Key Summary — When multiple ads from the same advertiser overlap in one auction, Meta lets in only the one with the highest total value and removes the rest from that auction. Creating a separate campaign for each creative makes our own campaigns take delivery opportunities from each other. A campaign that loses the auction can’t show the creative to the people it should reach, yet the results table records it as if the creative lost, so it’s easy to miss that the test was wrong. For small brands, every ad is live from the moment it’s turned on. Put creatives with the same goal into one ad set and let machine learning choose among them.
Branding & marketing stories from Soulpapamarketing.
Plenty of advertisers still run Meta ads with one campaign per creative, a 1:1 setup. The intent is to set a budget for each creative and run a fair A/B test.
This is exactly why our A/B tests fail even though we think we’re running them. What’s scarier is that we don’t know it failed. Our own ads took exposure from each other, but the results table shows one creative winning and one losing, cleanly, as if nothing was wrong.
What is auction overlap in Meta ads?
Meta prevents ads from the same advertiser from bidding against each other and pushing CPM up. When several of our ads could be shown to the same person, Meta compares each ad’s total value before they enter the auction.
Total value is determined by combining three factors: bid, estimated action rates, and ad quality. Meta enters only the one ad with the highest value into the auction. The other ads cannot take part in that auction. Meta’s official documentation calls this situation, where several of our ads overlap in one auction, auction overlap.
It is a mechanism that keeps our own ads from driving up each other’s costs. However, an ad that gets cut is not shown at all in that auction. Meta also notes that even if you split ad accounts, auction overlap can still occur if they are identified as the same advertiser.
Why is creating a separate campaign for each creative a problem?
Even within one ad set, only one creative enters any given auction. The difference is that budget and learning are pooled together. Meta explains that it decides where, when, and to whom an ad is shown through ad auctions and machine learning. Within the same ad set, each time the person, placement, or timing changes, the creative predicted to deliver one result from that person at the lowest cost is selected, and no matter which creative wins, spend and conversions accumulate in one ad set. This choice is the core of creative management.
Split campaigns 1:1 by creative and the picture changes. Each campaign runs its own budget and learning, so machine learning cannot distribute exposure across campaigns. Campaigns aiming at the same audience compete on total value in every auction, and the ones that lose lose that exposure.
What happens when you run a View Content campaign and a Purchase campaign together?
There’s a quick way to see how strongly auction overlap kicks in. Turn on two conversion (sales objective) campaigns targeting the same audience at the same time. Set one to optimize for View Content (product page views) and the other for Purchase.
When we ran this setup on an account we manage, spend concentrated in the View Content campaign. The Purchase campaign barely spent its budget.
Looking at the three components of total value explains why. The chance that someone who sees the ad will land on the product page is far higher than the chance they’ll buy. Because each Purchase result is worth so much, the Purchase campaign’s bid is set high enough to partly close the gap in estimated action rates. Still, most people who see the ad can open the product page but are unlikely to buy right away. For these impressions, the View Content campaign’s total value is likely to come out higher. So the Purchase campaign only enters the overlapping auctions when the person is predicted to have an especially high purchase probability.
The same thing happens when you split campaigns 1:1 by creative. The campaign for whichever creative has even slightly higher estimated action rates takes most of the overlapping impressions, and the other campaigns lose their exposure.
What happens to campaigns that lose auctions?
Campaigns that keep losing in auction overlap can’t spend their full daily budget. Because a different campaign enters each overlapping auction, the only exposure left for this campaign is the non-overlapping kind. Meta notes that an ad set typically exits the learning phase once it gets about 50 results in the week after its last significant edit, but split campaigns struggle to hit even that threshold.
But failing to reach 50 results isn’t the real problem. The real problem is that wrong data piles up. The losing campaign’s numbers come from impressions that missed the people it should have reached, and the winning campaign’s numbers mix in impressions it took without competing under the same conditions as our other creatives.
With this data, we turn creatives on and off, move budgets, and plan the next creative. Because the decisions are based on data, they look data-driven. But the starting data is wrong, so the more carefully you judge, the further you go down the wrong path.
How do creative skill and exposure conditions differ?
Ad numbers come from two things combined. One is creative skill, meaning the power to move the right person when they see it. The other is exposure conditions, meaning how often the creative appeared in front of which people.
When auction overlap occurs, exposure conditions tilt to one side. Ads featuring influencers or collaboration partners, ads with an eye-catching model, and ads that have run long and built up results tend to gather response signals quickly and have longer histories, so their predictions tend to come out high first. In overlapping auctions, these ads tend to enter first.
Conversely, ads that explain the product earnestly without a hook, or ads aimed at a specific customer segment, only perform when they reach the right person. But if they lose to other campaigns even in front of that person, the ad ends up mostly shown to the wrong people. The results table is then filled with those people’s responses. The table marks this ad as weak, but those numbers reflect exposure conditions far more than creative skill.
