Amazon Video Product Review Rules: A Compliance Workflow

Why Review Videos Succeed or Fail on Platform Rules

Every product review video gets judged twice. First by a viewer deciding whether to trust you, and second by a platform deciding whether to publish, suppress, or reject you. Most creators optimize obsessively for the first audience and barely think about the second. The result is predictable: a video that looks sharp, converts reasonably well, and then quietly gets demoted, age-restricted, or hard-rejected during upload.

The frustrating part is that compliance failures are rarely dramatic. They are small, boring details — an unlicensed music bed, a claim you can't substantiate, a synthetic voiceover nobody labeled, a burned-in caption that covers the safe area on mobile. Each one is fixable in minutes during pre-production and expensive after publishing.

This guide walks through the whole pipeline: what platforms evaluate, how to plan shoots and AI-generated assets, how to write disclosure that holds up, and how to run a quality-control pass that catches problems before a reviewer does. It is written for ecommerce teams, solo creators, and agencies producing review-style content at any volume.

The Five Compliance Layers of a Review Video

Before touching a camera or a generation tool, understand that compliance is not one checklist. It is five overlapping layers, and a video has to pass all of them.

Technical acceptability

Resolution, aspect ratio, frame rate consistency, bitrate, audio loudness, duration limits, subtitle handling, and safe-area framing. These rules are mechanical and easy to meet once you standardize your export settings.

Authenticity and disclosure

Who is speaking, whether they actually used the product, whether any segment is AI-generated, and whether a paid or gifted relationship is disclosed clearly enough that a casual viewer notices it without hunting.

Intellectual property

Music licensing, third-party footage, competitor logos, packaging artwork, screenshots of other sites, and product images you pulled from a search engine rather than from your own library.

Claims and substantiation

Health, safety, performance, durability, and comparative claims. The rule of thumb: if your script makes a factual promise, you need evidence in your files — test data, a spec sheet, a manufacturer statement — that supports the exact wording you used.

Category restrictions

Supplements, medical devices, children's products, weapons, adult products, and certain regulated goods carry extra rules about demonstration, imagery, and claims. A technique that works for a phone case can be disqualifying for a topical cream.

Pre-Production: The Checklist That Prevents Most Problems

Compliance is a research task before it is a production task. Build these artifacts once per product and reuse them across every video variant.

Verified facts sheet. Model number, dimensions, materials, what's in the box, price band, availability. Written down, not improvised on camera.
Claims ledger. Every factual statement the script will make, plus the source document that backs it. If a claim has no source, cut it or soften it to an opinion.
Asset license list. Music track and license type, stock footage IDs, brand-provided imagery, and permission emails. One spreadsheet column is enough.
Disclosure plan. Exact wording, where it appears on screen, at what timestamp it is spoken, and whether it belongs in the description as well.
Banned phrase list. Superlatives you cannot prove, medical outcomes, guaranteed results, competitor disparagement, and anything implying endorsement by a brand or retailer that has not endorsed you.
Shot list with intent. For each shot, note whether it is demonstration, context, or packaging. This makes it obvious later which shots can be generated and which must be real.

The single most common mistake at this stage is skipping the claims ledger because "it's just a quick review." Short videos compress claims into fewer seconds, which makes each one more prominent, not less. A twelve-second script can contain four separate assertions.

Capture and Generation Standards

Live capture that survives compression

Shoot at a higher resolution than you publish and downscale in the edit. Capturing at 4K and exporting at 1080p gives you room to reframe vertically and keeps fine product texture readable after the platform re-encodes your file. Keep frame rate consistent within a timeline — mixing 24p and 60p in the same sequence creates judder that reads as low quality even when nothing is technically wrong.

Audio standards matter more than most people expect. Record at 48 kHz, keep peaks around -6 dB, and target a final loudness around -14 LUFS integrated for most social and ecommerce destinations. A lavalier or a shotgun mic with a treated room will outperform an expensive camera with its built-in microphone every time.

Light the product, not the room. Two soft sources at roughly 45 degrees, one diffusion panel, and a neutral background will make a mid-range phone camera look better than a cinema camera under a ceiling fixture.

AI-generated assets done honestly

Generated footage is excellent for context shots: an environment the product belongs in, an abstract macro shot of a material, a stylized transition, a background for text. It is a bad idea for anything that functions as evidence — the product doing something, a result being achieved, a before-and-after comparison.

The distinction is simple: generation may illustrate, but it must not testify. If a viewer could reasonably conclude "this is what the product did," that shot should be real footage.

Practical quality checks for generated clips: inspect hands and fingers, watch for text that morphs or misspells, check that motion does not reverse mid-clip, and confirm reflections and shadows stay physically plausible. Keep a shot log recording which clips were generated and with which tool version, because you may need to answer a reviewer's question weeks later.

Disclosure and Synthetic Media: What Honest Looks Like

Disclosure fails for two reasons: it is technically present but visually buried, or it is written in language a normal person would not parse.

A workable structure has three touchpoints.

Spoken, early. Within the first fifteen seconds, say plainly whether you were paid, received the item free, or bought it yourself. One sentence is enough.
On screen, persistent. A small, legible label in a corner that stays visible for the portion of the video involving the sponsored or generated content. Not a two-second flash.
In the description. A written statement covering the same facts, which also serves as a record for your own team.

For synthetic narration or AI-presented segments, be specific rather than vague. "Narration in this video is synthetic" is clearer than "some content may be AI-assisted." If you use an AI avatar or cloned voice, say so once, early, and again in the description.

One more subtlety: staging can mislead even when every word is true. Filming a product in a kitchen you don't own, implying a test you didn't run, or using a prop result as if it were your own outcome are all authenticity problems. Reviewers and viewers both notice.

