Your AI images may already be labelled. Here's how to stay ahead of it, and when not to use AI at all.

So the sensible default is simple: be upfront about AI images, and never use them as proof. Get that wrong and it isn't a fine you risk, it's the lead. Below is the three-way test we use: mood, stand-in or proof.

Why might your AI image already be labelled?

The big image tools now mark what they make, usually in one of two ways:

• Invisible watermarks. Google DeepMind's SynthID hides a signal in the pixels themselves. In DeepMind's own explainer, the team describe training it against cropping, resizing, rotation and added noise, so the mark is built to survive the edits a designer would normally make.
• Metadata. Many tools write a provenance tag into the file. Google Merchant Center's image rules say "All images created using generative AI must contain meta data indicating that the image was AI-generated", and tell sellers: "Don't remove embedded metadata tags such as the IPTC DigitalSourceType property."

Platforms read these signals. Meta says it labels images on Facebook and Instagram when it detects "industry standard AI image indicators", and since July 2024 that label has read "AI info".

Saying nothing doesn't keep it quiet. It just means the platform tells your customer before you do, which looks far worse than having said it yourself.

What do UK advertising rules actually say?

There's no AI rulebook. The ASA put it plainly in its guidance on AI disclosure: "There is no blanket legal requirement in the UK to disclose the use of AI in ads." The test is whether your audience is likely to be misled if you don't.

Two points matter more than any label:

• A label won't rescue a misleading ad. The ASA's example is an AI image of a cosmetic product's effect that doesn't reflect real-world results. Its verdict: "disclosure alone is very unlikely to mitigate the harm caused by a fundamentally misleading message."
• You own the output. From the ASA's June 2026 guidance: "If an AI tool produces something problematic, the advertiser remains responsible for any output used in their ads."

Selling into the EU? The AI Act's transparency rules, including disclosing deepfakes, have applied since 2 August 2026.

How do you use AI images without losing trust?

We sort every AI image into one of three groups:

1. Mood and concept. An abstract hero image, a seasonal backdrop. Low risk. Keep the metadata intact; a loud label usually isn't needed.
2. Stand-ins for real things. Your product, premises, team or vans. If a customer would reasonably assume it's real, use a real photo or say clearly it's an illustration.
3. Proof. Before-and-after shots, results, reviews, case studies. Never AI. An AI "after" photo showing a finish your team didn't produce is the cosmetic example above in a hi-vis jacket.

Two habits keep you honest without slowing you down:

• Check your export, compression and background-removal tools aren't stripping the metadata.
• Keep a simple record of which assets are AI-generated, so nobody has to guess six months later.

If you're producing AI creative for Meta or Google ads, build these checks in from the first brief. It's part of how we run paid social for our clients.

Why does being upfront pay?

Trust is what your ad is really selling. A customer who feels tricked by one image doesn't just skip that ad; they start doubting the price, the reviews and the promise sitting next to it.

Writing "illustration" under an image costs you nothing. Getting caught can cost you the lead.

Use AI for the mood. Use the real thing for the proof.

Not sure where your AI creative crosses the line? Get in touch, and we'll take a look.

Sources

• Google DeepMind, "From deepfakes to DNA: the science of watermarking AI" (YouTube, 1 Oct 2026) — https://www.youtube.com/watch?v=HIUzrxQxTtw
• Google Merchant Center Help, image link [image_link] requirements — https://support.google.com/merchants/answer/6324350
• Meta, "Our Approach to Labeling AI-Generated Content and Manipulated Media" (5 Apr 2024, updated 1 Jul and 12 Sep 2024) — https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/
• ASA, "Disclosure of AI in Advertising: Striking the Balance Between Creativity and Responsibility" (29 May 2025) — https://www.asa.org.uk/news/disclosure-of-ai-in-advertising-striking-the-balance-between-creativity-and-responsibility.html
• ASA / CAP News, "AI and Deepfakes: Four Things Advertisers Need to Know Before They Hit 'Run'" (11 Jun 2026) — https://www.asa.org.uk/news/ai-and-deepfakes-four-things-advertisers-need-to-know-before-they-hit-run.html
• European Commission, Code of Practice on Transparency of AI-generated Content — https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content