Within a single ad set, things are different. Because creatives are selected per person, even an earnest ad can be shown to the people it fits.
Why don’t advertisers realize their A/B test failed?
The biggest problem is that the creative failed to reach the people it should have shown to, and the people who should have come in did not. The data starts to break down from here.
In front of the people who should have seen creative B, campaign A entered first in every overlapping auction. All that’s left for campaign B is the impressions A didn’t take. B’s CTR, conversion rate, and ROAS come from those leftover impressions. B may really have been weak. We just have no way to confirm it, because we never compared them under the same conditions.
Yet most advertisers don’t check whether auction overlap occurs when they split campaigns or ad accounts. They only look at which campaign’s numbers look better, head to head. That’s because the results table looks like a normal test. Campaign A has more spend and conversions, and campaign B has less of both. Read the numbers alone, and B looks like the loser. Spend and conversion figures alone can’t tell whether the creative is weak or whether it simply had fewer exposure opportunities.
So B gets turned off, and the next creative gets its own campaign and goes head to head with A. The wrong conclusion carries over into planning the next creative, and no clue remains to trace where it went wrong.
Meta also does not recommend unofficial tests that manually turn campaigns or ad sets on and off. Delivery efficiency drops, the results are hard to trust, and audiences can overlap. If the losing campaign failed to spend its daily budget, or ‘Learning Limited’ appeared in its Delivery column, that test result needs to be reexamined.
So how should Meta ad creatives be run?
Creative A/B tests are meaningless. Once you’re running ads, especially for small brands, everything is live. Budget set aside for testing is still money that has to generate sales, and if you split a small budget across versions, neither version can easily reach the results needed for learning (around 50 per week).
Don’t split campaigns and have people divide up the exposure. Put creatives with the same goal and the same optimization event into one campaign, and into one ad set where possible. Then let machine learning choose creatives to fit each person, placement, and time. If you want to manage budget at the campaign level, use Advantage+ campaign budget (formerly CBO).
Also settle on a single optimization event. If purchases are the goal, don’t keep a View Content–optimized campaign running alongside it on the same audience.
Seeing a creative with low spend within an ad set only tells you that machine learning picked it less often. Whether to keep that creative isn’t decided by building a separate test. Judge it by the per-ad performance accumulated in live delivery, and feed that into the next creative.
Meta recommends its A/B test feature, which compares non-overlapping audiences. However, each version requires its own budget, and Meta also notes that A/B tests require more budget. Brands with a small total budget often run out of money before results build up for each version.
While trying to test creatives ‘fairly,’ don’t tear up your very ticket into the auction.
Frequently Asked Questions
How can I tell whether auction overlap is happening in Meta ads?
If, among campaigns running on the same audience over the same period, one spends its full budget while the other keeps underspending, auction overlap is worth suspecting. If you use saved or custom audiences, go to the ‘Audiences’ screen in Ads Manager, select up to 5 audiences to compare, then click [Actions > Show Audience Overlap] to see the overlap percentage. Only audiences of at least 10,000 Accounts Center accounts (roughly 10,000 people) produce meaningful values.
How do I check whether I can trust the results of a creative A/B test that has already finished?
Start with the spend of the losing campaign. If it consistently failed to spend its daily budget, it had fewer exposure opportunities to begin with. From this result alone, it’s hard to say which creative is better. Also check whether the two campaigns’ audiences overlapped, and whether ‘Learning Limited’ appeared in the losing campaign’s Delivery column. If any one of these applies, don’t turn off the creative based on that result; it’s safer to pool the creatives into one ad set and run them again.
Can I run a View Content campaign to build a retargeting pool?
If you keep it running at the same time as a Purchase campaign on the same audience, the View Content campaign often takes the overlapping impressions. If you need the pool, it’s better to split audiences so they don’t overlap with the Purchase campaign, or to build the pool first at a different time and then move on to the Purchase campaign.
If weekly purchases are under 50, how do I resolve Learning Limited?
If campaigns with the same goal are split into several, start by consolidating them so results collect in one ad set. If weekly purchases still fall well short of 50 after consolidating, you can switch optimization to a more frequent event, such as Add to Cart, as Meta suggests. Even then, don’t keep a Purchase-optimized campaign running on the same audience.
Related Posts
- How to Do Brand Growth — Finding the Key Lever Among Five Tangled Metrics
- Beating Machine Learning with P(beat) — When should you replace the top Meta ad creative?
- Why Your Ads Hit an Efficiency Ceiling, and How to Revive Low-ROAS Ad Creatives
→ Our Brand Directing Approach — a performance and growth marketing agency that builds brands with data
<
p style=”font-size:0.85em;color:#888;margin-top:2em;”>Insights from Soulpapa Marketing — Korea’s digital marketing agency.
Original Korean article: https://soulpapa.co.kr/2026/10/13/meta-ads-auction-overlap-ab-test/
