By Lee Moran, Founder of Veritas Digital
Our team has historically run paid social budgets of up to £100K a month across retail, travel and lead generation, and for years the lookalike audience was the go too for targeting. Unfortunately, this isn’t the case anymore. Whilst Meta has not announced the end of lookalikes, the focus nowadays has been on using Advantage+. Below is more detail about what’s changed and why the focus is no longer just the audience.
What is Andromeda, and why should you care?
Andromeda is the system Meta uses to decide which ads are even eligible to compete for an impression. It sits in front of the auction, sifting tens of millions of live ads down to a shortlist for a specific person, in milliseconds.
Social Media Examiner described the introduction of Andromeda in late 2024 as effectively replacing the targeting systems advertisers had relied on for years, and the important part is who is now in charge. Previously you chose the audience, and Meta served your creative to it. Now Meta chooses the audience, using your creative to work out who that should be. Meta’s own engineering figures put the change at roughly a ten-thousand-fold increase in model capacity, with a 6% improvement in recall and 8% in ad quality.
What retrieval means in plain English
Think of it as a doorman rather than an auctioneer. The auction still exists and you still place a bid, but Andromeda decides which ads get through the door. An ad that is never shortlisted never gets the chance to win on price, which is why an account can go quiet without anything obviously breaking.
What a lookalike audience is
If you have not built one, the mechanics are simple. You give Meta a seed list: your purchasers, a CRM export, people who watched most of a video. Meta maps the behavioural patterns of that seed across its user base and returns the people who most resemble them. The percentage you pick, usually 1% to 4% (but can go up to 10%), is a percentile cut-off rather than a confidence score. A 1% lookalike is the closest matching one per cent of the population you selected.
This worked well for years because Meta could see a great deal of off platform behaviour. Apple’s iOS14 tracking changes took a large bite out of that. Ad tech platform Dotidot notes that pixel based lookalikes now perform significantly worse than their pre-2021 benchmarks, with many advertisers finding them only marginally better than broad targeting. The lookalike was already weakening post iOS14 before Andromeda arrived.
So are lookalike audiences now dead?
It is worth being precise here, because plenty of headlines on this are not. Meta has not deprecated lookalike audiences. The build tool is still there and campaigns using them are running today.
What changed is their status. A lookalike used to be a gate: if someone was not in the audience, they did not see your ad. It is now a suggestion. Ad tech platform Madgicx describes Meta as treating lookalikes as audience suggestions rather than hard targeting rules, and that is the whole story in a sentence. Only a short list of settings still work as genuine hard controls: location, language, minimum age, exclusions and Special Ad Category status. Almost everything else in the targeting panel is now a suggestion.
Anyone who has watched Google’s automation arrive will recognise the change. Inputs quietly demoted from controls to suggestions, with the interface still displaying them as though nothing has happened.
| The old model | The Andromeda model | |
| What you controlled | Who saw the ad, through interests, demographics, lookalikes and exclusions | Creative, budget, conversion goal and a short list of hard controls |
| Status of a lookalike | An absolute. If someone was not in the audience, they did not see the ad | A suggestion. The system can and does deliver beyond it |
| What determined delivery | Audience definition, then creative | Creative, which is read to infer the audience |
| Ideal account structure | Many ad sets, each narrowly segmented | Few ad sets, fed with genuinely varied creative |
| Main failure mode | Audiences too broad or overlapping | Creative too similar, and thin conversion signal |
| Where the work sits | Audience research and account admin | Creative production and data quality |
Why Advantage+ won, and it is not purely about performance
The honest reason most accounts moved to Advantage+ is not that someone ran a clean test and Advantage+ won. It is that Advantage+ is significantly easier to run.
A lookalike led account is a genuine build. Seed lists that need refreshing, overlap to police, exclusions to maintain across a dozen ad sets, and a budget sliced so thinly that each ad set struggles to gather enough conversions to learn from. It needs someone competent looking at it every week. Advantage+ is close to plug and play: give it a goal, a budget and creative, and it goes. Anyone who has watched Performance Max take over PPC accounts will recognise the pattern.
That matters when you read the comparisons, because they are rarely comparing like with like. The published figures do favour automation. Madgicx reports Advantage+ audiences running around 18% lower cost per acquisition than classic lookalikes, and Advantage+ Shopping delivering roughly 17% lower CPA and 16% higher ROAS than manual campaigns. Data from analytics platform Lebesgue, cited by Confect, puts broad targeting at 49% higher ROAS than lookalike targeting under Andromeda. Meta’s own testing claims a 22% ROAS increase for advertisers using its Advantage+ creative tools.
Treat all of that as directional. These are aggregate figures from platforms and vendors with a commercial interest in automation. In our experience the accounts with most to gain are the ones that were being neglected, and the accounts with most to lose are the ones already being run properly.
The argument you should be having is about creative
This is where we would rather you spent your attention, because the audience debate is largely settled and this one is not. Under Andromeda, your creative is your targeting. Social Media Examiner is blunt on the point: if your ads are underperforming, the problem is far more likely to be the creative than the settings, and broad targeting frequently beats manual interest targeting because the system reads the ad to decide who should see it.
