Meta's AI now picks your audience. That makes creative — not targeting — your real job
Chandan KumarFounder · Performance Marketing SpecialistListen to this article
The short answer
With Meta's Andromeda retrieval engine and the Advantage+ automation suite, the AI now decides who sees your ad — and it does it better than any manual audience stack. The old skill of interest-layering and lookalike-stacking is fading; narrowing the targeting just starves the model. The leverage moved to three things humans still own: creative is the new targeting (ship 8–12 distinct angles, not two polished films, and let the winners reveal the audience), feed it qualified conversions and values via the Conversions API (or it optimises toward cheap junk that never buys), and the offer and the funnel after the click (no automation rescues a weak offer, a slow page or a two-day callback). Go broad, win on creative and signal, and own everything on either side of the click.
On this page
For most of the last decade, being good at Meta ads meant being good at audiences. You layered interests, built lookalikes off your best customers, stacked exclusions, and split-tested it all until you found the pocket of people who converted. That skill is quietly dying. Meta's Andromeda retrieval engine — rolling out since late 2024 and now the backbone of the platform — plus the full Advantage+ automation suite have taken audience selection out of your hands and given it to a model that, honestly, does it better than any manual stack I've ever built. Meta's own numbers put a roughly 22% ROAS lift on advertisers who switched on Advantage+ targeting. So when a founder asks me 'what's your targeting strategy on Meta?', my honest answer in 2026 is: that's last year's question. The leverage moved.
The audience is now an output, not an input
Andromeda is, in plain terms, a much smarter matchmaker. It retrieves and ranks which ad to show which person from a vastly larger pool, far faster and with better recall than the old system. Meta reported meaningful jumps in retrieval quality from it, and you feel it in the accounts: broad targeting, which used to be where budget went to die, now routinely beats tight manual audiences. The model finds your buyer better when you give it room to look rather than boxing it in.
The mistake I see constantly is founders and junior buyers trying to 'help' the AI by narrowing it — five interests, a tight age range, three exclusions. All you're doing is starving the model of the signal and the search space it needs to do its job. In 2026, your audience is something the system discovers, not something you dictate. Your real job is to give it room to find the right people and then give it the signals to recognise them. Which brings me to the two things that actually matter now.
What is Andromeda?
Andromeda is Meta's machine-learning retrieval engine — a much smarter matchmaker that retrieves and ranks which ad to show which person from a vastly larger pool, faster and with better recall than the old system. It's why broad targeting now routinely beats tight manual audiences: the model finds your buyer better when you give it room to look rather than boxing it into a handful of interests.
Creative is the new targeting
When the machine decides who sees your ad, your creative becomes the thing that decides who responds. A scroll-stopping, problem-aware hook effectively self-selects its audience — the people it speaks to stop and convert, Andromeda learns from that, and it goes to find more of them. In other words, creative is now your primary targeting lever. You're not telling Meta who to find; you're showing it, one ad at a time, and letting the response do the targeting for you.
So we've completely changed how we brief. We don't make two polished films and pray. We ship eight to twelve distinct angles — different hooks, different pain points, different formats, a lot of it raw, UGC-style and shot on a phone, because that's what performs now. Each angle is a probe into a different segment. The winners tell us who the real audience is, which then sharpens the offer and the next round of creative. Volume and variety of angles, not production polish, is what feeds this machine.
How to brief creative when the AI picks the audience
- Ship many angles, not two films — 8–12 distinct hooks, pain points and formats per round.
- Treat each angle as a probe — it self-selects the segment it speaks to; the response does the targeting.
- Favour raw over polished — UGC-style, phone-shot creative usually out-performs glossy production now.
- Let winners sharpen the next round — the angles that convert reveal the real audience and the next offer.
Feed it conversions, not clicks
This is the same discipline that decides whether Google's AI works, and it's just as true on Meta: the automation is only as good as the signal you send it. If your pixel fires a 'lead' on every form submission and you never tell Meta which of those leads were real, Andromeda will optimise toward whatever is cheapest to collect — which is almost always low-quality, high-volume junk. Cheap leads that never buy will wreck an Advantage+ campaign faster than any targeting mistake ever could.
So we wire the Conversions API properly, server-side, and we close the loop. When a lead qualifies or a customer actually pays, that event goes back to Meta — ideally with a value attached, so a high-ticket enquiry counts for more than a low one. For one client we stopped optimising for 'leads' and started sending back 'qualified leads' only; the cost per lead on the dashboard went up, and the cost per actual customer fell by a third. The AI didn't get smarter. We just stopped lying to it about what success was.
By the numbers
On one account we stopped optimising for raw 'leads' and sent back only qualified leads via the Conversions API. The cost per lead on the dashboard went up — and the cost per actual paying customer fell by about a third. The AI didn't get smarter; we just stopped telling it that cheap junk was success.
What humans still own: offer, angle and the funnel after the click
It's worth being honest about what all this automation does not fix. Andromeda will not save a weak offer. Advantage+ will not rescue a landing page that loads in nine seconds, or a business that takes two days to call a hot lead back. The machine got dramatically better at one narrow job — matching an ad to a person likely to respond. Everything on either side of that click is still completely yours, and still completely manual.
That's actually good news, because it's where the durable advantage lives. Anyone can switch on Advantage+; the button is free and your competitor has it too. What separates accounts now is the strength of the offer, the sharpness of the creative angles, and whether the funnel after the click — page, follow-up, sales response — is built to convert the attention the AI is buying you. We've moved our hours out of the targeting tab in Ads Manager and into those three things, and that's exactly where they should be.
Key takeaways
- Go broad and let Advantage+ work — narrow manual audiences just starve the model.
- Creative volume and angles are your real targeting lever now; ship many, learn from the winners.
- Send qualified-conversion and value signals back via the Conversions API, or the AI optimises toward cheap junk.
Frequently asked questions
Should I use broad targeting or detailed audiences on Meta now?
Broad, in almost all cases. Meta's Andromeda engine and Advantage+ find your buyer better when you give them room to look, and narrow manual audiences (a handful of interests, tight age ranges, stacked exclusions) just starve the model of the search space and signal it needs. The audience is now an output the system discovers, not an input you dictate — your job is to give it room and then feed it the right signals to recognise the right people.
Why is creative called 'the new targeting'?
Because when the AI decides who sees your ad, your creative is what decides who responds — and the response is what trains the targeting. A scroll-stopping, problem-aware hook self-selects its audience: the people it speaks to stop and convert, