Our imagery

No synthetic photographs

We do not publish images that look like photographs but are not. A photorealistic image of an orchard, a facility or a shipment would be a claim that something was seen, and we would have no way to support it.

This is not a preference we apply case by case. There is no photographic register in our system to select, so there is nothing to switch on.

The four kinds of image

Charts are drawn from figures in our Index. Every figure carries its source, its basis and the date it was retrieved, and the chart shows them.

Diagrams are drawn by us from a fixed set of templates — a process flow, a supply chain, an orchard layout, a season calendar, a basis ladder. Any text in them is text a person wrote.

Botanical plates are generated illustrations in the manner of a nineteenth-century scientific flora: a single specimen on empty paper. They are drawings of a kind of thing, never of a particular one.

Editorial illustrations are generated flat prints, used for stories about policy, trade and markets where there is nothing specific to draw. They are conceptual by design.

The first two are built. The last two are generated. The standing masthead disclosure covers our automated newsroom; image provenance remains in metadata.

What a generated image may never contain

A generated image on this site may never contain:

• text, lettering, numerals, signage, logos, brand marks or flags
• an identifiable person
• a real, named or recognisable place, facility, port or landmark
• a depiction of the specific event an article reports
• anything imitating a record — a drawn chart, map, document, screenshot or satellite view
• any business in which our owner has an interest, whether named or implied

The last four are enforced before an image is requested rather than checked afterwards. Subjects come from a fixed vocabulary of generic descriptions — no place, company, person or event appears in any of them — and the short qualifier an editor may add is rejected if it contains a capitalised name or a number, which is what a country, a cultivar, a year or a tonnage looks like. The same owner-interest check that governs our articles runs over the image request too.

Text and photorealism are constrained in the image request. Every candidate is then checked against a fixed rubric that also rejects misleading botanical or mechanical anatomy. We describe that as a check rather than a guarantee because it is one. If a generated image is not good enough we fall back to a chart or a diagram, and if none of them fits we publish the article with no image at all.

No image is evidence

An image on Mac Daily never supports a claim. It does not anchor a figure, it is not a source, and it never appears as part of a conflict between sources. If something is true here, the words and the citation say so — never the picture.

One picture, one story

A generated illustration is used once. Every one we publish is fingerprinted, and the fingerprint is checked against everything already published before it can be placed again — so the same picture cannot appear on two stories, and a near-copy of it counts as the same picture. Where nothing distinct is available, the story runs its headline figure set in type instead. A second-best picture would be worse than none.

Some stories get no picture at all, on purpose. A tariff, a forecast or a trade ban has no appearance, and anything we put beside one would be decoration standing in for a scene that never happened. Those stories are illustrated by a chart drawn from their own figures — or by nothing.

How to check for yourself

Every image we publish carries its origin, its kind, and — for generated images — the model that made it and the exact prompt it was made from. That record travels in the file itself, in the standard IPTC field, so it survives being downloaded or resized: our generated images are marked trainedAlgorithmicMedia, and our charts and diagrams compositeSynthetic. No image we publish is ever marked as a digital capture, because none of them is one.

The same record is available through our public API, and where a provider supplies content credentials we preserve them. We never create them ourselves.

Related

Our use of automation generally, and the labelling required by the EU AI Act, is on the AI disclosure page.