produccion-ai
How to create hundreds of ads with artificial intelligence, without opening Canva
Published: July 2026 · By Wen López
The short answer: volume does not come from a magic tool, it comes from a system. With AI you can produce dozens of ad variants (image, copy and format) in the time it used to take to design one, but the value is not in publishing hundreds: it is in testing angles until you find the one that converts, discarding ruthlessly. This article covers the four-step system we use at the Lab, the real numbers behind the public cases, and why human curation is what separates an ad that sells from spam with machine aesthetics.
Why volume matters (and what it really means)
In AI advertising, volume is not vanity: it is a search method. Nobody knows in advance which angle will convert; you discover it by testing. The public cases are clear on this: the Kalshi commercial that aired during the NBA Finals was made by generating 300 to 400 clips to keep 15 usable ones, a yield under 5%. Coca-Cola reported 5 specialists processing 70,000 generated clips to reach one final spot.
The lesson is not "generate a lot". It is: assume most of what you generate gets discarded, and design your system so discarding is fast and choosing is deliberate. The full detail of those cases, with sources, is in how much does a commercial cost in 2026.
The four-step system
1. The brief locks the angle, not the AI. Before generating anything, define: who you are talking to, which problem you press on, what you promise and what your brand can NOT say. A one-page brief prevents a hundred useless variants. Without a brief, AI produces generic volume; with one, it produces variations of an idea with intent.
2. Batch generation. With the brief as the frame, AI produces the variants: different copy hooks, different visual styles, different formats per placement. This is where the real time saving lives: what took a team a week now ships in an afternoon. The tool matters less than the process: the same batch can be produced with image generators, ad suites or directed agents.
3. Curation with brand judgment. The step almost everyone skips. From each batch, a fraction survives: what looks well made, sounds like your brand and does not promise what you cannot deliver. A high discard rate is not the system failing; it is the system working.
4. Small test, data decides. The surviving variants go live with small budgets and clear objectives. Within days the data says which angle wins; the winner gets more budget and a new round of variations. The cycle repeats: the system learns every week.
AI gives you the hundred variants. The brief decides what they search for, curation decides which ones exist, and the test decides which one wins. Three human decisions per hundred machine executions.
· Rather have us do it?
The Lab produces your creatives with human direction and AI execution: variants to test without the structure of a traditional agency.
The uncomfortable stat that defines quality
That stat is the business case for curation: consumers do not reject AI, they reject what feels made without care. The pieces that pass your brand filter do not carry that penalty; the ones you publish just to publish, do.
Common mistakes when scaling ads with AI
- Generating without a brief. A hundred variants of nothing are still nothing. The angle is a decision that comes before the tool.
- Publishing the whole batch. Volume is for testing, not saturating. Publishing 40 mediocre variants burns audience and budget.
- Copying the aesthetic everyone already sees. If your ad looks generically "AI made", it inherits the 44% discomfort. Your own style is defined in the brief.
- Measuring vanity. Impressions do not pay bills. Tests are measured in clicks, cost per result and conversion, with attribution set up from day one.
- Scaling the winner without refreshing it. Every angle fatigues. The system does not end with the winner: the winner feeds the next round of variations.
· Before you run ads
Your ads bring people in; your AI visibility keeps them. Measure how your brand shows up in ChatGPT, Claude, Perplexity and Gemini. Free, 30 seconds.
You may also like
Frequently asked questions
Which AI can I use to create ads for free?
Image generators have free tiers that cover first variants, and copy can be iterated with any general-purpose AI assistant. The free-tier limit is usually not the quality of one piece, it is volume and style consistency across pieces: coherent branded batches need a system, not just a tool.
How many ad variants should I test?
Fewer than AI can generate and more than you are testing today. A healthy starting point: 3 to 5 genuinely different angles, with 2 or 3 executions each, and equal small budgets. Data eliminates fast; your job is making sure the variants are truly different angles and not the same idea in another color.
Do AI-made ads convert worse?
Not because they are AI: they convert worse when they look carelessly made, and that is where Kantar's stat weighs in (44% of consumers uncomfortable with AI ads). A piece with brand direction and curation competes head to head; the human filter makes the difference, not the tool of origin.
Do I need a designer if I use AI for my ads?
You need design judgment, which is not the same as operating design software. Someone on the team must decide what looks good, what is readable and what respects the brand. AI does the execution; a person sets the standard.
Sources: Kantar, Media Reactions 2025 · IAB, Digital Video Ad Spend 2025 · Cases with numbers: how much does a commercial cost in 2026