🎤 SMK 2026 · 🇸🇮 Ljubljana · 🤖 AI Creatives
Why AI Isn't to Blame for Average Ads. The SMK 2026 Talk Recap.
On 29 May 2026, Adward founder Ervin Prislan spoke at SMK Conference about why AI hasn't fixed mediocre advertising — and what actually separates the teams shipping standout creative from the teams shipping noise. This is the full recap with the prep framework, the prompting rules, and the scale math. SMK attendees: use code SMK at checkout for 30% off your first 3 months.
29 May 2026 · 12 min read
What was the SMK 2026 talk about?
The talk argued that AI is not the bottleneck in modern advertising — the prep work and context fed into the model are. The difference between a team shipping average creative and a team shipping standout creative happens before the prompt, not inside the model. Average prompts are 8–12 words. Average briefs are 5 sentences. The work that separates winners is the work that happens before anyone opens an AI tool.
What’s the 70/30 split in AI creatives?
70% of the result comes from prep work — product information, ideal customer profile, marketing angle, and references with rules. The remaining 30% is the concept selection and the prompt itself. Most teams flip the ratio: they spend 90% of their time tweaking prompts and 10% on prep, then wonder why their ads look like everyone else’s.
How does 1 brief become 64,800 ad derivatives?
Math from the talk: 1,500 base creatives × 3 ad formats × 12 languages, plus roughly 20% feedback iteration = 64,800 distinct ad derivatives from a single well-prepped brief. Manual format adaptation takes 7–12 minutes per creative; automated takes 40 seconds. Manual localization takes 15–20 minutes per language; automated takes 40 seconds with ~90% accuracy across 20 markets at once.
Get 30% Off Adward — Use Code SMK at Checkout →
In this article
- The Hook: AI Keeps Improving — So Why Are Ads Getting Worse?
- Why Every Ad Looks the Same in 2026
- The Prep Framework: 70% of the Result
- From 1 Concept to 1,000 Creatives
- The Scale Math: 1 Brief → 64,800 Derivatives
- Formats and Languages: 12 Minutes vs 40 Seconds
- 12 Things You Can Generate With Good Prep
- The 20/80 Split: Creative vs Derivatives
- Try Adward — 30% Off With Code SMK
- Frequently Asked Questions
The Hook: AI Keeps Improving — So Why Are Ads Getting Worse?
The talk opened with a line worth sitting with: “The AI that impresses you today is the worst version you’ll ever use.”
Every model release is more capable than the last. Image generation has gone from uncanny-valley monstrosities to photoreal product shots in 18 months. Text generation can match a brand voice with a few examples. Video, audio, layout — all of it improving on a curve that doesn’t seem to flatten.
And yet, scroll any feed in May 2026 and the ads all look the same. Same gradients. Same product-on-pastel-background. Same “AI-generated woman holding phone.” Same five hooks recycled across a thousand brands.
If AI keeps improving, why is average creative quality dropping?
That’s the contradiction the talk set out to resolve. The answer isn’t about the model. It’s about what humans put in front of the model — and what they don’t.
Why Every Ad Looks the Same in 2026
In 2020, creative teams had limited tools and limited creativity. You shot what you could afford, you wrote what you could think of, and the bottleneck was production cost.
In 2026, creative teams have unlimited tools and unlimited creativity — at least on paper. Anyone can generate a thousand variants overnight. The bottleneck has moved.
So why does the output look so uniform? Three reasons:
- Everyone uses the same tools. Midjourney, ChatGPT, Sora, the same handful of templates. The tool stack is converging.
- Everyone feeds in the same context. Generic product descriptions, generic ICPs, generic references. If your input is a commodity, your output is too.
- Everyone skips the prep. The average prompt is 8 to 12 words long. The average creative brief is 5 sentences. AI returns exactly as much intelligence as the prompter brings to the conversation.
The talk’s most uncomfortable claim: AI is a mirror. If your creative looks generic, it’s because your prep was generic. Blaming the model is easier than fixing your brief.
Skip the Generic — Try Adward With 30% Off (Code: SMK) →
The Prep Framework: 70% of the Result
The core of the talk is a four-part prep framework. Ervin’s claim: these four inputs account for roughly 70% of the final creative quality. The concept choice and the prompt account for the remaining 30%.
1. Product Information
Most product briefs read like an Amazon listing. That’s not useful. What the model needs:
- Why pick this product over alternative X? Not features — the actual reason a customer would switch.
- What changes for the customer after they buy? The before/after, in the customer’s life, not in marketing copy.
- Is the purchase impulsive, considered, or premium? The category shapes every visual decision.
- What validation exists? A specific review quote, a clinical test, a money-back guarantee — something concrete.
- The customer’s own words. Pulled from reviews, support tickets, sales calls. Not your words. Theirs.
The example used in the talk was Magna Hydration, where the brief moved from “electrolyte mix” to a specific scenario about hot Slovenian summers, a specific competitor it replaces, and a specific reviewer quote about cramping. That’s a usable input.
