Real UGC and AI ads: a Shopify creative testing guide
Separate customer proof from generated creative, check Higgsfield’s workflow limits and design an ad test that measures more than production speed.
A customer describing a product they used, a paid creator demonstrating it and a generated spokesperson are three different advertising inputs. A Shopify merchant can test all three formats, but should keep the source of the evidence clear. A realistic face does not establish a real customer experience.
This is a practical testing framework, not a merchant performance case study. We have no controlled campaign result showing that AI-generated creative beats real customer footage, or the reverse.
Start with the claim, then choose the format
Shopify’s UGC ads guide describes testimonials, product demonstrations, unboxings and lifestyle content. Use the format to answer a particular buying question: what fits inside the bag, how the fastening works, or what a customer noticed after using the product.
Keep three separate folders in your creative library:
- Customer evidence: the original review or recording, the person’s actual words and the agreed uses.
- Commissioned creator work: the brief, the product supplied, the creator’s recording and the approved advertising versions.
- Generated creative: the input assets, generated output and a record of which people, scenes or actions are synthetic.
Our recommendation is to approve the factual claim before producing variations. A generated scene can illustrate an idea; it cannot substantiate a claim that someone bought, tried or recommended the product. Keep a real demonstration available wherever appearance, fit or physical performance matters.
What the Higgsfield example establishes
In an October post, Adil Alimzhanov wrote:
people trust people, not ads
The post is labelled “Made with AI”. That makes it an illustration of synthetic creative, not customer testimony. It quotes a Higgsfield promotional post from 1 October describing ad workflows through ChatGPT. Those are vendor claims about production, not evidence of campaign returns. The original video players were unavailable during our check; the links provide the source context.
Separately, Higgsfield’s current Marketing Studio help page, checked on 23 October, describes template-based creator-style video in faceless, talking-head and silent formats. It says the template-based Studio is not yet available through Supercomputer, MCP or other agentic channels. Do not assume the social demonstration means every Studio template can be run through an agent.
The same help page lists a 15-second maximum per generation and warns that voiceover and pacing can vary. Its prediction score is an estimate, not a performance guarantee. Confirm the workflow you can actually access before promising a client a production schedule.
Run a test that answers one useful question
Here is our suggested starting design. Choose one product, one offer and one destination page. Compare a rights-cleared customer or creator cut with a generated explanatory cut addressing the same buying question. Keep the offer and core claim consistent. If the hook, audience and discount all change too, the result will not isolate the creative choice.
Write down the decision before launch: for example, whether the generated version is worth another production cycle at your acceptable acquisition cost. Set a spend ceiling and review period using your normal campaign economics. Use the advertising platform’s experiment tools where available; ordinary delivery can allocate impressions unevenly, so a dashboard comparison alone does not establish causation.
Record spend, purchases, revenue and production costs, including staff review time and rejected generations. Keep the attribution window consistent. A low-cost clip is not necessarily a low-cost acquisition, and a few purchases are not enough to establish a durable winner. Review returns and cancellations when they become available.
Approve the export, not just the preview
Check the actual exported file for product shape, packaging, spoken claims, captions, offer dates and destination URL. Confirm the intended uses with everyone whose face, voice or footage appears. Keep the source and any paid relationship clear to viewers, and review the relevant advertising platform’s requirements before launch.
For Performance Max, Google recommends uploading at least one video yourself even though its system can generate videos from supplied assets. Its guidance covers horizontal, square and vertical formats and 10–60-second videos. Plan the export for the placement rather than assuming a single cut fits every channel.
The useful outcome is a repeatable approval and measurement process. Our Moby video rollout report covers a separate workflow starting with a recording; this framework helps evaluate the finished advertisement whichever production tool you choose.
More from Marketing
See all →

