A creator delivered a review video for your product. The talking points are perfect, the lighting is beautiful, and something still feels… off. You cannot say exactly what. Before you approve the work and escrow releases, before you drop the file into a paid ad set, you need to know whether a real person filmed this or a tool generated it.
This is not an anti-AI article. A growing number of legitimate creators use AI to write hooks, clean up audio, or fix a background — the honest trade-offs are covered in the human versus AI UGC breakdown. The problem is not AI. The problem is an undisclosed synthetic person presented as a genuine customer, because you are buying believability and you deserve to know what you got for it. This is a practical checklist for the review step.
Why a checklist, not suspicion
You review UGC inside a 7-day window on ugc.ng. Approve, request revisions, or raise a dispute. That window is your only chance to verify before the deal closes and the creator is paid the next day. A checklist keeps that decision fast and fair — it separates "this looks generated to me" from "here are the specific things I need to check."
Fair also means proportionate. AI-assisted editing — a cleaned voiceover, a swapped background, AI-written captions — is a normal production choice and not a fraud. The line you care about is different: is the person in the video real, or synthetic? Is a claim being made by a face that does not exist? That is where the risk lives, and it is where the checklist focuses.
The micro-details: hands, teeth, and hair
Current generation tools have gotten dramatically better at people, but they still struggle most on the small things humans notice without knowing why. When you watch a review video at full screen, look at the details, not the storytelling.
Hands. Count the fingers. Watch hands move — through hair, around a product, in a gesticulation. Look at nails, knuckles, and how a thumb wraps a product. Hands remain the weak spot: wrong finger counts, fused fingers, a thumb that bends like a second wrist. If a hand holds the product at an odd angle that no real person would naturally pick, file a mental flag.
Teeth. Look at the mouth during speech. Synthetic mouths often render teeth as a single blurred block, with no separation, or a strange glassy sheen. Watch the inner lip move against the teeth — that collision is where generation slips.
Hair. Individual strands, flyaways, and how hair sits against the neck and shoulders. Generated hair tends to render as smooth, glossy masses with suspiciously consistent texture, and it often does not respond correctly to a slight head turn.
None of these is a slam dunk on its own — a bad phone camera can blur teeth too. But two or three of these tells together, on a video that otherwise looks too clean, is worth a closer look.
Background physics
Faces are getting hard to fake well. Backgrounds are still the easiest place to catch a generated video, because the physics there is unforgivable.
- Text. Watch every label, sign, and product box in the frame. Generated text warps — letters bend, spacing shifts between frames, a glyph randomly changes shape mid-clip. In a real kitchen or desk shot, text stays put.
- Reflections. Check mirrors, glass, windows, glossy tabletops. A reflection that does not match the motion of the person in front of it — or a reflection with slightly different features — is a classic miss. On a phone screen, look at the person's own screen: can you see the app, or a smeared rectangle of light?
- Lighting across cuts. A real clip has one light source direction for the whole take. If the shadow under the chin jumps or the light on the background shifts between angles with no visible lamp change, the scene was composited differently.
- Physics of objects. Products do not behave mid-air. A bottle that hovers for a frame, a chair edge that wobbles, a background that ripples when the person leans toward it and should displace nothing — these are render artifacts leaking through.
You want background realism under scrutiny, not just at a glance. Pause the video at a random frame and look at the edges: where the person's shoulder meets the wall, where hair meets the backdrop. Clean, hard edges everywhere, with zero motion blur on things that should move — that is as suspicious as messy edges, because real phone footage has imperfection.
Audio artifacts
Listen with the same attention you gave the visuals. Voices are generated well now, but the artifacts are predictable.
- No room tone. Real phone recording has background — the hum of a generator, traffic, a fridge, the echo of a room. Totally dead-silent audio in a claimed "I filmed this in my kitchen" video is a flag.
- Too-perfect pacing. Generated speech lands each word at a suspiciously even rhythm. Humans pause where they think, stumble, breathe. A flawless read with no breaths is a tell.
