Real photography or AI imagery? A rule for every image on a small-business website

2.1The rule: if a customer would rely on it, it must be real

Every image on a website makes an implicit claim. Some claims are about mood: this business feels calm, bold, warm, precise. Others are claims of fact: this is our shop, this is our team, this is a deck we built, this is what the product looks like when it arrives.

Mood images can be generated. Fact images cannot.

That sorts almost every image cleanly.

Real photography only:

• The premises, inside and out
• The people who work there
• The product, as sold
• Finished work: before-and-after, portfolios, case studies
• Anything that illustrates a result a customer might expect

Generated imagery can be fine:

• Abstract backgrounds, textures and patterns
• Illustrations and diagrams that explain a process
• Conceptual images for a blog post or a section divider
• Seasonal or campaign artwork that is obviously designed rather than documentary
• Placeholders during design, replaced before launch

A useful test: imagine a customer arriving at your shop, receiving your product or seeing your finished work, and then comparing it with the image on your site. If they would feel misled by the difference, the image needed to be real.

2.2Why fake “fact images” backfire

A generated photograph of a renovated kitchen is not neutral decoration. To a visitor, it reads as your work. If they later discover it is not, the damage is not limited to that image. They reasonably start wondering what else on the site was invented: the reviews, the years in business, the qualifications.

Generated people are the riskiest case. A “team” section populated with generated faces is a claim that those people exist and work for you. It also tends to look uncanny in subtle ways that viewers notice even when they cannot explain what is wrong.

Real photography, even imperfect phone photography, carries a kind of evidence that generated images cannot. Slightly uneven lighting, a real street outside the window and the actual tiles on the actual floor are what make it believable.

2.3What the platforms now expect

This is not only a matter of taste. The infrastructure around images is starting to expect disclosure.

Google’s guidance on generative content encourages site owners to consider “adding information on how your content was created” where it makes sense for the audience, including image metadata. For online stores it is stricter: its Merchant Center requirements say “AI-generated images must contain metadata using the IPTC DigitalSourceType TrainedAlgorithmicMedia metadata.”

Separately, the Coalition for Content Provenance and Authenticity (C2PA) develops open technical standards for “establishing content provenance and authenticity”, so that an image can carry a verifiable record of how it was made. The aim is that, over time, it becomes easier for anyone to check whether an image was captured or generated. It is sensible to assume your images will eventually be checked.

The practical implication: do not strip provenance metadata from generated images, and if you sell products online, never use a generated image as a product photo.

2.4Making real photography achievable

The honest objection is cost. Professional photography is a real expense for a new business. Some ways around it:

• Start with a phone and good light. Shoot near a large window, in the morning or late afternoon, with the flash off. Clean the lens. Hold the phone level. That alone beats most stock photography.
• Shoot a small set well. You need fewer images than you think: one strong exterior, two or three interiors or workspace shots, a handful of product or work photos, and a portrait of the owner. Ten good images can carry a small site.
• Build a shot list before you shoot. Write down every image the site needs and its shape (wide for a banner, square for a grid, tall for a phone hero).
• Document work as you go. Every finished job is a chance to photograph it. A service business that photographs every project for a year has a portfolio without a single dedicated shoot.
• Book a professional for the few images that matter most, typically the hero and the portrait, and do the rest yourself.

2.5Where AI helps real photography

There is a middle ground worth using: AI tools that edit your real photos rather than inventing new ones. Straightening, colour correction, removing a stray cable, extending a background to fit a wide banner or cleaning up a cluttered counter are edits that improve a real photograph without changing what it documents.

The line is the same as before. Editing presentation is fine. Changing the facts, such as adding a feature the product does not have, a view the shop does not have or a result the work did not achieve, is not.

2.6Describe every image properly

Whatever you use, give every meaningful image accurate alt text. Google calls alt text “the most important attribute when it comes to providing more metadata for an image”, and it is how people using screen readers learn what the picture shows. Accurate alt text for a generated illustration describes the illustration; it does not pretend to describe a photograph.

2.7The short version

• Generate mood. Photograph facts.
• Never generate people, premises, products or results.
• Keep provenance metadata intact.
• Ten real images beat fifty perfect fakes.

I founded UOVA, an AI website builder and creative studio in Toronto, so I am not against generated imagery. I think it is most valuable when it is used where it is honest. More notes on building websites at saltalash.vercel.app.