E
Elsa
Guest
Founders keep asking me two things about AI video. Won't people scroll past once they see it's fake? And won't the platforms bury it?
I think both questions rest on the same flawed premise: that the goal is to make the footage pass as real. It isn't. Viewers don't mind AI. They mind being fooled. What follows is what our team has learned making AI videos for our own product, and what we'd do again. For scale, last week the two of us produced more than a hundred.
Two kinds of AI video are blowing up on TikTok right now. The first runs on contrast: an elegant grandmother in a pink headscarf doing Pilates and handing out life philosophy. Her following is in the millions, and tens of millions once you count other platforms. Everyone can tell she's generated. They watch anyway, because there is nothing else like her on the internet. The second kind is the faceless POV, where a modern person lands in an ancient era and has to deal with whatever comes up.
Neither one tries to fool anyone. The audience is in on it, and that's why it works.
If you're building a character, build in contrast, and make it specific to you. Skip the default faces and the looks everyone has already seen.
For a startup this is a cheap bet. Token budgets are generous, and attention is the scarce thing at launch. Once you have it, you can steer it toward the product.
AI products behave a lot like consumer goods. Plenty of them do roughly the same thing, and people pick the one they trust. In 2026, with everything looking fake, a real person vouching for you is worth more than it used to be. Cross-border e-commerce brands hire spokespeople for exactly this reason. Early on, your spokesperson is you.
So put the founder on camera, and start the account months before launch (I covered this in our previous episode on packaging an AI product launch). If recording take after take is a slog, give an AI avatar your face and voice and let it deliver launch announcements or the company story, with your office behind you. Avatar and video models like HeyGen and Seedance can do this.
If you'd rather stay off camera, build a mascot. Muse did this well: a fuzzy character that people now associate with the product, threaded through every scene of its launch film. A mascot is easy to recognize, it avoids the uncanny valley because it isn't human, and AI video is good at animation. Once you have one, you can make videos around it indefinitely. It's the same reason fruit-drama videos are everywhere on TikTok.
There's a second payoff. In the AI era, a lot of users arrive through screenshots. Someone gets a result they didn't expect and sends it to a friend or posts it. If your mascot or another recognizable element lives inside the product, every screenshot advertises you. Picture an AI music tool with a small purple character in every corner of the interface. Someone sees a screenshot on X, wonders what the purple thing is, and goes to look. Keep it consistent across the product, the videos and every launch asset.
Here is the structure we use for TikTok and Reels.
Why it works:
On realism, be concrete in your prompts about what "real" means. Describe the lighting, the skin, the setting. Left alone, the model gives you lighting that feels off and a face that looks too polished. Phone selfies help too. Mirror selfies work because the face doesn't need to be clear, and you can still build a persona out of them.
Our team of two made over a hundred videos in a week because we automated most of the pipeline and connected it to a dashboard that tracks how each video performs. After that it's a loop: test, read the numbers, cut what doesn't work, make more of what does.
On TikTok and Reels, I'd run one official account plus three to five persona accounts. Personas aren't official, but they often work harder than the official account as a growth engine.
A persona has to connect to your product's selling point. If you build for ADHD, the persona might be someone with ADHD, a stressed office worker, or a productivity blogger. Audiences watch people who look like who they want to be, or who fill a gap in their attention. Fill that gap first and the product follows. Give value before you ask for any.
Make the persona specific. One successful example gave its character a visible gap between the front teeth. Small details like that are what make a face feel like someone. If you run a campus ambassador program, let the ambassadors help shape the personas. Real personalities beat a generic template.
On paid creators, I'm less bullish on big KOLs than I used to be. Algorithms don't always favor a large account, and follower loyalty isn't constant, so some high-priced creators with huge audiences get fewer likes, shares and comments per video than you'd expect. Newer, smaller creators (KOCs, in Chinese marketing terms) often pull strong traffic at lower rates. Early on, you can run AI assets with them. And since AI makes production so fast, your own personas are effectively KOCs.
Don't post one cut everywhere, and don't repost TikTok videos to Facebook or Instagram videos to TikTok as they are. Reuse the assets, re-edit for the feed.
