Why Your Ads All Look the Same — And How to Fix It

You generated ten ad creatives this quarter. Lined up side by side, they are one creative with ten sets of words in it — same layout, same rhythm, same block of text on the same coloured rectangle, in three approved shades.

You did everything right. You wrote a brand guide, you kept the palette consistent, you used the same template so the set would feel coherent. Your ads all look the same because consistency and variation were treated as the same problem, and they are opposites.

Then performance flattened, and the obvious diagnosis — creative fatigue — sent you to make more of them. Which produced more of the same.

Who has this problem

You are a founder running your own paid social, probably on a small budget, without a creative team to push back on you. You have read that creative is where most of the performance difference lives on modern ad platforms, so you have been dutifully producing volume.

If you have ever exported "three variants" that differed only in the accent colour and the button copy, and then wondered why the test told you nothing, this post is about that.

You are not lacking effort. You are running a loop with no source of difference in it.

Why this problem exists

Here is the mechanism, and it is worth understanding precisely, because the fix follows directly from it.

A generator that reads your brand record is a pure function of that record. Same brand in, same creative out. Your archetype picks the motion. Your palette picks the colours. Your positioning picks the words. Every one of those is a deterministic lookup, so ten generations produce one creative ten times. Changing the copy changes the words on it, not the concept.

That is the trap. The thing that has to vary for a test to mean anything is the concept — what is on the card, in what order, on what ground, moving how. Not the shade of the accent.

Randomising everything is worse, not better. The intuitive fix is to add noise: randomise the layout, the colour, the motion, the type. That produces incoherence — creatives that do not look like they came from the same company, which costs you the recognition that consistency was buying.

An art director does not randomise eight independent settings. They pick a route and commit to it. A poster is a coherent bundle of decisions. So is a data-led cut, a documentary treatment, a photographic one. Each is internally consistent and externally distinct from the others.

And a die repeats. Even with a set of routes, picking one at random will hand you the same route twice in a row often enough to matter. Ten rolls of a ten-sided die does not give you ten different results.

How to actually get creative variation

The Mark Platform's answer has two layers and a dealing mechanism. You can apply the same thinking with or without the platform.

1. Separate the layers: territory and constraint

A territory is a complete art-direction route — the poster, the data-led cut, the demonstration, the conversation, the photographic one, the documentary, colour blocking, negative space, kinetic type, the magazine spread. Each carries its own decisions about type, ground, structure and motion, and each has a note on when it wins and when it fails.

A wildcard is one Oblique-Strategies-style constraint laid on top: open cold, end on the number, halve the copy, one colour, invert one card, let one breathe.

Two layers, not one dial. The territory keeps the creative coherent; the wildcard stops the territory itself becoming the new template.

2. Walk a deck, do not roll a die

This is the small mechanical detail that does most of the work. The routes are shuffled into a deck and walked in order, with the position tracked per product. A deck does not repeat until it is exhausted. A die repeats immediately and often.

If you are doing this by hand, the equivalent is simple: write your routes on cards, shuffle once, and work through the pile before reshuffling. Do not pick whichever one appeals today — you will pick the same two forever.

3. Fence the variation so it cannot reach a guarantee

This is the part most people get wrong when they try to add variety, and it is why "make it more creative" usually makes it worse.

The draw is allowed to move what the brand already owns. It can rotate which brand colour carries the accent, flood a ground to a colour the brand records, invert to a brand neutral, go monochrome. It can change the structure, the motion, the emphasis.

It is not allowed to introduce a colour your brand does not have, write a new claim, or relax contrast, safe areas, the type floor or reading pace. Those are guarantees, and a variation mechanism that can reach them is not producing variety — it is producing defects.

Concretely: every derived colour goes back through the contrast check after the draw moves it. Type levers only multiply upward. Ground floods pick from your recorded palette or they do not happen.

4. Set the dial to match the moment

Variation is not always what you want, and this is where most advice on this topic is wrong.

• Holding one look across a launch? Turn the draw off. Every asset in a launch should look like the others.
• Producing a batch of genuinely distinct concepts? Turn it up, and expect roughly one in five to be wrong for you. That is the correct trade when you are hunting for a winner.
• Everything else? The middle setting — one route plus one constraint — makes two generations of the same template come back looking like they came from two different briefs.

The mistake is treating this as a quality setting. It is not. It is a distance from your default setting, and low is the right answer for a regulated brand or a launch.

5. Hold the draw constant across the variants of one test

Subtle, and it decides whether your A/B test is readable. Within a single generation, the three variants share the territory and differ only in the message under test. Re-drawing per variant would destroy attribution: "variant C won" tells you nothing when C differed in six ways at once.

Freshness comes from generating again, not from making the variants of one test more different from each other. This is the same discipline that makes funnel drop-off analysis worth doing — one variable at a time, or the result is not a result.

Why trust this approach

Because it is asserted mechanically rather than promised. Twelve consecutive generations must yield twelve distinct concepts, and no two routes may emit the same card sequence — both are tests in the repository, and both compare films rather than fields. An earlier version passed a naive version of that check while seven of ten routes were quietly emitting the same card order and differing only in easing. That is the exact failure this is built to catch, and it was caught by measuring output rather than counting the size of the deck.

It is also honest about what the mechanism cannot do. If your product record has no quantity in it, the data-led route cannot be built and is reported as unavailable with the reason and the fix — usually one number or one image URL. Silently skipping it would mean you never learn the route existed. And the deck cannot make a weak offer land: if the message is wrong, ten routes will show you ten well-composed versions of a wrong message. Get your positioning statement right first.

Frequently asked questions

Is this just A/B testing? No. A/B testing compares two things you already made. This is about making sure the two things are actually different enough to compare.

Will my brand still look like my brand? Yes — that is what the fencing is for. The draw only ever moves things your brand already owns.

How many creatives before I have a winner? Fewer than you think, if they are genuinely distinct. Many more than you think, if they are recolours of one concept.

What if I like one route best? Pin it. You can force a specific route when you want one, and reproduce an exact previous creative from its seed.

Your next step

Variation is a mechanism, not an attitude. Routes, constraints, a deck rather than a die, and a fence around the guarantees — that is the whole thing, and you can run it by hand on paper if you want to.

Or let the platform run it for you. See what the creative tools cover, or create your account and draw your first route →