A supplier sends a spotless production-floor image. Your sourcing team wants to know whether the facility belongs to that supplier and whether the equipment supports the product under discussion. An AI detector is an appealing shortcut, but a clean scan answers neither question.

On October 7, Google announced global access to the English-language SynthID Detector. It checks images, video and audio for watermarks within its supported ecosystem. That is useful context about some generated media. It is not a universal test of whether a photograph was taken at a particular factory.

Ask what the image is supposed to establish

Start with the procurement claim, not the detection score. A photo beside a statement about machining capacity is being asked to support a different proposition from a photo illustrating an office address. The first needs a connection to equipment, processes and products. The second needs a connection to location and occupancy. Neither proposition becomes true simply because no recognized watermark was found.

There are several separate questions: Was the image captured rather than generated? Where was it captured? Who operates the site? When was it taken? What work does the visible equipment actually perform? A genuine photo of a partner's facility can still be misleading if presented as an owned factory. An authentic but old image may not describe the present production arrangement.

Buyers do not need to resolve every question for every decorative photograph. They do need to resolve the questions attached to a purchasing decision. If a particular process determines whether a supplier belongs on the shortlist, ask for material showing that process and its relationship to the relevant product. A panoramic workshop image may be attractive, yet leave that relationship unexplained.

Requesting the original file can help establish a starting point, but the original alone is not conclusive. File information may be incomplete, and a photograph can be genuine while its caption is wrong. Treat the file, the explanation and other supplier evidence as separate pieces that should agree, rather than making any one piece a substitute for the rest.

Suppliers can make the question easier to answer

For Chinese manufacturers presenting themselves to international buyers, a short, precise caption is often more useful than a broad claim about advanced facilities. Identify the process shown, the product family it serves, the approximate capture date and whether the site is owned, rented or operated by another party. Publish only the details the business can substantiate.

Editing also needs a clear boundary. Cropping a photograph to emphasize a workstation does not have the same meaning as replacing the workstation with a generated machine. Removing a visual distraction can still affect interpretation if that distraction was relevant to the claim. The decisive question is whether the edited image represents a production condition that actually existed.

Concept illustrations remain useful. They can explain a component or show a proposed layout without pretending to be site evidence. Label their purpose next to the image. Do not place an illustrative layout under a heading that implies it documents installed equipment. Buyers should not have to infer the difference from the visual style.

A longer video is not automatically stronger proof. Continuous footage may show spatial relationships that separate photos miss, but an edited demonstration can omit the setup that makes a test meaningful. If the footage supports a performance claim, the product identity and test conditions matter more than cinematic polish. The business should be able to explain what happened outside the visible frame when that affects the result.

When an image triggers concern, an AI scan can suggest a useful next question. It cannot clear the supplier's capability claim. Avoid assembling a set of detector scores and selecting the most favorable result; that turns a limited diagnostic into an apparent certificate it was never designed to provide.

Our earlier discussion of a content provenance register concerns how a team records where material came from. Procurement requires an additional judgment: whether that material supports the particular statement being made. Known provenance and sufficient proof are related, but they are not interchangeable.

Review the few images that carry the most weight in a supplier presentation. Write down the exact claim each is meant to support, then remove claims the image cannot establish. Keep attractive decoration if it helps the page, but do not let decoration do the work of manufacturing evidence. That gives both the supplier and the buyer a more useful conversation than arguing about whether an algorithm thinks a picture looks artificial.

Sources