2. Ideal Customer Profile
Most ICPs are demographics. “25 to 45-year-olds with disposable income.” That’s not an ICP. That’s a census category.
A usable ICP describes:
- Who the product speaks to as a person, not as a segment.
- The exact moment their problem appears. Time of day, room they’re in, what just happened.
- Their own words for the problem. Pulled from reviews and calls, not invented in a Notion doc.
The example used was Everyday Dose. The bad ICP: “millennials who want healthier coffee.” The good ICP: “people who feel jittery and anxious by their second cup, who’ve already tried matcha and didn’t like the flavor, and who use the phrase ‘I crash hard at 2pm.’” That’s an ICP the model can do something with.
The trap: “25 to 45-year-olds” tells the model nothing. It speaks to no one because it tries to speak to everyone.
3. Marketing Angle
The angle is the emotional or rational tension that grabs attention in the first three seconds. It’s the answer to: why should anyone stop scrolling for this ad?
Most teams pick one safe angle and apply it to everyone. Two traps to avoid:
- One angle for everyone. Speaks to no one. Different customers care about different things — wrap each angle around its own audience.
- Copying competitors. Looking at what’s working in your category and doing the same thing averages you out. If three brands are running “save time” ads, the fourth one should run “look like a pro” — not “save time faster.”
The example used was Peak Footwear, where the same product had three angles: performance for runners, status for office workers, regret-prevention for parents buying for teens. Three distinct creative directions. Same product. Same model. Wildly different output.
4. References and Rules
Most teams throw references at the model like a Pinterest mood board and hope for the best. That doesn’t work. Useful reference inputs are:
- Format rules. Where the logo goes. How tall the headline is. Where the product sits in the frame.
- Correct renders. Existing examples of “good” output the model can pattern-match against.
- What you like AND what you dislike. “Use this lighting” is half the instruction. “Don’t use this lighting” is the other half.
The example used was a Slovenian e-commerce brand called ULTRA, which had a strict rule that products always sit in the lower-right third of the frame, headline left-aligned at the top, with a specific orange (#FF4A1C) only on call-to-action chips. With those rules in the prompt, every output was on-brand. Without them, every output was generic.
The trap: dumping 20 references on the model without context. The model can’t tell which parts you wanted from each reference. Fewer references with clear rules beats more references with no rules.
From 1 Concept to 1,000 Creatives
Prep gives you context. Concepts give you creative direction. The talk laid out 10 concept types that, with good prep, scale into thousands of variants:
- Product-Led — the product as the hero, isolated and clean.
- Before & After — the transformation the customer experiences.
- Problem-Oriented — show the pain, then resolve it.
- Social Proof — review screenshots, testimonials, stats.
- Comparison Table — you vs. the obvious alternative.
- UGC — looks like a real customer shot it on their phone.
- Lifestyle — product in the context of someone using it.
- Magic Moment — the single best second of the customer experience.
- Founder Story — the why-we-made-this voice.
- Seasonal — tied to a moment in the calendar.
These aren’t templates. They’re directions. Each one can fork into hundreds of variants once the prep is solid.
Prompting Rules: 100–250 Words, Priority Order
Once you have the prep and the concept, the prompt itself is mechanical. The rules from the talk:
- Length: 100 to 250 words. Shorter than 100 and you starve the model. Longer than 250 and the model loses what mattered most.
- Priority order matters. Put the most important constraint first. Models weight early tokens more heavily.
- Define format, style, lighting, mood. Vertical 9:16 or square 1:1. Editorial or candid. Studio softbox or harsh sun. Optimistic, urgent, calm.
- Place the product in the scene. Not “a kitchen” — “the product sits on the counter to the left of a half-cut lemon.” Specificity rewards you.
A 100-to-250-word prompt with prep behind it consistently beats a 500-word prompt with no prep.
The Scale Math: 1 Brief → 64,800 Derivatives
Here’s where prep work pays off geometrically. From a single well-prepped brief, the talk walked through the math:
| Variable | Count |
|---|---|
| Base creatives generated | 1,500 |
| × Ad formats (square, vertical, horizontal) | × 3 = 4,500 |
| × Languages (EU markets) | × 12 = 54,000 |
| + Feedback iteration (~20% of variants re-generated) | + 10,800 |
| Total derivatives | 64,800 |
That’s from one brief. With good prep.
Without good prep, you generate 1,500 mediocre creatives and the math doesn’t matter — none of them perform.
The geometric multiplier is what makes prep work worth it. A one-hour deeper brief produces 64,800 differentiated ad derivatives. The same hour spent tweaking prompts on a thin brief produces noise at scale.
Run This Math With Adward — 30% Off (Code: SMK) →
Formats and Languages: 12 Minutes vs 40 Seconds
The two operational bottlenecks in creative ops are format adaptation and localization. Both are where AI’s leverage is largest.
Format Adaptation
| Approach | Time per creative | Success rate |
|---|---|---|
| Manual designer work | 7 to 12 minutes | High but slow |
| Automated AI resize | ~40 seconds | ~75% pass-through |
For a campaign with 1,500 base creatives, manual adaptation across three formats is 750+ designer-hours. Automated adaptation is roughly 17 hours of compute, with a 25% reject rate that gets re-prompted.
