Advantage+ audience will not fix a broken pixel or a polluted custom audience. It starts from your data and spends into the mistake faster. Before you hand it your budget, check three things in Events Manager: that events arrive at all, that the event you optimise for is the real one, and that it carries a value.
Every figure below comes from screens of Meta ad accounts I manage, recorded on 28 September 2026 and anonymised. One of them scored 100 out of 100 on an audit tool I built and was still not telling Meta which leads were worth anything.
Why Meta's AI cannot fix a broken pixel
If your cost per lead jumped and the leads turned to rubbish after you switched on Advantage+ audience, it is tempting to blame the targeting. Usually the problem starts earlier: the signal Meta learns from.
Search Engine Journal made the same point about AI agents in general: they act on audience data at a scale no team can, so wrong data spreads faster, and visibility can rise without the same rise in qualified traffic or conversions. Advantage+ audience works the same way.
Check 1 comes first, because a dataset that receives nothing makes checks 2 and 3 pointless. If your results dropped for another reason, start with why Meta Ads performance drops.
How Advantage+ audience actually works
According to Meta's help centre, Advantage+ audience starts from your audience suggestions: custom audiences, lookalikes, age, gender and detailed targeting. It prioritises those first, then searches more widely when Meta thinks that is likely to improve performance.
- Locationsa hard limit the AI will not expand beyond
- Minimum agea hard limit
- Languagesa hard limit
- Custom audience exclusionsa hard limit
Everything else is a suggestion, so it starts from your data. If your dataset is empty, it starts from nothing, and it still spends at a scale no team could.
Check 1: is anything arriving at all?
This is Events Manager on an account I manage, two datasets side by side. The first received 20,300 events in the last 28 days. The second has never received events. Not a quiet week: no event activity, ever.
- Open Events Manager, Datasets
List every dataset in the business.
- Find the one connected to the ad account you spend from
Not the one with the nicest name, the one actually linked to that ad account.
- Read its activity
If it shows no events, every campaign optimising on it is learning from nothing.
On another account I manage, no dataset was connected to the ad account at all.
If this check fails, stop and fix the connection before anything else. My conversion tracking service starts exactly here.
Check 2: the event name is not the event
A Lead event can match by name and still be wrong. The pixel sits on the thank-you page, and some of your leads never load it. Events Manager shows a Lead event, the ad set optimises for Lead, and nobody notices the count is lower than the real number.
- Pick one week
The same seven days in both places.
- Count leads in your CRM
The real number your sales team works from.
- Count the Lead event in Events Manager
Same week, same event the ad set optimises for.
Your optimisation event is wrong, and no targeting AI recovers that. Fix where the event fires before you change a single audience.
For the campaign side, see my Meta Ads lead generation guide.
What the connected dataset actually sent
Here is what that connected dataset sent last month: Page View, View Content and Search. Three events. Nothing else was listed, so there was no lead or purchase event to optimise for.
The 17.5% is Meta's estimate in its own interface, for that pixel. It is not a result measured on that account.
Event match quality and deduplication
Event match quality is a score out of 10, shown as Poor, OK, Good or Great, for how well the customer information you send with server events matches events to Meta accounts. It is available for web events, and Meta explains it in its match quality guide. On that account every event scored 4.4.
The pixel and the Conversions API were both firing, with no event ID. Without an event ID Meta cannot deduplicate the browser and server copies of the same event, and Meta's documentation says that can affect measurement, attribution and delivery optimisation for the event you optimise for.
- Send an event IDThe same ID on the browser event and its server twin, so Meta counts it once.
- Read match quality per eventEspecially the one your campaigns optimise for.
- Cover the events that matterA dataset sending only page views gives Meta nothing to optimise a lead campaign on.
For the setup side, see my data layer push and Google Tag Manager pages.
Where a bad event spreads: custom audiences and lookalikes
This is where one broken event becomes a much bigger problem. Custom audiences are built from website traffic, and lookalikes are built on those custom audiences.
A lookalike is only as good as the list under it. Feed it a list that is missing real leads, and Meta learns the wrong pattern, then Advantage+ audience starts from that pattern.
✅ Build from
- Lists that include every real lead or buyer
- Audiences rebuilt after the Lead event is fixed
❌ Avoid building from
- A website audience fed by an event that misses leads
- A dataset that only sends page views
Check 3: does the event carry a value?
This is the audit tool's report on Account A for the last 90 days: 100 out of 100, grade A. The report's own summary says that score reflects no policy violations and no technical errors. It does not mean efficient.
Conversion value tracking was not configured, so return on ad spend read 0.0. Meta was counting leads with no idea which leads were worth anything.
- Pass a value in the Lead event
A value on each lead tells Meta which leads are worth more.
- Or send revenue back later
Send closed revenue back to Meta as offline conversions, so it can see which leads actually paid.
That is why I read the findings, not the grade.
What a 32 out of 100 account looked like
Same tool, Account B: 32 out of 100, nine findings, four critical.
- Three campaigns spending with nothing trackedMoney going out with no conversions coming back.
- One audience spending with nothing to show6% of account spend at stake.
A campaign that reports nothing cannot be optimised by Meta either; its delivery is guessing. Put that account on Advantage+ audience and the guess just gets a bigger budget.
What to do this week
Run these in order. Each one makes the next easier to read.
Only then decide whether Advantage+ audience deserves more budget. The same order applies if you compare it with Google Ads: fix the signal first.
Get the 47-point Meta Ads audit checklist
The 47-point Meta Ads audit checklist is free to open, with no email wall. It starts with tracking and covers Conversions API deduplication, event match quality and lookalike source size.
Open Events Manager, Datasets, and find the dataset connected to the ad account you spend from. If it shows no events, nothing else in this guide matters yet.
Want someone to run these checks for you? I do this as a Meta Ads specialist in Dubai; book a call and I will tell you what your data is teaching Meta.
