InsightsNovember 11, 2026
How to Make AI Video Ads That Actually Sell
How to make AI video ads that sell, not just click: what to hand to AI, a six-step workflow, what 2026 research shows and when you have to label.
Matthias SteffensAI video ads work best when AI handles the slow parts of production and people keep the parts that sell: the idea, the real product footage and the final call. The workflow has six steps: collect your footage, brief once, let AI find scenes and write hooks, add voice and captions, multiply into versions, then review and label. In 2026 studies, AI-made ads matched human-made ads on clicks, but fully AI-generated ads trailed them on predicted sales.
Key takeaways
- Hand AI the repetitive jobs (footage search, hook variants, voiceovers, resizing) and keep people on the idea, the product proof and the final approval.
- AI ads matched human ads on clicks across more than 500 million impressions in a 2026 Taboola study, but fully AI-generated ads scored 5 points below Ipsos's sales benchmark while human ads scored 11 above.
- Creative drove 49% of incremental sales from advertising in NCSolutions' meta-study of nearly 450 campaigns, so testing versions is where AI pays off most.
- 82% of ad executives think Gen Z and Millennial consumers like AI-generated ads; only 45% of those consumers do (IAB, 2026).
Most ad buyers already use AI for video. In IAB's 2026 Digital Video Ad Spend & Strategy report, nearly two in three video ad buyers said they use generative AI for video creative, up from half in 2025, and they expect 43% of their ad assets to use it by 2027. The question is no longer whether to use AI, but where it helps and where it hurts.
What should AI do in a video ad, and what should stay human?
A video ad is a set of jobs, not one task. AI is faster than any editor at the repetitive ones and weaker at the ones that decide whether an ad sells: the insight, the proof that the product works, and the judgment of which version fits the brand.
| Job in the ad | Let AI do it | Keep a person on it |
|---|---|---|
| Finding footage | Search every file for the right scene in seconds | Choose the story and the hero shots |
| Hooks and script | Draft five to ten variants of the opening line | Approve claims, prices and tone of voice |
| Voice | Generate voiceovers in any language | Check pronunciation with a native speaker |
| Missing shots | Generate filler scenes, backgrounds and end cards | Keep the product itself real |
| Formats and versions | Build 9:16, 4:5 and language versions | Decide which versions go live |
| Review | Check every version against the brief | Give the final approval |
Audiences draw the same line. In IAB's 2026 study with Sonata Insights, 82% of ad executives believed Gen Z and Millennial consumers feel positive about AI-generated ads, but only 45% of those consumers did, a gap that has grown from 32 to 37 points since 2024.
IAB's advice: use AI to raise creative quality, not only to cut costs. The brands we work with land in the same place, with real people on camera at the core and AI shots only where they help, such as a filler scene or an end card; one agency we spoke with rules out AI voices for a client entirely.
How do you make AI video ads, step by step?
The six steps follow the order a production team already works in. The top row is what you decide; the bottom row is what AI does.
Collect the footage you already own
Everything starts with material AI can use: product shots, creator videos, past ads, behind-the-scenes clips. Put it in one searchable place, such as an asset library, and ask creators and agencies for raw clips and clean exports.
A file with subtitles burned into the picture is a finished ad, not footage: it cannot be re-cut into a new hook or language.
Footage of real people is worth the effort: Google's ABCD guidelines for YouTube ads tell advertisers to feature people. Plan for time, too: one large consumer brand we work with found only subtitled files in its folder of finished ads, and collecting the raw footage took about two months. For usage rights, see our guide to UGC ads.
Write one brief, not ten
AI needs the brief a good editor needs, in one paragraph: product, offer, audience, market, length and anything that must appear, such as a code or a date. Describe the outcome, not the edit. In a November 2026 demo, one spoken paragraph was enough: a 30-second launch ad for two new milkshake flavors in London, with a discount code to use at the counter.
