AI generated images on websites: when they read as fake, and where they can work
AI generated images on websites are quick and cheap, and visitors are getting good at spotting them. They read as fake when they stand in for real people, places or products, and they can mislead when they do. They work best where nobody expects a photograph to be evidence: textures, abstract backgrounds, concept art for an article, or early mock-ups.
Why they often look fake
- Too perfect. Even lighting from nowhere, flawless skin, spotless surfaces and a glossy finish no real camera would produce in that setting.
- Wrong details. Hands, teeth, jewellery, text on signs and packaging, and the joins between objects are where errors show, even as models improve.
- Generic composition. The model's idea of “a cosy café” is an average of many cafés, so it looks like nowhere in particular.
- Mismatched sets. Each image is generated separately, so lighting, colour and lens character drift between them.
- The house style. Many images share a recognisable look: soft cinematic light, warm-and-teal colour, shallow depth of field on everything.
Some of these tells will fade as models improve. The underlying problem won't: an image of a business that doesn't show the business.
Where they can work
| Use | Risk | Why |
|---|---|---|
| Textures and abstract backgrounds | Low | Nobody expects them to depict anything real |
| Concept images for blog posts and articles | Low to moderate | Readers expect illustration; avoid realistic people and events |
| Mock-ups during design | Low, if replaced | Useful placeholders while real photos are taken |
| Atmosphere: weather, landscapes, materials | Moderate | Fine if generic; misleading if implied to be your location |
| Products you sell | High | Customers rely on the image to know what they'll receive |
| Team, customers, premises, past work | High | Presents people and places that don't exist as if they do |
Why real photos still matter
A visitor deciding whether to book a physiotherapist, order a cake or hire a builder is looking for evidence. Real photos, even plain ones taken on a phone, show the actual room, the actual cake, the actual finish on a kitchen. That evidence is the point. A synthetic image can be beautiful and still answer the visitor's question falsely. Once a visitor notices one generated image of the team, they start doubting the reviews and the prices too.
Advertising and consumer protection rules in many countries prohibit misleading presentations of products and services, and a generated image of a product that differs from what's delivered can fall foul of them. Stock photos on websites discusses the same line for stock photography.
Rights and labelling
Ownership of AI-generated images is unsettled and varies by country. The US Copyright Office, for example, has taken the position that material generated without sufficient human authorship isn't protected by copyright, which matters if you want exclusive use of an image. Check the terms of the tool you used for commercial rights, and avoid prompts naming living artists or real people. Labelling images as AI-generated is good practice where a visitor might otherwise take them as real, and some platforms and jurisdictions are introducing rules about it.
If you do use them
- Decide the role first: decoration or concept, never evidence.
- Generate a whole set in one session with one style description, then reject outliers, so lighting and colour match.
- Inspect at full size for errors in hands, text and edges.
- Write honest alt text that describes what the image shows, not what it pretends to be.
- Replace mock-up images with real photos before launch.
Accessibility
Generated images need the same care as any other: a text alternative for meaningful images under WCAG 2.2 success criterion 1.1.1, no important text baked into the picture, and contrast checked wherever text sits on top. Generated images often contain garbled pseudo-text in signs or labels, which is meaningless to everyone and shouldn't be described as if it were real.
Images in Anticode
Anticode doesn't generate images. It places the photos you add to a build: each is checked by its first bytes and stored under your account, and the model sees them, so it can put each in a sensible section. Design-engine pages can also draw on a curated library of real photographs from Unsplash, each with written alt text, for atmosphere while you gather your own. Sentinel checks every image has a text alternative before the page can be published.
Questions people ask
Should I use AI images on my business website?
For decoration and concepts, they can be fine. For your team, premises, products and work, use real photos; visitors treat those as evidence.
Can I copyright AI-generated images?
It depends on the country and how much human authorship is involved. In the US, purely generated material isn't protected by copyright. Check the tool's terms for commercial use.
Do AI images hurt SEO?
Being generated isn't in itself a known ranking problem. The bigger risks are lost trust and images that mislead about what you offer.