A Repeatable AI-Assisted Production Workflow

Step 1: Research and script skeleton

Write the script in three blocks: hook, demonstration, verdict. Put every factual statement in a claim slot rather than free prose, so it is easy to check against the ledger. Aim for a hook under five seconds, because drop-off in the first three seconds determines most of your performance.

Step 2: Generate or capture assets

Shoot demonstrations first, in one batch, using the same lighting setup and camera position for every product in a session. Then generate supporting B-roll: environment shots, macro textures, transitions, and title backgrounds. Generate two variants of each clip; generation is stochastic and one of two attempts usually wins.

Step 3: Assemble and edit

Use a master timeline template with pre-set letterboxing, caption styles, disclosure label position, and end-card timing. Templates are the highest-leverage compliance tool you have, because they make the correct choice the default choice.

For narration, record human voice where credibility matters and reserve synthetic voice for list sections, spec readouts, or multilingual versions of the same script. Always keep the source script and the audio file together in the project folder.

Step 4: Quality control pass

Run the checklist in the next section. Treat it as a gate, not a suggestion.

Step 5: Export and metadata

Export with a consistent filename convention — product, format, language, version — and write the title, description, and subtitles in the same session. Metadata written three days later is metadata written badly.

Quality Control: The Review Before the Review

Watch the finished video on a phone at roughly half brightness with sound off, then again with headphones on. Those two passes catch most real problems.

First three seconds. Is the product identifiable and is the premise clear?
Text legibility. Captions and on-screen text must survive small screens and low brightness.
Safe areas. Nothing important within the outer margins, so platform UI doesn't cover it.
Caption accuracy. Auto-captions mishearing a product name or a specification is a factual error you are now publishing.
Audio consistency. No sudden loudness jumps between spoken and music-only sections.
Music licensing. Track, license, and expiry documented.
Claims check. Read each claim aloud against the ledger.
Disclosure check. Spoken, on screen, and in the description.

Budget thirty minutes for this pass per finished minute of video. It feels slow for the first three videos and then becomes automatic.

Common Rejection Triggers and How to Fix Them

Comparison without evidence. Saying a product beats a named competitor requires documented testing. Fix: reframe as personal experience, or remove the name.

Generated footage presented as a demonstration. Fix: replace with real footage or relabel as an illustration with on-screen text.

Before-and-after imagery. Often requires documented conditions and timestamps. Fix: show the timeline and state the conditions.

Time-lapse implying speed. A compressed clip makes a slow process look fast. Fix: add a duration label.

Unlicensed music or footage. Fix: use a licensed library and log the track ID at the moment of download.

Visible third-party logos. Clothing, background signage, and competitor packaging all count. Fix: reframe, cover, or choose neutral wardrobe.

Undisclosed synthetic narration. Fix: one spoken line plus a description note.

Misleading price or availability. Prices change; recorded claims go stale. Fix: say "at the time of filming" or avoid the number entirely.

Scaling: Templates, Batching, and Version Control

Once a workflow works for one product, the goal is volume without variance drift. Three systems make that possible.

Templates. Master timelines for horizontal and vertical, each with locked disclosure placement, caption styling, and export presets.

Batching. Shoot five products in one session with identical lighting and audio. Then generate all B-roll in one sitting, then edit, then run QC in one final batch. Context switching is where consistency dies.

Version control. Keep a compliance log per product: script version, claims sources, asset licenses, generation tool versions, disclosure wording, and publish dates. When a rule changes, you can find every affected video in minutes instead of auditing your entire library.

Also build a prompt library for generated shots — reusable descriptions for "studio macro on dark surface," "morning window light interior," and similar — so every video in a series shares a visual language.

Decision Criteria: AI Generation, Real Footage, or Both

Use real footage whenever the product is the subject of a claim. Use generated footage when the goal is mood, environment, or transition. Use both when you need scale: real demonstration plus generated context is the standard hybrid for high-volume review libraries.

Consider three trade-offs when deciding. Iteration speed favors generation; legal defensibility favors capture. Cost favors generation for B-roll and capture for demonstration. Disclosure burden favors real footage, since generated segments add labeling requirements you must maintain on every variant, including translated versions.

Frequently Asked Questions

Can AI-generated video be used in a product review at all? Yes, for illustrative and contextual purposes, provided it is disclosed and never substitutes for evidence of the product's performance.

Do I need to disclose a free product? Yes. Whether you paid for the item is one of the first things a viewer should know, and reviewers treat hidden gifting as an authenticity problem.

How long should disclosure stay on screen? For the duration of the sponsored or generated segment. A brief flash is not disclosure.

What if my AI tool adds a watermark? Remove it only if your license permits, and never crop to hide a watermark you don't have rights to remove.

Do translated versions need their own disclosure? Yes, in the target language, spoken or written as the format allows.

How do I handle prices that change? Use relative language, or timestamp the statement on screen.

Is synthetic narration allowed? Usually, with disclosure. If the voice is presented as a real named person, that is a different and much larger problem.

How often should I re-check my library? Whenever a platform updates rules, and at least once per quarter. Run the compliance log and flag anything affected.

Building a Durable Compliance Habit

The teams that ship review videos consistently without rejection are not the ones with the biggest legal budget. They are the ones who made compliance structural: templates that place disclosure by default, scripts written in claim slots, asset logs filled in at download time, and a QC gate that nobody is allowed to skip because the deadline is close.

Start with one product. Build the facts sheet, the ledger, and the disclosure plan. Shoot the demonstration, generate the context, assemble in a template, and run the QC pass out loud. Then repeat the exact same sequence on the next product, and the next. By the fifth video, the process takes a fraction of the time and the rejections stop — because the errors you used to catch during upload are now caught before you ever hit export.