Google's ABCD framework makes a good checklist: Attention (get to the heart of the story fast), Branding (show the brand early and often), Connection (help people think or feel something) and Direction (ask for one clear action). Google cites a Google and Kantar analysis in which ads that followed the ABCDs showed up to a 30% lift in short-term sales likelihood; Kantar notes that a predictive model scored 11,000 ads, so the lift is modeled, not measured in market.
Let AI find the scenes and draft the hooks
With the footage indexed, a search that took an editor an afternoon takes seconds: in the same demo, the agent searched 780 files, about 350 GB, and assembled an ad with three hooks, one body and one call to action. Read every hook against the brief first, because the opening carries the most weight. If the scenes are wrong, fix the brief, not the timeline.
Add voice and captions
AI voiceovers save the most time: the voice is generated next to the edit in seconds and regenerated until it sounds right, and a voice library or a cloned voice of your presenter keeps the sound consistent.
Language versions are worth it. In CSA Research's survey of 8,709 consumers in 29 countries, 76% preferred to buy products with information in their own language and 40% never buy from websites in other languages.
Translations need one check, because a translated voiceover can run longer than the original: in an Italian version from a May 2026 demo, the cut needed extra scenes to fit. Captions should follow the final voice and stay inside each placement's safe zones, and Google recommends repeating the on-screen call to action in the voiceover.
Multiply one ad into versions
Versions are where AI pays off most, because creative is the biggest sales lever you control. In NCSolutions' meta-study of nearly 450 consumer goods campaigns, creative drove 49% of incremental sales from advertising, more than twice the next factor, brand, at 21%. And demand keeps rising: in Adobe's 2025 survey of more than 1,600 marketers, 71% expected content demand to grow more than five times by 2027, and 37% said their paid social lacks enough fresh variation.
Write several hooks and endings for one body, render every combination, export the formats you buy and translate the set. In the May demo, five hooks and two calls to action made ten variations, 20 videos in two formats and 40 with an Italian version.
Change one thing per version, so the results show which hook or offer worked. Our ad volume calculator shows how many new creatives your spend level usually needs each week, and our guide to ad fatigue shows which part of a tired ad to replace first.
Review, label and export
Review is where the time goes: in the same Adobe survey, 89% of marketers said content passes three or more approval stages, and 58% spend more than 40% of their time on reviews and approvals. Check claims, prices, clip rights, captions and every AI-generated scene, collect comments in one shared link instead of chat threads, label what needs a label (see below) and export all formats together.
Do AI video ads perform as well as human-made ads?
On clicks, yes. On sales and memory, not yet, at least when the whole ad is generated. Three studies point the same way.
Taboola's study with researchers from Columbia, Harvard, TU Munich and Carnegie Mellon compared matched pairs of AI-made and human-made ads from the same advertiser, campaign and day, across more than 500 million impressions. AI ads had a slightly higher raw click-through rate, 0.76% against 0.65%, and performed comparably under the tightest controls. The winners were AI ads that did not look like AI, and a large, clear human face was one of the strongest trust signals.
Ipsos and Syracuse University's Newhouse School paired 10 human-made ads, produced before 2021, with fully AI-generated versions from the same brief and showed them to 3,000 US respondents. Only 13% of those who saw an AI ad were at least somewhat confident it was AI. But on Ipsos's sales-validated measures, the human ads beat the benchmark by 11 points on average, while the AI ads fell 5 points below it, doing worst where the brief needed storytelling or emotion.
NIQ's neuroscience study suggests why. More than 2,000 people watched AI-generated ads, about 150 of them with EEG. They spotted most AI ads intuitively and found them more annoying, boring and confusing than traditional ads, and even high-quality AI ads produced weaker memory activation, which NIQ says could weaken the motivation to act.
Ipsos draws the same conclusion we see in practice:
AI is a powerful tool, but the data shows that the human capacity for storytelling and emotional connection still creates a measurable competitive edge. The future is humans and AI working together.
So let people find the idea and film the proof, and let AI turn that material into the versions the platforms ask for. That is where AI pays off in creative testing.
