A few years ago, choosing creative was like a game of chance. Marketers intuitively bet on “favorite” images, team chats exploded with arguments, and budgets burned through tests faster than coffee in a thermos.
Today, everything is different. In place of emotions and “eyes,” algorithms have come, cold, systematic, and unbiased. They see what we miss: the pace of editing, the model’s angle of view, an overly bright background, or colors that don’t inspire trust. AI is no longer just an “assistant,” but a tool that decides the fate of an advertising campaign before it even starts.
But do all algorithms really work? And which ones help find the very “golden” creative that flies off the first test?
How did AI learn to “see” creativity?
What used to be called “creative’s sense” has today turned into a set of measurable indicators. Artificial intelligence doesn’t look at videos or banners like a human would, it breaks them down into data, compares them with millions of previous cases, and looks for patterns that correlate with high performance indicators.
AI models that work with creatives are based on three levels of analysis:
1. Text level how copywriting analytics works
Algorithms analyze:
Natural language processing (NLP) models are used for this, similar to GPT or Claude, but specially trained on marketing text corpora.
Result: the algorithm can assess whether the text sounds convincing, clear and is suitable for a certain audience.
2. Visual level, when AI literally “looks” at the image
Modern computer vision algorithms that use convolutional neural networks (CNN) analyze thousands of visual parameters:
Services like CreativeX, Pattern89 or Vidooly combine these parameters with historical campaign metrics and predict how much the visual will “get” to the target audience.
3. Level of performance: learning from success patterns
AI connects to data from advertising platforms: CTR, CPM, CVR, viewing time, number of interactions, creative fatigue (ad fatigue). The system then compares:
“What visual or text characteristics are repeated in creatives with the highest CTR or lowest CPA?”
This is called CTR-based learning — the algorithm builds a predictive model that can estimate with a probability of 70–90% which creative will work best, even before it is launched.
4. Who is already doing it in practice
5. Why is this important?
AI does not replace creativity — it makes the invisible measurable. Previously, the team tested dozens of options and hoped to “guess”. Now the algorithm immediately shows which creatives have a statistical advantage and this allows you to save up to 30–40% of the budget at the pre-test stage.
In essence, AI is not a designer, but an analyst with a memory for millions of creatives. It does not come up with ideas, but it knows exactly what worked yesterday — and can suggest what will work tomorrow.
TOP-3 AI algorithms worth knowing
There are dozens of algorithms today that promise to “find the best creative”. But only a few actually work, those that have access to large amounts of historical data and are already integrated into advertising ecosystems. We have selected three areas that have proven effective in the practice of arbitrage, e-commerce and performance marketing.
Meta Advantage+ Creative — the brain of Facebook advertising
Meta has long turned its advertising system into a self-learning mechanism. Advantage+ Creative is an AI module that analyzes previous campaigns, predicts CTR and automatically combines elements (images, texts, CTA buttons) into the most effective options.
How it works:
For whom: Facebook and Instagram advertisers, especially with large creative sets.
Cons: Lack of transparency AI “knows” what will work, but does not explain “why”.
AdCreative.ai / Kreateable / Pencil AI — performance assessment and rating
This is a new generation of AI tools that do not just generate creatives, but also assess their potential conversion before launch.
What they do:
Feature: the algorithm is trained on real advertising data (CTR, conversions, CPC) from thousands of accounts, not just on “theoretical” design samples.
Result: teams reduce the number of tests by 3-4 times, weeding out “weak” banners before they are even loaded.
For whom: e-commerce, affiliate projects, SaaS and digital agencies that want to minimize budget losses on tests.
Minus: AI can give biased estimates if trained on examples that are irrelevant to your niche.
Vidooly / CreativeX / Pattern89 — video creative intelligence
Video is the most difficult format for prediction, but this is where AI demonstrates the greatest breakthrough. Services like Vidooly, CreativeX, or Pattern89 use deep learning models that evaluate over 300 video parameters before launch:
Based on these factors, AI predicts whether the video will be able to hold the user for the first 3 seconds — a key point for advertising survival on TikTok, Reels, or Shorts.