Founders should have personal accounts, and I'd encourage the whole team to have one too, even starting from zero followers. Treat it as a place to talk: comment, connect, share what you're learning. If you're building an AI health product, share what to eat for breakfast, explain why you're building it, and show up under established health accounts. Don't only market.
Then run an official product account for launches and demos. Coding models have made launch videos cheap. Pair a strong coding model with Remotion and you get the polished Silicon Valley launch-video look in a fraction of the time, so you can ship demos and product updates as video instead of text alone.
Every platform weights video more heavily now, so give your threads some motion. Generate a 2 to 3 second animated cover and attach it to the first post. It signals that the thread contains video, which gives exposure a lift.
Everything above is aimed at consumer products. B2B lives in cold email, outbound and LinkedIn.
AI tools can find your target accounts, whether that's investors or recruiting firms, and automate the outreach. Video fits in here. Generate a short video for each recipient, with their website on screen, explaining how your product helps their site or their hiring needs, and embed it in the email. It looks like you recorded it just for them, and most of the work is automated. Plain cold emails get low open rates, and a tailored video gives people a reason to click. Say it's AI-assisted. The same no-trickery rule applies.
On LinkedIn, use video to introduce your service, your approach and your customer success stories, instead of text-only posts.
Pick one platform. Build one character with a real point of contrast. Make ten versions of the sandwich and watch which hooks get taps. We'll cover the actual workflow, prompts and automation setup in a follow-up.
I think both questions rest on the same flawed premise: that the goal is to make the footage pass as real. It isn't. Viewers don't mind AI. They mind being fooled. What follows is what our team has learned making AI videos for our own product, and what we'd do again. For scale, last week the two of us produced more than a hundred.
People will watch AI video that gives them something new
Two kinds of AI video are blowing up on TikTok right now. The first runs on contrast: an elegant grandmother in a pink headscarf doing Pilates and handing out life philosophy. Her following is in the millions, and tens of millions once you count other platforms. Everyone can tell she's generated. They watch anyway, because there is nothing else like her on the internet. The second kind is the faceless POV, where a modern person lands in an ancient era and has to deal with whatever comes up.
Neither one tries to fool anyone. The audience is in on it, and that's why it works.
If you're building a character, build in contrast, and make it specific to you. Skip the default faces and the looks everyone has already seen.
For a startup this is a cheap bet. Token budgets are generous, and attention is the scarce thing at launch. Once you have it, you can steer it toward the product.
Trust is the real problem
AI products behave a lot like consumer goods. Plenty of them do roughly the same thing, and people pick the one they trust. In 2026, with everything looking fake, a real person vouching for you is worth more than it used to be. Cross-border e-commerce brands hire spokespeople for exactly this reason. Early on, your spokesperson is you.
So put the founder on camera, and start the account months before launch (I covered this in our previous episode on packaging an AI product launch). If recording take after take is a slog, give an AI avatar your face and voice and let it deliver launch announcements or the company story, with your office behind you. Avatar and video models like HeyGen and Seedance can do this.
If you'd rather stay off camera, build a mascot. Muse did this well: a fuzzy character that people now associate with the product, threaded through every scene of its launch film. A mascot is easy to recognize, it avoids the uncanny valley because it isn't human, and AI video is good at animation. Once you have one, you can make videos around it indefinitely. It's the same reason fruit-drama videos are everywhere on TikTok.
There's a second payoff. In the AI era, a lot of users arrive through screenshots. Someone gets a result they didn't expect and sends it to a friend or posts it. If your mascot or another recognizable element lives inside the product, every screenshot advertises you. Picture an AI music tool with a small purple character in every corner of the interface. Someone sees a screenshot on X, wonders what the purple thing is, and goes to look. Keep it consistent across the product, the videos and every launch asset.
The sandwich template
Here is the structure we use for TikTok and Reels.
- Hook, 1 to 2 seconds. An AI-generated person in a POV or reaction shot. Wide eyes, a real-looking desk, a caption like "I have three hours until my exam."
- Middle. A real screen recording of your product, with your link attached.
- Ending. A closing line from a real creator or an AI-generated clip.
