One or two ad sets to begin with, and three or four creatives inside each one. Every ad set learns on its own and needs its own volume of conversions, so opening more boxes does not multiply the budget, it only divides it and stretches the time each box needs before it has enough data to decide anything.
Three levels, one kind of decision in each
Meta's structure has three floors and each one holds a different kind of decision. Getting confused about which floor a decision lives on is behind almost every messy account: somebody opens a new ad set to test a creative, or duplicates a whole campaign to change a headline, and three months later nobody can read what happened.
| Level | What is decided there | What touching it causes |
|---|---|---|
| Campaign | The objective and, if you put it there, a shared budget | Changing the objective is starting over |
| Ad set | Audience, placements, conversion event, schedule and bid | The ad set goes back into the learning phase |
| Ad | Creative, copy and destination | Impressions are redistributed inside the ad set |
Of the three, the middle one sets the pace. Learning happens at the ad set, not at the campaign and not at the ad, and that is where the platform counts the conversions it needs before it stops exploring and starts sharpening. Everything else in this piece follows from that single sentence.
The practical rule, then, is that an ad set is a hypothesis about who you are talking to, and an ad is a hypothesis about how you say it. If what you want to test is an audience, it goes in separate ad sets. If what you want to test is a hook, it goes in separate ads inside the same ad set. That one distinction prevents half of the chaotic accounts people inherit.
Why splitting the budget stretches the test
Budget does not multiply when you divide it into more boxes, it only divides. With the same money, going from one ad set to five means each one receives a fifth and takes five times as long to gather the conversions it needs to leave the learning phase.
On top of that comes a second effect the arithmetic does not show. When two ad sets aim at people who look very much alike, they compete for the same impressions and delivery stops being even: one takes most of it and the other barely runs. The comparison you meant to make between the two is no longer a comparison, because they were not given the same chances. How to build audiences that genuinely do not tread on each other is in the Meta Ads audiences guide.
That is why a sensible starting structure is narrow: one campaign, one or two ad sets, and the creatives inside them. You open a third ad set when there is a specific question that can only be answered by separating it, and with enough budget for it to reach the threshold on its own. What that threshold is, and how to work out the floor it demands, is in the minimum campaign budget guide.
How many ads fit inside one ad set
Three or four, and genuinely different. Delivery inside an ad set does not split evenly: it finds the best responding ad early and pushes it, so the spare variants are left with a handful of impressions each and never gather enough data for their result to mean anything.
Genuinely different means another hook, another format or a different angle on the same product, not the same video with a different button colour. Two variants that resemble each other do not produce one reading, they produce two halves of the same reading, and they eat the impressions the third one needed, which was the one actually testing something. What counts as a variant and when to rotate is in the ad creative guide.
There is one case where creatives do belong in separate ad sets: when a piece needs a format that only exists in certain placements and you want to control where it shows. That decision is not a creative one, it is a placement one, and placements live at the ad set.
Budget at the campaign or at the ad set
Both options exist and neither wins every time. Put at the campaign, the system allocates between ad sets according to what it sees, which works well when the audiences are similar and what you care about is the aggregate result. Put at the ad set, you guarantee spend in each box, which is what you need when one of them has to run no matter what.
That second case is more common in an agency than it looks. If a brand pays for its own investment separately, or a client insists that one product line has its own spend, the budget has to live where the accounting separation is, not where the algorithm would prefer to put it. The convenience of letting the system allocate is paid for with an awkward conversation when the client asks why their campaign spent half of what was agreed.
Meta has renamed these options several times, so the label in your account may not match the one in a tutorial from two years ago. What has not changed is the question underneath: who decides the split between ad sets, you or the system.
Everything in this article happens in one place in GoFeed.
Try it freeDuplicate instead of editing
Changing the audience, the conversion event or the bid of an ad set that is already running sends it back to the beginning of the learning phase. Since that is going to happen either way, the useful question is not how to avoid it but what you want to keep while it happens.
Editing the original ad set leaves you without a reference: the history is still there, but it mixes two different configurations under one name, and in a month nobody will know which part of those numbers belongs to which. Duplicating and changing the copy leaves the original intact as a comparison point, even paused, and turns the change into something you can describe in one sentence.
Two warnings about duplicating. The copy starts its own learning from zero, so it inherits nothing but the configuration. And if you leave both running at once on the same audience, you are back in the previous section's problem: two boxes competing for the same impressions on half the budget each.
One simple discipline keeps it all readable: one change at a time, written down with its date. When somebody asks in two months why the cost per result rose in the second week of October, that note is worth more than any dashboard.
When you run the structure of several brands
Everything above gets harder because of the number of accounts, not because of its difficulty. One brand with two campaigns is easy to hold in your head; eight brands with two campaigns each is sixteen, and the typical mistake stops being structural and becomes an oversight: a paused ad set nobody switched back on, or a daily budget left over from last month.
The ads dashboard reads Meta, Google, LinkedIn, TikTok and Pinterest and leaves campaigns, ad sets and ads in the same list, with their spend and their result beside them, so the weekly review happens in one sitting. Every ad account hangs off its own brand, so one client's work never mixes with another's.
Meta is also the platform where the whole hierarchy can be built from there: create a campaign, duplicate an ad set, put the budget at whichever level you want, and see which ad sets are still in the learning phase. On Google, LinkedIn, TikTok and Pinterest campaigns are created on their own platform, and a Google campaign is never created or renamed outside Google Ads.
That difference matters to the weekly routine more than it sounds. On a Meta account the review and the change happen where you saw the number; on the others, what you gain is the single list to decide from, and the change is made where the campaign lives. Knowing in advance which of the two you are doing saves the half hour of bouncing between tabs with eight accounts open.
Any profile can have it, whether you run clients, your own brands or your company's marketing, from the second tier of its plan upwards, and the detail of the module is on the ads page.