A Repeatable Framework for AI Ad Creatives (LBWT)

Doing ads by feel gives you uneven results. One week you get a good image and a muddled headline. The next week it is the other way round, and you cannot say why. A framework fixes that by giving you the same four moves every time, so the only thing that changes is what you learn.

This post lays out a framework for AI ad creatives that fits the order of Lesson 4.5. It is a process, not a promise. It will not guarantee a result, but it will make your results easier to read. The lesson itself is here: Make ad creatives in minutes.

Each round, set your Look, make a Batch, write the Words, run a clean Test, then feed what you learned into the next round.

The four moves at a glance

The framework is called LBWT here, short for Look, Batch, Words, Test. It is a name for this post, not an industry standard.

1. LookWrite a short style note so every image comes out in your brand's look.
2. BatchMake ten different image ideas in one sitting and keep the best.
3. WordsWrite five plain headlines for each idea, from five angles.
4. TestLine the ads up so only one thing changes, run them, read the result, and record it.

Then it loops. The record you write at the end of Test becomes part of the Look note for the next round.

Move 1: Look

The Look is a short paragraph you paste into every image request. It has four parts:

• Colours, in plain words.
• Mood, such as warm, calm, bold or playful.
• Scene type, such as close up product on a wooden table, soft daylight.
• Avoid list, such as no text in the image, no faces, no clutter.

Why it matters: without it, each batch looks like a different company made it. Customers who see two of your ads should recognise both.

Keep the Look alive

After each round, update the Look. If a certain kind of scene performed well, add it. If an image style keeps producing odd results, add it to the avoid list. Over a few rounds this note becomes a real asset, written in your own words. The same habit of keeping a reusable brief is covered in how to brief AI like a senior hire.

Move 2: Batch

The Batch is where you make volume. The rule is ten different ideas, not ten edits of one idea. Use five idea families so you cover the ground:

• Product close up.
• Product in use.
• The result or feeling.
• The problem scene.
• A plain background with room for a headline.

Make more than you need and cut. Reject anything with garbled letters, odd hands or details that your product does not really have. If a customer would feel misled on opening the box, the image goes.

Make the shape early

Platforms crop. Meta's guidance lists 4:5 as the recommended ratio for single image ads in Facebook Feed and 9:16 for Stories and Reels. For tall placements it advises keeping key text and logos away from the edges so they are not covered (Meta Business Help). Make your best ideas tall first.

Move 3: Words

For each image idea, write five headlines from different angles:

• The result the customer gets.
• The problem they are tired of.
• A proof point you can really back up.
• The offer and the deadline.
• A direct question.

Keep the words plain. Read each aloud. If you would not say it to a customer across a counter, rewrite it. Never let AI invent proof. If it suggests a number, an award or a review, check it is real or delete it.

Move 4: Test

The Test is what makes the other three worth doing. The rules are simple:

• Change one thing. Meta's A/B testing guidance says ad sets should be identical except for the variable you are testing.
• Give it time. Meta recommends tests of at least 7 days, and says shorter tests may be inconclusive (source).
• Do not edit mid test.
• Write down the result before you do anything else.

A sensible order is images first (same headline, different images), then headlines on the winning image, then format (still against short video).

If you advertise on Google, its Performance Max guidance recommends creating variations of assets to find top performers, and waiting 2 to 3 weeks before deciding to replace low performing assets (Google Ads Help). The rule of thumb holds across platforms: be patient before you judge.

A worked example (illustrative)

This is an invented example to show the loop. The business and every number are made up. No real results are being reported.

Say you run a small bakery that sells weekend sourdough boxes. Here is round one:

• Look: "Warm cream and deep green, friendly and a bit rustic, close up on a wooden table in soft daylight, no text in the image."
• Batch: ten images across the five families, trimmed from fifteen after you reject two with strange crusts and three that look too glossy.
• Words: five headlines for the three best ideas, such as "Fresh bread on your table by Saturday" and "Order by Thursday, collect Saturday".
• Test: the same headline on five images, run for a week, with budget, audience and link identical.

At the end you write one line in your record: "Close up product on wood beat the in use scenes. Next round: more close ups, test headlines." That sentence goes into the Look note. Round two starts ahead of round one.

Making it a weekly habit

The framework is designed to fit a small slot in your week.

Times depend on you. The point is the rhythm: a block of making, a long quiet period of testing, a short block of reading.

Common ways the loop breaks

• Skipping the record. Without a written result, each round starts from zero.
• Changing two things. If the image and headline both change, you will not know which mattered.
• Reading too early. A few clicks over a day is noise. Wait for the full test period.
• Ignoring the page. If the ad works and the page does not, people click and leave. Fix the landing page with help from the lessons on a website and a page that sells.

For the numbers to read at the end of each round, see measuring AI ad creatives.

Start your first loop

Pick one offer, write your Look, and make your first ten images. Run the loop three times before you judge it. After three rounds you will have a Look note shaped by real tests and a record of what worked for your customers.

Lesson 4.5, Make ad creatives in minutes, takes you through round one with every step shown.

Questions people ask

Is LBWT an official method?

No. It is a name used in this post for a simple four step habit. Use any name you like.

How often should I run a round?

As often as your budget and attention allow. A round needs at least a week for the test, so weekly or every two weeks is realistic.

What if my test shows no clear winner?

That is a result. It means the thing you changed did not matter much to your audience. Pick a bigger difference for the next round.

Do I need a paid ads account to use this?

You need somewhere to run the test, but the Look, Batch and Words moves are useful for any creative work, paid or organic.