Get customers · Lesson 4.5
10 Ways to Get Better AI Ad Creatives (Pick Two)
There is no shortage of advice on AI ad creatives. Most of it is a long list you will never finish. This post gives you ten tactics, says what each one is for and how to try it, and then asks you to do something unusual: pick two. Trying two things properly beats skimming ten.
None of these come with a promised result. Ads depend on your product, your price and your customer. What each tactic does give you is a clearer way to find out what works for you. If you want the full hands-on route, Lesson 4.5 covers it: Make ad creatives in minutes.
Choose two tactics from this list, run them for a week, and keep a note of what changed.
1. Write a style note and paste it every time
Describe your brand look in a short paragraph: colours, mood, scene type, what to avoid. Paste it at the start of every image request. This stops each new batch drifting away from the last one. It takes ten minutes and pays back on every ad you make afterwards.
2. Make ten different ideas, not ten small edits
When you ask for variations, AI tends to hand you near copies. Push for real differences: a close up, a scene in use, a bold plain background, a "before" scene, a "result" scene. Different ideas teach you something. Tiny edits mostly teach you nothing.
3. Write five headlines from five angles
For each image, write headlines that come at the offer from different sides: the result, the problem, the proof, the offer and a direct question. The lesson on ad creatives uses this same five headline habit. It gives you a fair comparison of words, not just pictures.
4. Put one idea in each ad
An ad that tries to say three things usually says none. Meta's guidance on text overlays says not to communicate too many messages because ads usually have only one call to action (Meta Business Help). Cut each ad down to one offer and one next step.
5. Make the vertical version first
A lot of ad viewing happens on phones, and platforms crop to fit. Meta lists 9:16 as a vertical format for Stories and Reels. For Stories and Reels it also advises leaving the edges clear of key text and logos so the interface does not cover them. If you design tall first, you can usually crop to a squarer shape later without losing the message.
6. Turn your best real photo into a short video
You do not need new footage. Take your strongest real photo and add gentle movement with a video tool, such as a slow zoom or a light sweep, then end on your offer. Watch the whole clip. AI motion can bend objects, so reject any version that shows something your product does not do.
7. Test one thing at a time
If you change the image and the headline and the audience together, you cannot tell which change mattered. Meta's A/B testing advice is to test one variable so your ad sets are identical apart from that thing, and it recommends tests of at least 7 days (source). Test images first, then headlines on the winning image.
8. Give the platform several versions to work with
Google's Performance Max guidance lists best practices such as covering every asset type, creating variations of assets to find the top performers, and adding as many assets as possible, up to 15 headlines, 5 descriptions and 20 images (Google Ads Help). It also says to wait 2 to 3 weeks before deciding to replace low performing assets. The pattern is useful even if you do not run that campaign type: make many versions, then wait long enough before judging them.
9. Match the ad to the page it links to
A great ad that lands on a confusing page wastes the click. Read the headline, then look at the page. Does the page repeat the same promise within a few seconds? If not, fix the page. Lesson 4.1 on building a website and the lesson on a page that sells cover that side.
10. Keep a swipe file of your own winners
After every test, save the winning image, headline and a one line note on why you think it won. Next month, start from that file rather than a blank page. Your own tested results beat anyone else's examples. For more ideas to model, see 12 AI ad creatives examples you can adapt.
How to choose your two
Not every tactic suits every business. Use this quick filter:
| If your problem is | Start with |
|---|---|
| Ads all look different from each other | 1 and 2 |
| Pictures are fine but nobody clicks | 3 and 4 |
| You have no budget for photos or video | 6 and 2 |
| You launch ads but never learn anything | 7 and 10 |
| Clicks come in but nobody buys | 9 and 4 |
A week plan for two tactics
Say you pick tactics 1 and 7. Here is what a week could look like:
- Day 1: write the style note and make ten images.
- Day 2: choose five and write one headline for all of them.
- Day 3: launch a test where only the image changes.
- Days 4 to 9: leave it alone. Resist the urge to tweak.
- Day 10: read the results, record the winner in your swipe file, plan round two.
The leaving it alone part is hard and important. Changing a live test halfway ruins what you were trying to learn.
What these ten have in common
Look at the list again. Almost none of it is about the AI. It is about being clear: one offer, one customer, one idea per ad, one change per test. AI makes the options cheap. Your job is to decide what is worth comparing and to be honest about what the numbers say.
If you only do one thing after reading this, write the style note and make your first ten images. Lesson 4.5, Make ad creatives in minutes, takes you through that first sitting with every step shown.
Questions people ask
Which tactic should I try first?
The style note. It is quick, free and improves every later step.
How many tactics should I use at once?
Two. More than that and you cannot tell what made the difference.
Will these tactics definitely lower my cost per click?
No one can promise that. They make your testing clearer, which helps you find what works for your own audience.
Do I need a big budget?
You need enough budget for the test to collect meaningful numbers. That depends on your market, so start small and see how quickly results arrive.