Three creatives are live, one performs better than the other two, and nobody knows why. That is the normal state of a badly posed test, and you recognise it because it produces no learning: the winner is kept, the other two are dropped, and next month starts from zero again.

A useful test is not better designed, it is better disciplined. Two rules written before launch are enough to turn a spend into knowledge.

What separates a test from a launch

A launch looks for a result. A test looks for an answer to a question asked in advance. If the question is not written first, the result gets interpreted afterwards, and an interpretation made afterwards is an opinion.

The question fits in one sentence, of the kind "does showing the product in the first three seconds improve click-through". It has to be phrased so that numbers can answer it yes or no.

This guide covers that arbitration. Producing the creatives themselves, what the script must contain and how disclosure is planned at the framing stage, belongs to UGC for paid advertising in France.

Isolating one variable, where possible

The most useful discipline is to change only one thing between two creatives. Same hook, same product, same length, one difference, and the result tells you something about that difference.

It is an experimental discipline, not a law of the craft. Two cases depart from it legitimately.

Testing a whole concept

Comparing two complete approaches, a testimonial against a demonstration for instance, is a valid test and often the most useful early in a campaign. It does not say why one wins, it says which one deserves digging into. Variables get isolated on the next round.

When volume does not allow it

Isolating a variable multiplies the number of variants, therefore the budget each needs to reach a readable threshold. On a small budget, two clear concepts beat six nuances that will never reach anything legible.

So the practical rule is the reverse of what you read everywhere: test concepts when looking for a direction, isolate variables when optimising a direction you have already found.

The cut rule, written before launch

This is the part almost nobody writes, and it is what separates a test budget from a wasted one.

The three numbers to write before launching

  1. The maximum budget per variant, past which you stop whatever happens.
  2. The minimum duration, below which you do not even look at the numbers.
  3. The gap you count as real, rather than a coincidence of a few clicks.

Those three numbers do not need to be perfect, they need to exist before the numbers arrive. Their only function is to take the decision away from you at the moment you will be least objective.

Without them the same thing always happens: the creative you believed in most stays live too long because excuses get found for it, and the one you disliked gets cut too early.

The usual objection is that a written rule stops you reacting to what you observe. It conflates two things. Nothing forbids stopping a test in flight if something obvious happens, a targeting error, a broken link, a comment flagging a product problem. What the rule forbids is extending a variant simply because you like it, and that is precisely the decision nobody makes clear headed in the moment.

One last precaution: never change targeting, budget or landing page while a test runs. The day one of the three moves, the test no longer compares creatives, it compares situations.

What is left when the test ends

A winner is not enough. A creative has a shelf life, and today's will be tired in six weeks.

What is left are the written learnings, one line each: which hook won, on which audience, against what. Three lines a month make, after a year, a knowledge of the product nobody else has, and it feeds the next briefs. It also settles arguments: a question already tested does not get retested because a new person on the team finds it interesting.

What is also left are the losers, which should not be thrown away. A creative that did not work in advertising can serve perfectly well on a product page or in an email, where it has a different job to do: there it does not have to stop someone scrolling, it has to reassure someone already looking.

The file worth more than the videos

A four column spreadsheet is enough: the date, the question asked, the winning variant, the gap observed. It fits on one page after a year and it answers the question every new brief raises, the one nobody can answer from memory: what has already been tried.

What a test will never tell you

It will not tell you why. Numbers say one thing worked better, never the reason, and the explanation you give yourself is a hypothesis for the next test, not a conclusion.

Nor will it tell you whether your product is of interest. A creative test compares ways of presenting the same offer; if the offer itself finds no audience, every variant loses together, and that is the one result to read as a product signal rather than a creative one.

It says nothing either about what happens after the click. A creative can win every attention test and sell nothing, because the page receiving the traffic does not carry on what it promised. That is a frequent cause of failure and it is corrected elsewhere than in the video, in landing pages that receive a UGC video.

Finally, it says nothing durable on too small an audience. Below a sufficient volume, the gap observed between two variants is noise, and reading it as a lesson produces briefs built on nothing.

Sources

Checked on 12 September 2026. This guide is not legal advice. Where this guide and the official source disagree, the source prevails.