In agency practice: such systems allow you to assess before launch which videos will “survive” to a view-through rate of 80%+, and which will drop out after the 2nd frame.
For whom: large affiliate teams, digital productions, brands with video traffic.
Cons: requires large data sets and corporate access fees.
These AI systems are not a “magic button,” but analytical filters that allow you to focus testing on the most promising options. On average, teams that use such algorithms halve the time of the pre-launch stage and increase the ROI of tests by 20-40%.
AI does not decide for you which creative is “brilliant” — it simply removes the noise, leaving room for what is really worth testing.
How to integrate AI selection into your creative process?
Today’s arbitration is no longer divided into “people versus machines.” The most effective teams work in the Hybrid Creative format — when the analytical accuracy of AI is combined with human intuition and context. The algorithm sees patterns, and the creative person sees meanings. And only together do they create what really “shoots.”
For AI to really help, and not confuse the cards, it is important to integrate it into the process systematically.
At this stage, AI does not replace the creative, but expands the field of options. Tools: ChatGPT, Ideogram, Gemini, Perplexity. What we do:
- we form a short brief or audience pain point;
we ask AI to generate several concepts from humorous to emotional;
we select those that have the potential to “hook” the first 3 seconds of attention.
The human task is to assess whether the idea fits the context, trends and language of the target audience.
Tools: Midjourney, Runway, Kaiber, Pika Labs. AI helps create prototypes of banners, storyboards and short videos without the participation of designers. What we do:
- we create a series of options with different visual accents;
analyze the reaction of the team or focus group;
leave 3–5 options for further evaluation.
Human role: adjust style, facial expressions, color and cultural subtext that AI does not always understand (for example, how not to cross the line between provocation and toxicity).
Tools: AdCreative.ai, CreativeX, Pattern89, Pencil AI. This is the analytics stage: the system checks each prototype for dozens of parameters — CTR forecast, contrast, text size, color scheme, expression of emotions.
What we get:
- “Creative Score” or performance rating;
recommendations on which elements should be changed (font, CTA, phrase length);
preliminary forecast of the cost of click and expected conversion.
At this stage, AI plays the role of an editor: it does not say what is beautiful, but what works.
No algorithm gives a 100% guarantee, so the final word is based on real metrics.
- we run a microtest with several creatives that received the highest AI score;
analyze CTR, CPC, CPA, retention;
return to AI with new data for further training of models.
This is how you create a cycle of continuous improvement — “AI → human → test → data → AI”.
Why And can’t you trust it completely?
AI doesn’t feel sarcasm, cultural allusions, or the context of war, politics, identity—what makes communication alive. It can predict a click, but it won’t predict a reputational failure or meme potential.
That’s why the role of the creative doesn’t disappear—it changes: from an “idea generator” to an algorithm curator. A person forms the task, sets the tone, and makes sure that the ad remains human.
Conclusion: a new role for the creative
AI didn’t come to take away the work of creatives, but only chaos. That endless stream of “lucky” tests when you upload twenty banners and pray that at least one will take off. Now, instead of “trial and error,” there are accurate analytics, patterns, and forecasts.
And this means that the role of the creative has changed. From an “idea machine” to an algorithm curator — someone who sets the direction, asks the right questions, and understands why this particular creative has a chance to work. The creative becomes a data designer: he chooses the input parameters, interprets the results, and adds what no AI sees — emotion, irony, human cultural code.
AI has learned to see colors, emotions, and patterns, but it doesn’t feel context. Therefore, no matter how tempting automation may be, it is the human being who remains the final filter between the algorithm and the audience.
Test with your head, not your emotions. The algorithm is not a competitor. It is your new partner in the creative game, who can count but not feel. And that is why you are needed both.