Job ads on Facebook: how they find your candidates without age targeting

What is not allowed

For job ads, Facebook and Instagram require a special category, “Employment”. It removes the settings that could discriminate:

• no targeting by age or sex;
• no targeting by postcode or small radius: the area has a minimum size, about 17 km in radius when I measured it on 27 August 2026;
• no “lookalike” audiences built from your former employees.

A job offer must be visible to anyone able to do the work, whatever their age or gender. At Rafa, every ad is declared in this category, without exception.

So how does the ad find the right people?

By watching who responds. At first, the ad is shown fairly widely in the area. Facebook then looks at who stops, who clicks, who completes the application, and then shows the ad first to people who behave in the same way. It is not a portrait made of age or sex: it is a portrait made of behaviour.

This is the principle Meta itself describes, under the name of the learning phase. Meta recommends that an ad set gets about 50 conversions a week to leave it. That is not my figure: it is a benchmark from the platform.

One point I cannot hide from you

My 2026 landing page ads receive 8.9 applications a week at the median, and the most productive one 31.2. None of the 17 reaches 50 a week, even counting every application received. In other words, all my ads stay in a “limited” learning phase, in Meta's sense.

I do not know exactly what that costs, because I have not run the opposite experiment. I only know that it is the normal condition of a local recruitment ad: little volume, narrow audiences.

The role of the page: signals, not just applications

When an ad gets only a few applications, Meta has little to learn from. So on each page I measure three moments: the person arrives, they start applying, they send their application. These three signals are more numerous than applications alone, and they act as a fallback below the threshold of 50. With a built-in form, Facebook sees the application, but not the path that leads to it.

I have not isolated the effect of these signals on price: it is a design reason, not a measured result.

And your personal data?

People ask me this often, and rightly so. My rule fits in two sentences: I pass no direct identity data (name, email, phone) to Meta. After consent, only pseudonymous technical identifiers (advertising cookies, IP address, browser) are shared, to measure how well the ads perform.

What this changes for you

• Let the ad run for a while. Judging after two days means judging on almost nothing: the ad has not had time to show what it can do.
• Do not change everything all the time. A major change partly starts the Meta learning phase again from zero.
• Take care with the area. It is the only targeting allowed: it may as well be right.

In summary