Creatives and A/B tests
8x CTR difference between the best and worst creative. A/B test structure, sufficient volume for significance.
In eight years in performance, I haven’t seen a single channel in which creativity matters less than 50% of the result. On my projects between the best and worst creativity in one in the office, the difference in CTR is eight times, the difference in CPA is four to five times. This is not “designer magic”, this is a working system: formulate hypotheses, generate options, test with the right volume, find winners. This chapter is about system.
why does creativity weigh so much?
Auction algorithms for 2026 (Direct, VK, Telegram Ads) are optimized for CTR in the first 24–72 hours of the campaign. A creative with a CTR of 3.5% gets 2-3 times more impressions for the same bid than a creative with a 1.2% CTR. Further effect cumulates: the best creative gets more data, the algorithm learns more accurately, the target audience narrows down to the most relevant one, CPA drops.
This means: it is impossible to “win” in the office with bad creative, no matter how hard you changed the bets. This is a density auction: either you pay a lot or you have a steep creative Most often it’s either one or the other.
Anatomy of a working creative
Any working creative (banner, video, text) consists of four layers:
- Hook The first 1.5 seconds or the top half of the banner. The goal is to stop the scroll. Not “hello, our company,” but a provocation, a specific figure, an unexpected visual, a recognizable face.
- Value (offer). What do you offer and why is it useful to a specific person? segment. Not “quality services”, but “we will deliver the rug in 24 hours in Moscow.”
- Proof (proof). Figures, reviews, cases, guarantee. Removes objection “how do I know that you won’t cheat.”
- CTA (call). Specific single action: “leave a request”, “book”, “find out price”. Without a CTA, the person looked and left.
Most “bad” creatives lose one or two layers. The hook that is most often lost is and the man simply did not stop to look. Learn more about advertising creatives in blog article.
a/b test structure: what we change at a time
The main law of the A/B test: we change one variable. If you changed the title and picture at the same time - you won’t be able to understand what exactly worked. List of variables for priority test:
- Hook/visual. The biggest impact on CTR.
- Offer. The main headline and its promise.
- Social proof. Reviews vs numbers vs guarantee.
- CTA wording. “Buy” vs “Order” vs “Find out price”.
- Landing page. If the CRs are very different, we try a different landing page.
how much volume is needed for significance
The most common mistake is choosing a “winner” after 200 clicks and 5 conversions. On this the volume of “victory” of one option may be an accident. Minimum volume for statistical significance at the 95% level:
| basic CR | increase 10% | increase 25% |
|---|---|---|
| 1% | ~40,000 impressions per option | ~6,500 impressions per option |
| 3% | ~13 000 | ~2 100 |
| 5% | ~7 500 | ~1 200 |
| 10% | ~3 500 | ~600 |
The numbers are approximate, but the order is correct. If base CR is 3% and you catch an increase of 10% - you need 13,000 impressions for each option to understand “victory” real or accidental." Less is fortune telling. Details about the methodology - separate blog article.
how to generate creatives quickly in 2026
Previously, 5 creatives = 5 days of designer work. Now 5 creatives is an hour Midjourney + Photoshop + Canva. Generation duty cycle:
- We formulate 3–5 hypotheses (different “hooks”: fear of missing out, benefit, provocation, social proof).
- For each hypothesis - a prompt in Midjourney v6 / Nano Banana Pro / DALL-E (1 minute on prompt).
- We run the finished images through Canva with a brand template: title + CTA + logo (15 minutes for 5 creatives).
- Texts for creatives - ChatGPT with the role-prompt “performance-copywriter”, 3 options header for each creative.
- Load it into your account and run A/B. After 7 days - we leave 1-2 winners, Let's generate a new pack for new hypotheses.
Learn more about AI tools for creatives - chapter 16.
creative burnout: what to do
Any creativity burns out. On cold traffic - in 7–14 days (the audience gets used to it and stops noticing). On retargeting - in 3–7 days (audience is narrow, frequency hits the ceiling quickly).
Signs of burnout:
- CTR drops by 30% in 5–7 days with the same budget
- CPC grows without changing the rate
- Frequency per user exceeds 10 impressions per week
- Complaints and negative reactions are growing
What to do: have 3-5 fresh creatives on each channel. When we see burnout - turn off the burnt out one, start a new one. Creative update cycle - every 2–3 weeks on cold traffic, every week on retargeting.
typical failures in a/b tests
- Test everything at the same time. 12 creatives on a small budget - none of them will gain significance. No more than 3-4 options at a time.
- Stop the test early. "One is leading in two days" This is noise, not results.
- Change two variables. If you changed both the title and the image - it's impossible to know what worked.
- Ignore segment. Creativity works on cold, not on retarget (and vice versa). We test under the same conditions.
in the next chapter
Chapter 10 - Landing and Conversion. The structure of a landing page that gives 4–7% CR, five typical mistakes of the first screen and the link “creative → landing page” without loss of meaning.