AI-generated photos on a company website: what the buyer actually sees

On a manufacturer's website, an AI-generated photo of the plant, the team or the product does more damage than a stock image. A stock image shows a real plant that is not yours, and everyone assumes as much. A generated image shows a plant that exists nowhere, yet presents itself as yours. A buyer who has walked real shop floors spots the gap, and from that moment doubts your real photos too. There are a few narrow places where a generated image does no harm. This page names them, after explaining why everywhere else is expensive.

Not a better stock photo, a worse one

The generated image is often pitched as the natural next step after stock: cheaper, no licence to track, and above all customisable. The customisation is exactly the problem.

A stock photo is generic, and it looks generic. Visitors read it as an illustration. A generated image can put your logo on the building, your colours on the conveyors, a part that looks like yours on the assembly bench. It no longer presents itself as an illustration. It presents itself as evidence. And it is evidence of something that never happened.

The page on stock images vs real photos covers the choice between a bought image and a commissioned one. The question here is narrower: what happens when an image looks like yours without being yours.

What a buyer who knows a shop floor notices

A prime contractor's purchasing lead, a process engineer or a quality manager does not look at a plant photo the way an ordinary visitor does. They read it. And a generated image fails on exactly the details they read.

• Machine guards missing, or placed where no standard would put them.
• Personal protective equipment worn wrong, or not at all, in an area where it is mandatory.
• Gauges, labels and safety signs covered in illegible characters.
• A machine whose shape matches no real model, or a weld bead too regular for any person to have laid it.
• Spotless, empty aisles in a plant that is supposed to run three shifts.

Take a machining subcontractor in the Beauce that wants onto the short list of a heavy equipment manufacturer. Its home page shows a five-axis machining centre bathed in light, generated in a few minutes. The buyer is interested, asks for a qualification visit, and the machine is not on the floor. The subcontractor never lied about its real capacity. It simply let an image lie on its behalf. The conversation that follows is no longer about tolerances.

The cost does not stop at the offending image. Once a buyer has recognised one generated image on your site, every other photo becomes suspect, including the ones taken in your real plant. You have spent credibility that your real photos had earned.

The same goes for the team. A distributor in Montérégie that puts generated faces on its "Our team" page is introducing customers to people they will never meet. The day the real sales rep walks in, the customer compares.

The risk beyond credibility

Three points are worth knowing before you publish. None of this is legal advice.

The first is advertising. The Competition Bureau states that under the Competition Act it is against the law to advertise or market something in a false or misleading way. Among its advertising dos and don'ts, it says not to use photos or illustrations that differ from the product being sold. It also explains that courts consider the general impression a representation conveys, not only its literal meaning. No one actually needs to be deceived for a practice to be found misleading. The Bureau pages consulted do not mention AI. Applying them is our own reasoning: a generated image of a product, a plant or a capability you do not have runs into these general rules.

The second is ownership. According to the federal government's consultation paper on copyright and generative AI, case law ties authorship to a natural person who exercises skill and judgment. The paper finds that criterion far less likely to be met for works produced by generative AI systems from short instructions. It adds that who would be the author and first owner remains an open question. It is a consultation paper, not a statute or a ruling. The takeaway is simple: the hero image on your site may not belong to you as clearly as a photo commissioned under contract.

The third is the trail. LinkedIn shows an icon on images and videos signed with C2PA Content Credentials. Clicking it can show whether AI was used to generate or edit the content, and with which tool. LinkedIn also says it is not yet possible to identify and label all AI-generated content. So this is not a reliable detector. It is a sign that an image's provenance is becoming visible, sometimes on the very platform where your buyers are looking at you.

The narrow places where a generated image is harmless

It would be dishonest to say a generated image is never useful. It is harmless wherever nobody could take it for evidence.

• An illustration of an abstract concept at the top of a blog post: traceability, compliance, supply chain.
• A texture, a pattern or a background that acts as a surface and shows nothing real.
• An internal mock-up used to prepare a shoot, the way you would use a sketch, that never leaves the company.
• A concept of a future expansion, clearly presented as an illustration, the way an architect's rendering would be.

The test fits in one question. Could a buyer take this image as proof that something exists at your company? If the answer is yes, it has to be a real photograph. The plant, the people, the equipment, the parts and the installations at customer sites all land on the wrong side of that test.

What you can demand today from whoever builds or feeds your site: a written list of every generated image on it, and a commitment that none of them shows the plant, the team or the product. The answer that should end the conversation is "nobody will notice the difference". Your buyer will.

That is the position printed on this site: "Certified Human Content", "NOT PROMPTS, NOT SLOP. CRAFT." AI generates images. It does not generate a prime contractor's trust. For anything that has to serve as evidence, the answer is a planned photo day, as the page on corporate photography explains, and more broadly the video production and content page. In the Signal Program, executive portraits are shot on Hasselblad medium format at 100 MP, retouched and delivered in 48 hours.

When this does not apply

A service firm with no physical site to show, such as an engineering consultancy or a software publisher, has less to lose. Its buyers are not assessing a shop floor. A generated illustration of a concept stays an illustration there, as long as it does not stage fake customers or fake employees.

The same holds for a company that does not have a plant yet. A young company in Saguenay presenting its future building can show a concept, provided it is labelled as one. The problem is never the generated image itself. It is the generated image passing itself off as a photograph.

Finally, if your site has no photos at all, do not fill the gap with a generated plant. One honest photo of the building, taken on a weekday morning, is worth more. An empty space leaves a question open. A false image answers it badly.

Sources

• Competition Bureau Canada, "Deceptive marketing practices", https://competition-bureau.canada.ca/en/deceptive-marketing-practices, fetched 30 September 2026.
• Competition Bureau Canada, "The general impression test", https://competition-bureau.canada.ca/en/deceptive-marketing-practices/general-impression-test, fetched 30 September 2026.
• Competition Bureau Canada, "Advertising dos and don'ts", https://competition-bureau.canada.ca/en/deceptive-marketing-practices/types-deceptive-marketing-practices/advertising-dos-and-donts, fetched 30 September 2026.
• Innovation, Science and Economic Development Canada, "Consultation Paper: Copyright in the Age of Generative AI" (date modified 2024-12-05), https://ised-isde.canada.ca/site/strategic-policy-sector/en/marketplace-framework-policy/consultation-paper-consultation-copyright-age-generative-artificial-intelligence, fetched 30 September 2026.
• LinkedIn, "Content credentials | LinkedIn Help", https://www.linkedin.com/help/linkedin/answer/a6282984, fetched 30 September 2026.