Do you have to label AI video ads?
Often, yes, and the rules depend on the platform and the market.
- TikTok allows significantly AI-generated or AI-edited content in ads only with its AIGC label or your own clear disclaimer. TikTok's advertising policy counts completely AI-generated images, video or audio as significant, as well as making a person appear to do or say something they did not; small color or lighting fixes need no label.
- In the EU, Article 50 of the AI Act applies from August 2, 2026. Anyone who uses AI to create or manipulate realistic image, audio or video content that could pass as authentic, a deep fake in the law's terms, has to disclose it. Our EU AI Act Article 50 checker shows which of your ads are affected.
Disclosure costs less than many advertisers fear: in the IAB study, 73% of Gen Z and Millennial consumers said knowing an ad was AI-made would raise or not change their likelihood to buy, and more than half want AI video and AI images disclosed. Yet less than half of advertisers using generative AI always disclose.
Labels are easier to set when you know which scenes are AI-generated. Content credentials, the C2PA metadata that many generators now attach, record that; our content credentials checker shows what a file carries.
How to make AI video ads with Solid
Solid is the AI creative engine for brands and their agencies, and each of the six steps above has its place in it:
- Footage: the Smart Library cuts every upload into scenes, transcribes and tags it, so any shot is one description away.
- Brief: the Solid Agent takes the brief in plain words and follows your Brand Hub guidelines.
- Scenes and hooks: the agent picks scenes from your library and writes hooks and scripts; the AI Video Generator fills only the shots you never filmed, with models such as Veo and Seedance.
- Voice and captions: pick a voice or clone your presenter's, and fine-tune script, voice and cut in the Video Editor.
- Versions: the agent builds every format, plus language versions from the Grow plan.
- Review and export: in Review and Export, comments stick to the frame, clients review through guest links, and approved ads export together.
One brand we work with put a new end card on 300 ad files this way, after doing it by hand the first time. See how the Solid Agent works, or start with your own footage.
FAQ
Can AI make a whole video ad from scratch?
Yes. Text-to-video models such as Veo, Sora and Seedance can generate a complete spot from a prompt. In the 2026 Ipsos and Syracuse study, fully AI-generated ads were hard to tell from human-made ones but predicted weaker sales, and they worked best for simple, product-driven briefs.
What is the best AI tool for video ads?
It depends on the job. Generators create new footage, avatar tools put a synthetic presenter on screen, and AI editors turn the footage you already own into new versions. Our comparison of the best AI video ad generators shows which type fits which team.
How long does it take to make an AI video ad?
Once your footage is indexed, a first cut with several hooks takes minutes; review takes longer than production. Plan a day for your first batch and less for every batch after it.
Ready to build your first AI video ads?
Turn the footage you already own into tested versions in every format and language.
Start free trial or Book a demo
About the author
Matthias Steffens is Founder Associate (GTM) at Solid, the AI creative engine for brands and their agencies. He talks with brand and agency ad teams every week about how they plan, produce and test video ads. Connect on LinkedIn
Sources
- IAB: 2026 Digital Video Ad Spend & Strategy Report, July 14, 2026
- IAB and Sonata Insights: The AI Ad Gap Widens, January 15, 2026
- Ipsos and Syracuse University: AI ads and sales impact, May 18, 2026
- Taboola and university researchers: AI ads vs human creative, January 28, 2026
- NIQ: Consumer attitudes toward AI-generated ads, December 12, 2024
- NCSolutions: The Five Keys to Advertising Effectiveness, 2023 update, via Circana
- Adobe: Content demand survey of more than 1,600 marketers, June 16, 2025
- CSA Research: Can't Read, Won't Buy (B2C), July 7, 2020
- Google Ads Help: The ABCDs of effective video ads; Kantar: ABCD validation
- TikTok: Advertising policy on misleading and false content, April 2026
- European Union: AI Act (Regulation 2024/1689), Article 50
- Solid demos and customer calls, May to November 2026 (customers not named)