A few years ago, running Facebook ads well meant building audiences by hand: interests, lookalikes, exclusions, dozens of ad sets. Today the platform's delivery system does most of that work itself. Broad targeting often performs as well as careful segmentation, and sometimes better.
That shift moved the main lever somewhere else. When the algorithm decides who sees an ad, the ad itself becomes the targeting. A message about planning for retirement finds people thinking about retirement. A message about side projects finds people with evenings to spare. The creative tells the system who to look for.
What this changes in practice
Variety of ideas matters more than variety of ad sets. We test many distinct concepts rather than many small variations of one. A new angle on the same product usually teaches us more than a new button color.
Each ad should speak to one person. An ad that tries to reach everyone gives the system nothing to work with. A specific situation, a specific frustration, a specific hope: that is what attracts the right viewers.
The first second does most of the work. In a feed, the opening frame or line decides whether anything else is seen. We write and design the hook first, then build the rest of the ad around it.
A simple testing rhythm
- Start with three to five clearly different concepts.
- Give each one enough budget to produce a meaningful signal.
- Keep the winners, retire the rest, and write new concepts inspired by what worked.
- Repeat every week or two. Audiences tire of an ad faster than most marketers expect.
Honest creative lasts longer
There is a temptation to chase attention with exaggerated claims. It works briefly, then it fails: ads get rejected, accounts get restricted, and the people who do click arrive with expectations no product can meet. Ads that describe a real benefit plainly stay in delivery longer and bring in customers who stay.
The algorithm finds the audience. The message decides whether it is the right one.
Our creative process reflects that. Every concept starts from a real customer problem and a claim we can support. The rest is testing.