Volume is not the same as diversity
This is the expensive misunderstanding. The old playbook was to find one winning image and run fifteen headline variants against it. That now works against you. Meta’s visual recognition treats an image with slightly different text over the top as essentially the same image, and where it judges an account to lack creative diversity it can respond with higher CPMs. You pay a surcharge for repetition and never see a line item for it.
Meta has since surfaced a Creative Similarity metric in Ads Manager, which ad tech platform Optimizer flags as the clearest signal yet of how your library is being scored. Confect’s guidance is to aim for twenty to thirty genuinely different assets across formats. Different message, different visual, different format: change two of the three, or it is not a new asset in any meaningful sense.
The other half is signal quality
Creative feeds the retrieval stage. Conversion data feeds everything after it. If your event data is thin, duplicated or measuring the wrong thing, Andromeda will confidently retrieve the wrong people and Advantage+ will scale that mistake efficiently. Dotidot’s assessment is that the platform now prioritises first-party inputs and expects the Conversions API (CAPI) to supplement browser based pixel data. That is no longer a nice to have.
What this means for retail and lead generation accounts
For retail, the work moves upstream into creative production and feed quality. If your product data is untidy and you can produce four new assets a quarter, no amount of restructuring will rescue the account. The constraint is the studio, not the ad platform.
For lead generation the risk is larger. Seed lists are usually smaller and noisier, so the loss of precision bites harder. More importantly, if you optimise toward raw form fills, a system this powerful will find you an enormous number of people willing to fill in a form and no revenue at all. Feeding qualified or closed leads back into the platform, rather than raw enquiries, is now the highest value thing most lead generation accounts can do.
What we would do this quarter
- Stop building new lookalikes for prospecting campaigns and put the effort into creative. Keep the seed lists clean anyway, because good first-party data improves everything downstream.
- Count your genuinely distinct creative concepts, not your ad variants. If the honest number is under ten, that is your bottleneck.
- Check Creative Similarity in Ads Manager and treat a high score as a cost problem, not a design opinion.
- Verify the Conversions API is live and matching well, and check what your conversion event represents. A lead is not a sale.
- Before comparing this quarter to last, note when Advantage+ was first used. Much of what looks like performance change is structural change.
Frequently asked questions
Can I still create lookalike audiences on Meta?
Yes. The tool has not been removed and existing campaigns still run. What changed is that a lookalike no longer restricts who sees your ad. It informs the system rather than instructing it, so expect delivery beyond the audience you defined.
Is Advantage+ always better than a manual setup?
Not always, but it is better for most accounts most of the time, and it’s a lot less work. The clearest exceptions are small budgets, niche categories and B2B accounts with genuinely high quality customer data, where a well built seed can still earn its place.
How many creatives do I need?
More than you are running, almost certainly. Published guidance points toward twenty to thirty distinct assets, with the emphasis firmly on distinct. Fifteen versions of the same photograph with different headlines is one asset wearing fifteen hats, and the system will recognise this.
My cost per acquisition rose after switching to Advantage+. What went wrong?
Check three things before blaming the campaign type: whether your conversion event or attribution settings changed at the same time, whether your creative library is too similar to sustain broad delivery, and whether retargeting is being absorbed into prospecting. In many accounts the switch coincided with other changes, and the numbers are not comparable without adjusting for them.
The bottom line
Lookalike audiences have not been switched off. They have been overruled, which for practical purposes amounts to the same thing. Meta has taken the audience decision away from you and handed you a different job in exchange: producing enough genuinely varied creative and feeding the platform clean data via the Conversions API.
That is a harder job than building a 1% lookalike, and less comfortable, because it cannot be finished in an afternoon and does not sit in a settings panel. But it is where the performance now lives, and pretending otherwise is expensive.
If you would like an honest look at whether your paid social account is set up for how Meta works now rather than how it worked in 2021, that is exactly the kind of work we do at Veritas Digital. You can see how we approach paid social here.
Get in touch with us
Not sure whether your creative or your data is the thing holding the account back? Get in touch with us and we will take a look. No jargon and no hard sell, just a clear, commercially grounded view of what is working, what is not, and what it would take to fix it.
References
- Social Media Examiner – Facebook Ad Algorithm Changes for 2026: What Marketers Need to Know: https://www.socialmediaexaminer.com/facebook-ad-algorithm-changes-for-2026-what-marketers-need-to-know/
- Madgicx – Your Guide to Meta Lookalike Audiences in 2026: https://madgicx.com/blog/why-facebook-lookalike-audiences-are-worth-your-ad-spend
- Dotidot – Lookalike Audiences Meta: What Works Now in 2026: https://www.dotidot.io/post/lookalike-audiences-meta-what-works-now-in-2026
- Confect – Meta Andromeda: The Ultimate Guide to Meta Ads in 2026: https://confect.io/tactics/meta-andromeda-2026
- TheOptimizer – How to Test Ad Creatives on Meta After the Andromeda Update: https://theoptimizer.io/blog/how-to-test-ad-creatives-on-meta-after-the-andromeda-update-2026-playbook
- 1ClickReport – Meta Andromeda Update, Explained: What Changed and the Settings That Recover ROAS: https://www.1clickreport.com/blog/meta-andromeda-update-2025-guide
