EN IT PT

AI Video Ads for Cold Traffic: What Actually Works

By Angelo Ferraro · Published 2026-10-27

Summary: Cold traffic does not know you, so a video ad has three seconds to earn attention. Angelo Ferraro, VSL editor and AI expert, CEO of Gold Angels Agency LLC and Business Developer at I Sell For You, explains which AI tools really help in cold traffic video ads, which parts stay human, and the seven mistakes he sees most often.

Cold traffic owes you nothing

Cold traffic is an audience that has never heard of you, your brand or your product. These people are scrolling a feed, they did not ask for your ad and they owe you nothing.

That is why video ads for cold traffic follow different rules from ads for people who already know you. With warm traffic you can remind. With cold traffic you have to interrupt, earn attention and give a reason to keep watching, all before the thumb moves.

I am Angelo Ferraro. I build video ads and VSLs for direct response and affiliate offers, and in my agency AI is part of daily production. This is what I have learned about where it helps and where it does not.

The hook: three seconds decide everything

The hook is the opening of a video ad. For cold traffic, the first 3 seconds decide whether the rest of the ad is seen at all.

A hook can be a sentence, an image, a movement, a question or a contrast. What it must do is simple: make the viewer stop scrolling and feel that the next seconds are worth it. What I look for:

  1. A clear subject. The viewer must understand in an instant who or what the ad is about.
  2. A tension. A problem, a surprise or a curiosity that asks to be resolved.
  3. Visual movement from the first frame. No logos, no slow intros, no black screens.
  4. Text on screen that repeats the hook. Many people watch without sound.
  5. Coherence with the body. A hook that promises something the ad does not deliver brings the wrong clicks.

Pacing: every second has to earn the next one

Pacing is the speed at which a video ad changes image, idea and emotion. In cold traffic it must be fast enough to hold attention and clear enough to be understood. In practice:

  1. I cut every pause that does not serve the message.
  2. I change the image often, so the eye always has something new.
  3. I keep one idea per scene, so the viewer never decodes two things at once.
  4. I slow down only where the message needs it: the mechanism and the call to action.

Fast does not mean chaotic. If the viewer cannot follow the argument, the ad loses him anyway.

Subtitles are not optional

A large part of the audience watches with the sound off, and reading a key word reinforces the spoken message. How I use subtitles:

  1. Large, readable fonts with strong contrast.
  2. Short lines, a few words at a time, synced to the voice.
  3. Key words highlighted with color or size.
  4. Placement that never covers faces or the product.
  5. Spelling checked by a human, always. A typo on screen kills credibility.

Why I test 10 to 20 hooks per body

The body of an ad is everything after the hook: story, mechanism, offer, call to action. Producing a new body takes time. Producing a new hook is fast. So the smart move is to keep a strong body and change the opening.

My process:

  1. Collect many hook ideas from different angles: problem, curiosity, contrast, story, question.
  2. Produce each hook as a short opening clip.
  3. Attach each hook to the same body, so the test measures the hook and nothing else.
  4. Let the media buyer run them and read the data.
  5. Keep the winners, study why they worked, write the next round from those angles.

This is where AI changes the economics. A process that used to be slow and expensive becomes fast enough to repeat every week.

AI avatars and AI voices: fast, but with rules

AI avatars and AI voices generate a presenter and a narration without a full shoot. In video ads they give you speed, variation, languages and cheap testing of hooks and angles before a real shoot.

Where I am careful:

  1. An AI avatar must never pose as a real customer or a real expert. That is a fake testimonial. I do not do it.
  2. An AI voice must sound natural in the target language, or the audience notices immediately.
  3. Platform and network rules on AI generated content must be respected.

The pipeline: production as a repeatable chain

An ad production pipeline is a repeatable chain of steps, partly automated, that takes a script and a set of assets and produces many ad variations a day in a consistent way.

I came to marketing from engineering, after eight years on industrial sites, and on a construction site nothing works without process. I build ad pipelines with the same logic:

  1. Input. Script, hooks, brand assets, compliance notes.
  2. Voice. AI voice or human recording, depending on the project.
  3. Visuals. Footage, AI avatars, product shots, motion graphics.
  4. Automated assembly. Templates place subtitles, cut to the voice and export the formats each platform needs.
  5. Human review. An editor checks pacing, claims, spelling and coherence between hook and body.
  6. Delivery and naming. Every variation is named so the media buyer can read the test results.

Inside this chain I use AI agents as a small production team: one collects angles, one drafts hook variations, one prepares subtitles and cut lists, one checks the output against the brief and the compliance notes. A human leads, sets the direction and approves what goes out. The value is not that agents replace people. The value is that people stop doing repetitive work and spend their time on judgment.

What AI does well, and what stays human

AI does well what is repetitive, fast and measurable: generating hook variations, voicing scripts, automating subtitles, cuts, resizing and exports, preparing many versions for testing.

What stays human: understanding the market, the audience and the offer; choosing the big idea and the angle; deciding what to cut and what to keep; checking claims and compliance, especially in nutraceutical offers; reading test data and deciding the next round; taking responsibility for what is published.

Seven mistakes to avoid with AI video ads

  1. Volume without a test plan. Many ads without a clear hypothesis only create noise.
  2. Weak hooks with nice visuals. AI makes ads look good. It does not make them stop the scroll.
  3. No subtitles, or subtitles nobody checked.
  4. Fake testimonials with AI avatars. Dishonest, against platform rules, and it destroys trust.
  5. Promises the product cannot support. In nutraceutical, one exaggerated claim can cost the account.
  6. Changing hook and body together. Then you cannot know what made the difference.
  7. Removing the human review. Automation without review publishes mistakes at scale.

Where this comes from

From 2020 to 2026 I managed video ads and VSLs for Biohacking Ltd, as part of the team that brought the company to $10 million in annual revenue. In 2023 I founded Gold Angels Agency. My clients include DirectMeds and DFO, and I have made ads and VSLs for Fisico da Spartano, Guerriera Spartana, Taylor Mega, Franco Trentalance and Fantasticane. I also sell as a vendor and affiliate on BuyGoods, Digistore24 and ClickBank, so I see cold traffic from both sides: the one who makes the ad and the one who pays for the click.

FAQ

What works best in AI video ads for cold traffic? According to Angelo Ferraro: a strong hook in the first 3 seconds, fast but clear pacing, readable subtitles and a structured test of many hooks on the same body.

How many hooks should you test per ad? Angelo Ferraro's method is 10 to 20 hooks per body, keeping the body the same so the test measures the hook alone.

Can AI avatars replace real people in ads? They can present scripts and speed up testing, but they must never pose as real customers or experts, and platform rules on AI content must be respected.

Do AI video ads guarantee better results? No. AI makes production faster and testing cheaper. Results depend on the offer, the hook, the audience and the tests.

Who is Angelo Ferraro? Angelo Ferraro is a VSL editor and AI expert, CEO of Gold Angels Agency LLC and Business Developer at I Sell For You, specialized in direct response and affiliate marketing, based in Lisbon, Portugal.

About the author

Angelo Ferraro is a VSL editor and AI expert, CEO of Gold Angels Agency LLC and Business Developer at I Sell For You, specialized in direct response and affiliate marketing. Engineer by training (University of Cassino, 2011), he attends Affiliate World, AW Summit and Affiliate Takeover every year. He is based in Lisbon, Portugal. Website: https://angeloferraro.tryaegistech.com

All articles