Why it works:
- The first seconds decide everything. If someone taps your cover and stays, you've won maybe half the battle. If they don't, the video was never watched. That's why POV formats are so common. Test your covers constantly.
- It sidesteps the fake problem. People get suspicious when they stare at a synthetic face for a long time. Show it for one or two seconds and nobody has time to work out whether it's real.
- The middle is reusable. The screen recording stays the same. You only swap the first two or three seconds and the ending, so you can vary the character's actions, expressions, ethnicity and POV script and see which version gets taps.
- You can ride trends. Check which skits, dances and sounds are popular on TikTok, or have AI pull trending videos for you, then fit the sandwich into them.
On realism, be concrete in your prompts about what "real" means. Describe the lighting, the skin, the setting. Left alone, the model gives you lighting that feels off and a face that looks too polished. Phone selfies help too. Mirror selfies work because the face doesn't need to be clear, and you can still build a persona out of them.
Make volume cheap, then measure it
Our team of two made over a hundred videos in a week because we automated most of the pipeline and connected it to a dashboard that tracks how each video performs. After that it's a loop: test, read the numbers, cut what doesn't work, make more of what does.
Accounts: one official, several personas
On TikTok and Reels, I'd run one official account plus three to five persona accounts. Personas aren't official, but they often work harder than the official account as a growth engine.
A persona has to connect to your product's selling point. If you build for ADHD, the persona might be someone with ADHD, a stressed office worker, or a productivity blogger. Audiences watch people who look like who they want to be, or who fill a gap in their attention. Fill that gap first and the product follows. Give value before you ask for any.
Make the persona specific. One successful example gave its character a visible gap between the front teeth. Small details like that are what make a face feel like someone. If you run a campus ambassador program, let the ambassadors help shape the personas. Real personalities beat a generic template.
On paid creators, I'm less bullish on big KOLs than I used to be. Algorithms don't always favor a large account, and follower loyalty isn't constant, so some high-priced creators with huge audiences get fewer likes, shares and comments per video than you'd expect. Newer, smaller creators (KOCs, in Chinese marketing terms) often pull strong traffic at lower rates. Early on, you can run AI assets with them. And since AI makes production so fast, your own personas are effectively KOCs.
Adapt by platform
Don't post one cut everywhere, and don't repost TikTok videos to Facebook or Instagram videos to TikTok as they are. Reuse the assets, re-edit for the feed.
TikTok: exposure is the whole job. Be loud and grab attention in two seconds. You don't need an official account here; personas do the work.
Instagram: feels more polished and more branded. This is where an official account makes sense, in your brand colors, with the educational content that has a longer shelf life.
Facebook Reels: more natural lead-ins, less theatrical than TikTok.
X
Founders should have personal accounts, and I'd encourage the whole team to have one too, even starting from zero followers. Treat it as a place to talk: comment, connect, share what you're learning. If you're building an AI health product, share what to eat for breakfast, explain why you're building it, and show up under established health accounts. Don't only market.
Then run an official product account for launches and demos. Coding models have made launch videos cheap. Pair a strong coding model with Remotion and you get the polished Silicon Valley launch-video look in a fraction of the time, so you can ship demos and product updates as video instead of text alone.
Every platform weights video more heavily now, so give your threads some motion. Generate a 2 to 3 second animated cover and attach it to the first post. It signals that the thread contains video, which gives exposure a lift.
B2B plays a different game
Everything above is aimed at consumer products. B2B lives in cold email, outbound and LinkedIn.
AI tools can find your target accounts, whether that's investors or recruiting firms, and automate the outreach. Video fits in here. Generate a short video for each recipient, with their website on screen, explaining how your product helps their site or their hiring needs, and embed it in the email. It looks like you recorded it just for them, and most of the work is automated. Plain cold emails get low open rates, and a tailored video gives people a reason to click. Say it's AI-assisted. The same no-trickery rule applies.
On LinkedIn, use video to introduce your service, your approach and your customer success stories, instead of text-only posts.
Where to start
Pick one platform. Build one character with a real point of contrast. Make ten versions of the sandwich and watch which hooks get taps. We'll cover the actual workflow, prompts and automation setup in a follow-up.