Start with the question, not the ad
Meta's own guidance on A/B testing says to choose a hypothesis before you choose what to test. Its example is about audiences: you might expect one audience strategy to outperform another for your business. The same habit works for creative. Before anyone designs an ad, write down what you think will happen and why.
A hypothesis turns "let's try a video" into something you can learn from. If the result goes the other way, you still know something you did not know before.
A template
For [who], changing [one thing] from [A] to [B] will improve [a measure tied to the goal], because [what we know about these people]. We will read it at [review point] using [where the numbers come from].
Every part earns its place:
- One thing. The opening line, the kind of proof, the setting or the format. Change two things at once and nobody can say which one mattered.
- A measure tied to the goal. A shop cares about purchases, a clinic about booked appointments, a software firm about demos its sales team accepts. Clicks and views help explain a result; they are rarely the result.
- A reason. The "because" comes from customer insight: reviews, sales calls, support questions, what people search for. It is what makes the next test smarter whatever happens.
- A review point. Decide in advance when you will look, so a good or bad first day does not decide for you.
Examples across different goals
These businesses are invented to show the shape of a hypothesis. They are not clients and the outcomes are unknown.
- Online shop. For people who already carry a water bottle, changing the opening line from a general benefit to a specific time of day ("still cold at 4 pm") will improve purchases per thousand people reached, because reviews keep mentioning long workdays. Review after two weeks, using purchases recorded on the site.
- Hair salon. For people nearby who have never visited, opening with how quickly a time can be booked, instead of a finished style, will improve completed first bookings, because the most common question at the desk is how far ahead the salon is booked. Review against the salon's own calendar.
- Software company. For operations managers at small firms, leading with one weekly chore the software removes, instead of a feature tour, will improve demo requests the sales team accepts, because that chore comes up on nearly every sales call. Review after a longer window, since the decision takes time.
Run it as a real test
Meta recommends using its A/B testing tools rather than "testing informally, such as by turning ad sets or campaigns on and off manually", which it says can lead to overlapping audiences and unreliable results. Its tools split the audience so nobody sees both versions, and Meta recommends giving both versions the same budget.
Signs of a weak hypothesis
- It aims at something vague, like "better engagement".
- It changes the image, the headline and the offer all at once.
- The measure has nothing to do with what the business needs.
- There is no reason, so a result cannot shape the next idea.
- There is no review point, so the test ends whenever someone gets nervous.
The claim inside the ad still has to be true
A hypothesis is about which honest message works better. It is never a licence to test claims you cannot support. In Canada, the Competition Bureau explains that a claim about a product's performance or effectiveness must be based on an "adequate and proper test" carried out before the claim is made, and that this covers words and images on websites and social media too. If "still cold at 4 pm" is the hook, someone should have checked that it is.