards had publicly banned AI art weeks earlier, so the slip read as hypocrisy. Long-time artist Dave Rapoza publicly quit after nearly 17 years of working with them.

The lesson: the AI wasn't the scandal. The denial was. A brand's community will verify you whether you like it or not, and getting caught contradicting yourself costs far more than saying it upfront.

2. Promise a world you can't deliver: Willy's Chocolate Experience

In February 2024, an events company sold tickets to a Wonka-style family attraction in Glasgow. The website was full of dreamy, candy-coloured visuals that were obviously AI-generated, complete with garbled AI text.

Families arrived to a sparse warehouse with a few props. The event was cancelled the same day, parents called the police, and the organisers promised full refunds. It became a global meme.

The lesson: AI made it effortless to show something that didn't exist. A visual is a promise. When the image promises more than reality delivers, you haven't saved on production. You've borrowed trust you can't repay.

3. Disclose honestly, still get burned: Selkie

Selkie, a fashion brand known for romantic, artist-led prints, did the "right" thing in January 2024. It openly said its Valentine's collection prints were built using an AI image generator as a base, which the founder then painted over.

Backlash was immediate anyway. Founder Kimberley Gordon defended the process but said she would not use AI-generated images in future Selkie collections.

The lesson: disclosure isn't a shield on its own. Selkie's customers pay for the feeling of human artistry, so AI hit the exact thing they were buying. Research backs this up: people's preference for human labour is stronger when a product is symbolic rather than purely functional.

Why it happens: effort, proof and peers

Three old ideas explain all three cases.

People price effort. In a classic 2004 experiment, readers rated the same poem higher, and judged it worth more money, when told it took 18 hours to write instead of 4. AI collapses visible effort to seconds. Unless you show where the human work went, people assume there wasn't any.

Authenticity needs proof. Researchers Kent Grayson and Radan Martinec split authenticity into two kinds: indexical, a real, traceable link to the thing itself, and iconic, a convincing resemblance to it. AI is brilliant at iconic and has nothing indexical to offer. Willy's Chocolate Experience was pure iconic authenticity with zero indexical proof behind it.

Peers deliver the verdict. A brand tells its story, but customers verify it with each other in comments, groups and reposts. Wizards of the Coast learned that its own fans were the most effective AI detectors in the world. You don't control that second channel. You can only give it evidence that holds up.

Boulder's rules for using AI without losing trust

We use AI. Pretending otherwise would break rule two. Here's how we keep it from costing our clients' credibility.

1. Never let AI make a promise reality can't keep. Your product, your space, your team and your food are shown as they really are. AI can build the world around them, never replace them.
2. Disclose before you're discovered. One plain line beats a three-day crisis. Say what AI did and what people did.
3. AI is the apprentice, not the master. It's good for exploring, moodboards, variations and clean-up. The decisions, the final craft and the sign-off stay human. That's exactly the setup the research says people accept.
4. Know your category. If customers pay for artistry (fashion, crafts, wine, illustration, anything giftable), the bar is higher. If your product is purely functional, people are more relaxed.
5. Audit your suppliers. Wizards of the Coast blamed a vendor. Your audience won't care whose fault it was. Put an AI clause in every brief and contract.
6. Make the human layer visible. Sketches, rejected options, process shots. When people can see the effort, AI stops looking like a shortcut. (More on this in our next post on showing the work.)

Before you post your next AI visual

• Does this image show anything the customer will expect to find in real life?
• If someone asked "is this AI?", would we be comfortable answering in public?
• Is there a short, plain disclosure line ready?
• Can we point to the human decisions in this piece?
• Does our audience buy our product for its craft or artistry?
• Did every vendor on this job confirm how they used AI?

AI won't kill your brand. Hiding it, overpromising with it, or using it where people expect a human hand will. Trust is the one asset you can't generate.

Not sure where your brand stands? poke us and we'll look at it together.

Written by Bukhra, with AI assisting the drafting.