Language Localization
| Approach | Time per creative | Accuracy |
|---|---|---|
| Manual translator + designer | 15 to 20 minutes | ~98% but slow |
| Automated AI localization | ~40 seconds | ~90% across 20 markets simultaneously |
The 90% vs 98% gap matters less than it sounds — the 8% gap is mostly minor copy polish that a native reviewer can fix in seconds, not deep mistranslations. And the time gap is the difference between launching in 20 markets on Monday vs. launching in 20 markets sometime next quarter.
The trade-off the talk emphasized: you do not need 98% accuracy on 100 markets. You need 90% accuracy on 20 markets, today. Optimize for shipping.
12 Things You Can Generate With Good Prep
The last third of the talk was a tour of what good prep unlocks. With the 4-part prep done well, a small team can ship:
- Product photography — studio-grade renders without a studio.
- Ad scripts — short, mid, long-form video scripts in your brand voice.
- Social templates — Instagram, TikTok, LinkedIn, Pinterest formats.
- Comparison pages — you-vs-competitor landing pages.
- YouTube thumbnails — A/B-testable thumbnails by the hundred.
- Packaging mockups — for retailer pitches and internal review.
- Video production assets — b-roll, transitions, lower thirds.
- Newsletter creative — header images and inline visuals.
- Catalog imagery — product catalogs at scale across SKUs.
- Pitch decks — investor and partnership decks with consistent visuals.
- Campaign concepts — full creative directions before any execution.
- UGC-style concepts — looks-like-a-real-person ads without casting.
A team of three with good prep can now do what a team of fifteen needed to do in 2020. That’s not hyperbole — it’s the practical takeaway.
The 20/80 Split: Creative vs Derivatives
The closing argument of the talk reframes how teams should spend their hours:
- 20% on creative work. The brief, the concept, the prep framework. The thinking.
- 80% on derivatives. The variants, the formats, the languages, the iterations. The doing.
Most teams in 2020 were the inverse: 80% on doing, 20% on thinking. The doing was expensive. AI inverted the economics. Doing is now cheap; thinking is the moat.
The closing line of the talk: smaller teams can now compete with corporations — but only if they spend the 20% on the right things. Spend the 20% on prompt tweaks instead of prep, and the 80% is just noise at scale.
Try Adward — 30% Off With Code SMK
Adward is the AI ad creative tool built specifically for European markets. It bakes the prep framework from the talk into a workflow:
- Product Library — paste a URL, get a clean, prep-ready product profile in 30 seconds.
- ICP builder — guided prompts to push past demographics into actionable customer profiles.
- Reference rules engine — upload references, define rules per reference, get on-brand output.
- 20+ EU languages — native-quality localization, not machine translation.
- 30+ ad formats — auto-resize for every platform, 75%+ pass-through on first try.
SMK attendees get 30% off the first 3 months. Use code SMK at checkout. Available on all paid plans, from €25/month.
Start Free — Use Code SMK for 30% Off →
Frequently Asked Questions
Who was the SMK 2026 talk for?
The talk was aimed at performance marketers, creative leads, founders, and growth teams who use AI tools daily but feel like their output is converging on the same generic look as everyone else’s. If you’ve prompted Midjourney 500 times and your ads still look generic, this talk was for you.
Where can I watch the recording?
The SMK conference organizers will post the official recording in the weeks following the conference. Check smk.si for the 2026 conference archive. If you’d like a notification when the recording is live, reach out to projects@conversion-design.com and we’ll send the link directly.
Does Adward actually implement this prep framework?
Yes. The prep framework in the talk is the same framework baked into Adward’s product workflow — Product Library captures the product info, the ICP builder forces specificity beyond demographics, and the reference rules engine prevents the “throw 20 references at the model” trap. The talk is the philosophy; Adward is the implementation.
How is Adward different from generic AI tools like Midjourney or ChatGPT?
Generic AI tools optimize for general-purpose output. Adward optimizes for performance ad creative in European markets specifically. That means: native localization for 20+ EU languages (not translation), 30+ ad-format auto-resize, GDPR compliance with EU-region data servers, and trained on European ad performance data. The talk’s framework works on top of any AI tool — Adward is the version that doesn’t make you build the workflow from scratch.
What is the SMK code and how do I use it?
SMK is a 30%-off discount code for new Adward subscribers. Use it at checkout to get 30% off your first 3 months on any paid plan. It works on monthly and annual billing. One code per account.
When does the SMK code expire?
The SMK code is valid through 31 August 2026 for new subscribers. Active SMK conference attendees who want an extension can reach out to projects@conversion-design.com.
What’s the most important takeaway from the talk if I only have 30 seconds?
Stop blaming the model. Spend more time on prep — product info, ICP, marketing angle, references with rules — and your generic ads will start looking specific. The model isn’t the bottleneck. Your brief is.