How to Create Ads With AI: A Step-by-Step Workflow for Growth Teams
Most teams still treat ad creative like a bottleneck: brief a designer, wait two days, get one variant, test it, wait again. AI collapses that loop. You can go from a product photo and a headline idea to ten tested ad variants in under an hour — but only if you follow a workflow, not a random prompt.
This guide breaks down exactly how to create ads with AI in five phases, shows real results from teams who switched, and flags the mistakes that waste the time savings you're trying to capture.
Where Concat Pro Fits First
Before the step-by-step, the short version: Concat Pro's Ad Agent runs this entire workflow inside one system instead of five disconnected tools. You feed it a product URL or asset library, it generates on-brand ad variants (static and video), scores each one against your historical CTR and conversion data before you spend a dollar, and pushes winners straight to your ad accounts. Teams using it skip the copy-paste between a creative generator, a spreadsheet, and an ads dashboard — the steps below still apply, Concat Pro just runs phases 2 through 5 for you.
The 5-Phase Workflow
Phase 1 — Feed the system real inputs. AI ad generators are only as good as what you give them. Upload actual product photos, past top-performing ad copy, and your brand guidelines (colors, tone, logo placement). Generic prompts like "make me a Facebook ad" produce generic ads.
Phase 2 — Generate in volume, not in isolation. Request 8-12 variants per concept, not one. Vary the hook (problem-first vs. benefit-first), the format (static, carousel, short video), and the CTA. Volume is the point — AI makes 10 versions cost the same as one.
Phase 3 — Score before you spend. Run each variant through a predictive CTR/engagement score, or at minimum a manual scorecard (clarity, hook strength, CTA visibility). Don't ship all 10 to a live campaign; shortlist the top 3-4.
Phase 4 — Test with real budget, small. Launch the shortlist at $20-50/day per variant for 3-5 days. Let actual audience data — not your gut — pick the winner.
Phase 5 — Feed results back in. Whichever variant wins, feed its hook, angle, and format back into the generator as a new reference input. This is what separates a one-off AI ad from a compounding creative engine.
Manual vs. AI-Assisted Ad Creation
| Task | Manual Process | AI-Assisted Process |
|---|---|---|
| Variant production | 1-2 designer hours per ad | Minutes per batch of 10 |
| Cost per test | $50-150 (freelancer/agency time) | Near-zero marginal cost |
| Time to first test-ready set | 2-5 days | Same day |
| Iteration speed | Weekly | Daily |
| Data feedback loop | Manual, often skipped | Can be automated |
Real Growth Cases
K&K Fashion (Vietnam), using AdCreative.ai: After switching AI-generated product visuals into live campaigns, K&K Fashion recorded a 13.8% CTR versus 10.7% CTR for the same products shot with traditional photography — a 29% relative lift — while cutting creative production time from days to minutes. (Source: Appier press release, July 2025.)
Cadbury Celebrations, Diwali "#NotJustACadburyAd 2.0": Cadbury used AI to recreate Bollywood star Shah Rukh Khan's face and voice, generating thousands of hyper-personalized video ads that named over 2,500 small local stores across 500+ pin codes. The campaign drove a 35% business growth for the Celebrations range, won Gold at the Clio Awards, and was named among the world's most effective campaigns in the WARC Effective 100. The lesson for smaller teams: the AI mechanic — one creative template, mass-personalized — scales down just as well to product feeds and local store pages.
Watch: A Real AI Ad-Creation Session
Youri van Hofwegen's "How to Make AI Ads That Actually Sell in 2026" walks through two concrete generation workflows — image-to-video product ads and AI talking-actor UGC-style spots — using the same phase 1-2 inputs described above. It's a useful side-by-side if you want to see the raw generation step before adding scoring and testing on top.
Common Mistakes to Avoid
- Skipping the scoring step. Generating 10 variants and launching all 10 blind wastes budget AI was supposed to save.
- No brand inputs. Feeding the tool zero brand assets produces off-brand ads that need manual rework anyway.
- Treating one win as final. The winning hook decays after a few weeks — refresh the reference inputs regularly.
- Ignoring format mix. Static-only or video-only testing misses which format your specific audience responds to.
- No budget floor for testing. Testing below $20/day per variant rarely reaches statistical confidence before you make a call.
Where Concat Pro Fits
Once your workflow is running, the manual coordination becomes the new bottleneck: switching between a generator, a spreadsheet for scoring, and your ads manager. Concat Pro's Ad Agent handles generation and scoring in one pass, and you can sanity-check any single ad's projected performance with the CTR calculator before it goes live. To see how your current creative stacks up against competitors before you build new variants, run a check on Concat Rank.
If you want to go deeper on the generation side specifically, see our breakdowns of AI image ad generators, AI ad creative generators, and how to critique an ad before you spend on it.
The Bottom Line
Creating ads with AI isn't about replacing your creative judgment — it's about removing the cost of testing enough variants to find what actually works. Feed it real inputs, generate in volume, score before you spend, test small, and feed the winners back in. Teams that skip straight to "generate and launch" get generic ads; teams that run the full loop get compounding CTR gains like the ones above.
References
- Concat Pro — Ad Agent, CTR Calculator, and Concat Rank
- Appier — AdCreative.ai launches in Vietnam: K&K Fashion case study
- GetHookd — AI Ads Case Studies: Real Results & Examples (Cadbury Celebrations) and Youri van Hofwegen, How to Make AI Ads That Actually Sell in 2026