- Breath and mouth-sync. The biggest giveaway in dubs and generated voices: breaths that do not align with the mouth, or the voice describing a product at a constant volume even when the person turns away from the microphone.
- Extraneous sounds. A real review has incidental noise — packaging crinkle, product click, the sound of a lid opening. If the only sound is the voice, someone stripped the track, and you should ask why.
None of this means the video is worthless. It means you have a verification question for the creator — here is what to actually ask.
The realism tells that are too perfect
Some gives are about quality, not mistakes. Synthetic content frequently looks better than real mobile footage:
- Flawless skin with perfect light, no motion blur, and studio-level exposure in a video the creator says was shot on a phone in a bedroom at 7pm.
- Every frame in sharp focus. Real handheld footage has slight focus hunting and camera shake. Rock-steady, perfectly focused, with cinematic bokeh on a phone claim.
- A face that never blinks the way humans blink. Odd blink frequency and fully symmetrical, emotion-consistent expressions throughout.
- A delivery too fluent for the script. If the creator nails a dense, brand-written script word-perfectly on take one with no script glances, weigh that against how the rest of their portfolio talks naturally.
The honest principle: real people are inconsistent. A video where the person is literally flawless in every single way is not evidence of a great creator — it is evidence that you should ask how it was made.
What to ask the creator for
The cheapest, most reliable test is not forensic analysis. It is asking the creator to show their work. Any real creator can do this in two minutes, over WhatsApp, before you approve. Any synthetic pipeline will struggle.
- The raw file and any edits. The unedited first take, screen-recorded if necessary, or at minimum the editing timeline. A real creator has the original clips.
- The workflow in two sentences. "I shot it on my iPhone 12, two takes, voiceover cleaned in CapCut, background noise removed." That is a complete, believable answer. Vague answers are a flag.
- A quick live proof. Ask for a 5-second clip of the same person holding two fingers up near their face, saying today's date or your brand name. Real people comply instantly; generated faces break immediately.
- Location and product-in-hand proof. "Send a photo of the product on your desk right now" is beautifully hard to fake convincingly.
- Whether AI was used anywhere. Straight question, and you are entitled to a straight answer. On ugc.ng, portfolio quality is the moderation standard, and asking about the pipeline is a normal brief requirement — no follower count is involved, just the ability to do the job honestly: how to hire a UGC creator covers the vetting from the ground up.
If a creator admits to AI-assisted editing, that is not disqualifying. If they refuse to show raw files or dodge the workflow question, you have your answer — and your answer is "do not approve yet."
When it is fine anyway
Fairness cuts both ways. There are genuinely legitimate uses of AI in UGC that you might accept happily:
- AI-assisted, disclosed. Real human, real product, real review — with AI-written hooks, cleaned audio, or a background fix. This is a production choice. If the person is real and the claim is true, you are buying the same believability, just faster. Accept it — so long as the creator was upfront about the AI.
- Fully generated, disclosed and labelled. A brand that intentionally runs a synthetic character as an ad, labelled as AI per platform rules, is making a style choice — not committing fraud. The AI content rules for brands guide covers the disclosure obligations you inherit.
- Faceless content. Hands-and-product videos with no face are often the best value in the market precisely because they sidestep the whole "is the face real" question.
The rule to hold: you need to know which of these three situations you are in before the video runs as your ad. To paraphrase the human-versus-AI argument: if the audience suspects the person was generated, the believability advantage evaporates — and platforms expect realistic synthetic content in ads to be disclosed.
Your approval gate
You have 7 days. Use two of them on this checklist. Check hands, teeth, hair, background physics, audio, and the too-clean tells. Then ask for raw files and a live proof. Decide whether the pipeline is human, AI-assisted, or generated, and whether you are comfortable with the answer as your ad creative. If the creator answers honestly and the content performs, you have found a reliable asset. If they disappear when you ask for a live proof, you have dodged a problem, and the escrow design means nobody got paid for work you did not approve.
Post your brief, review submissions with this list in hand, and only approve what you can honestly vouch for.