Should we use AI-generated or AI-edited product photos on product pages and in catalog ads, and should we label them?
Our answer
Use AI around the product, never on the product itself, so backgrounds and lifestyle scenes are fine but the product stays as photographed. Keep the AI metadata on generated images for Google. Labels on ads follow Google AI asset labels and Labeling AI ads and AI voiceover.
- Confidence
- Medium
- Strongest proof
- 8 official rules
- 3 datasets, 1 practitioner, 2 opinions or claims
- Sources
- 9 sources
- 14 statements
- GM
IIOCBIAAP
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentThe main image must show the exact item sold, in its exact color, pattern, finish and material (Google). Google still allows AI-generated images if their AI metadata stays in place, and its own Product Studio removes backgrounds and generates scenes.
Shoppers say they trust AI product images less. Of 2,273 US shoppers Baymard Institute (ecommerce UX research firm) surveyed, 60% said knowing an image was AI-generated would lower their confidence. And 59% of Australians doubted AI-advertised products would match their description (Auros.ai, AI data firm).
The image is judged against the product. The UK's ad regulator ruled an ad misleading when footage of seemingly real or AI-generated puppies stood in for a robot toy (Osborne Clarke, law firm). Meta's catalog rules ask for images that accurately show the product.
Our judgment
We drew the line at the product itself by combining four sources: Google's exact-item rule, Baymard Institute's advice to keep authentic photography for core product views, the IAB's split between prompt-made images and routine post-production, and the UK ruling against imagery that stood in for the real product. Google's guidance only suggests saying how content was made where that makes sense. So on your own product pages we put the effort into keeping AI off the product rather than into a page label.
Our judgment, built from Google, Baymard Institute, IAB and Osborne Clarke: our reading of these sources, not a finding from any one of them.
What the sources say
14 statements from 9 sourcesGoogle's Merchant Center rules say the main image must clearly show the exact item sold, variant images must match the exact option, and AI-generated images must carry metadata saying so.
“All images created using generative AI must contain meta data indicating that the image was AI-generated”
Google requires AI-generated product images to keep IPTC metadata marking them as AI-made, and AI-written titles and descriptions to go in structured attributes flagged as trained_algorithmic_media.
“Don't remove embedded metadata tags such as the IPTC DigitalSourceType property from images created using generative AI tools, for example Product Studio.”
Google Merchant Center lets merchants label a product image or video as made or edited with AI, which adds a disclosure to the ad's "How this ad was made" panel. A detected label stops automatic image enhancements.
“If an AI label is detected, automatic image enhancements like overlay cleanup from Merchant Center Automations won’t run.”
An Auros.ai survey of 4,000 adults reports 59% of Australians do not trust AI-advertised products to match their description, and 44% are less likely to buy if they know AI was used. Stated attitudes.
“Close to half (44%) say they are now less likely to purchase a product if they know AI was used in its advertising.”
Baymard's June 2026 survey of 2,273 US online shoppers found 60% say knowing a product image is AI-generated lowers their purchase confidence, and 17% would not buy at all. Stated attitudes.
“Retailers should therefore exercise caution when introducing AI-generated imagery, prioritizing authentic photography for core product views.”
Meta's catalog image rules: at least one JPEG or PNG per product (8 MB max), 4 or more for carousels, a white backdrop, no text over the product, and one aspect ratio, 1:1 or 4:5.
“Don't include text in your images that overlays the product, calls to action, promo codes, watermarks or time-sensitive information like temporary price drops”
Google says generative AI can help with research and structure, but generating many pages without adding user value may break its scaled content abuse policy. Models predict words, not retrieve facts.
“using generative AI tools or other similar tools to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse”
The IAB's voluntary 2026 framework asks for labels on AI-generated images and video and synthetic people that could mislead, not on routine retouching or ad copy, and cites a 31.5% click-through drop from labeling.
“regardless of subsequent human refinement, editing, or compositing”
Common Thread Collective says standard edits like cropping or brightness do not trigger Meta's AI label, and outside detection relies on C2PA metadata. No data, and Meta has published none on performance.
“Meta has not published data suggesting the AI info label directly reduces ad performance.”
Meta automatically labels ads made or significantly edited with its own generative AI tools, and third-party AI content when it detects C2PA metadata; the label sits in the About this ad menu.
“automatically label ad content that has been created or edited using some of Meta's generative AI tools”
A department store association argues AI visuals pay off through consistent image sets across a catalogue and accurate images that can cut returns, and advises A/B testing image formats. No data.
“Accurate visuals can also reduce returns, especially in clothing and furniture, where many returns stem from mismatches between online images and delivered products.”
Meta says its About this ad will include AI info labels on ads made or heavily edited with its generative tools, and it will start detecting third-party AI tools. Photorealistic AI humans get a label next to Sponsored.
“the label will appear next to the Sponsored label (not behind the three-dot menu).”
Photoroom's UK survey reports 31% of consumers say visible AI-generated marketing makes them trust brands less and 63% say inconsistent product imagery makes a seller look unreliable. No sample size.
“31% said visible AI-generated marketing makes them trust brands less”
The UK ad regulator upheld complaints against a robot-dog toy ad that showed real or AI-generated puppies instead of the toy, finding it exaggerated the product and was misleading.
“Caution should be exercised when using AI-generated or other imagery to represent a product rather than the actual product itself”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer