How to make a UGC-style recommendation video from a website

Last updated 2026-08-24 · 7 min read

What is the short version?

A UGC-style video fails by being too good. The lighting, the pacing and the read all have to sit slightly below advertising polish, because the entire persuasive mechanism is that this looks like a person rather than a campaign.

1. Go to Vizard Agent and give it the product's website.
2. Say you want it to sound like a real person recommending it, not a polished ad.
3. Say the length and the platform.

What do you need before you start?

A URL. Vizard Agent reads the site for the product, the positioning and the visual identity, and generates the presenter and the demonstration screens, so a link is genuinely the whole input. What is worth adding is any specific thing you want mentioned, since a site says everything and a recommendation says one thing.

• The website. Vizard Agent reads it and captures it.
• The angle. What a friend would actually say about it.
• The persona. Roughly who is recommending it.
• The length. Thirty to forty seconds.
• Your own footage. Optional. Real footage beats a generated presenter when you have it.

What do you type into Vizard Agent?

Say what it should not be. Vizard Agent treats "not a polished ad" as a real constraint on the read, the pacing and the framing, and that negative instruction shapes the result more than any description of the product would.

Prompt

Variants worth knowing:

• Product mockups. Clean screens showing the thing working, cut between talking shots.
• B-roll inserts. Everyday footage that breaks up the talking head.
• Your own face. If you have footage, use it instead of a generated presenter.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real UGC-style video. The decisive step is the seventh, and it goes against instinct: rather than making the narration cleaner, Vizard Agent tightened the gaps to make it sound less composed.

1. Researches the website's details and offerings and captures a full-page screenshot.
2. Extracts the brand assets and reads the site's full text across several sections.
3. Searches for a natural, conversational voice rather than a commercial one.
4. Generates the narration and a realistic presenter portrait, then reviews the portrait.
5. Transcribes the narration for word-level timings and inspects the segment timings.
6. Tightens the pauses in the narration for snappier, less scripted pacing.
7. Generates the talking presenter in separate segments, reading the error log and splitting the audio when a longer pass fails.
8. Sources upbeat acoustic music and authentic b-roll inserts, then generates clean mobile interface mockups of the product.
9. Rebuilds the mockup screens as vector typography, generates word-timed captions, conforms every cut to vertical, mixes with measured ducking, then refines the caption contrast and safe-zone placement and re-renders.

Step 6 is the counterintuitive one. A voice model leaves even, professional gaps between sentences, and that evenness is what makes a read sound like an advert. Closing those gaps is most of the difference between a UGC video and a commercial.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 38.4 seconds, AAC audio. Vertical, thirty-eight seconds, a generated presenter cut against product mockups and b-roll, with word-timed captions placed inside the platform's safe zones.

Thirty-eight seconds is about as long as a recommendation holds. The format works on the premise that someone is telling you about something they like, and past forty seconds that premise stops being believable.

When does this not work well?

This format asks a generated person to pass as a real one, and that single fact is the source of both its effectiveness and every problem it carries. Vizard Agent builds it convincingly, and the questions that matter here are about honesty rather than about technique.

• A generated presenter is not a real customer. Do not present it as a genuine testimonial.
• Some platforms require disclosure. Synthetic presenters and paid promotion both have rules.
• Real footage beats generated. If you can film yourself or a customer, do.
• Claims come from the site. Vizard Agent narrates what the website says.
• Too polished defeats it. The moment it looks like an advert, the format has failed.

How do you fix a result that came back wrong?

Say what feels staged. Vizard Agent keeps the site content, the narration with its tightened timings, every presenter segment, the mockups and the caption tests, so a re-read or a pacing change is a re-render rather than a rebuild.

• "It still sounds like an advert." The pauses tightened further and the read re-cast.
• "The mockup does not look like our app." Rebuilt as vector screens from the real interface.
• "The captions are under the platform buttons." Repositioned into the safe zone, as this run did.

How does Vizard Agent compare to doing it yourself?

By hand this means either paying a creator and waiting a fortnight for a video you cannot revise, or recording it yourself and discovering that reading your own script aloud sounds nothing like talking. The pacing is the part that gives it away, and it is very hard to fix by re-recording.

Common questions

Does this work for a service as well as a product? Yes. Vizard Agent reads whatever the site describes and builds the mockups around it, and a service simply gets screens showing the workflow rather than the product.

Is a link really enough? Yes. Vizard Agent reads the site for the product, the copy and the visual identity, and captures the pages it needs.

Should I use a generated presenter? Only if you have no footage. A real person is always better for this format.

How long should a UGC video be? Thirty to forty seconds. Any longer and the premise stops holding.

Can it show the product working? Yes. Vizard Agent builds clean mockup screens of the interface and cuts them between the talking shots, which reads better than a screenshot of the website.

Why tighten the pauses? Because even, well-spaced delivery is the single clearest signal that something was scripted and recorded rather than said. Closing those gaps does more for believability than any change to the wording.

Do I have to disclose it? Check your platform's rules on synthetic media and paid promotion. That decision sits with you.

Will the captions be readable? Yes. Vizard Agent checks contrast and keeps them inside the platform's safe zones.

Does Vizard Agent check the finished cut? Yes. It reviews a contact sheet, verifies the loudness and re-rendered this one after refining the captions.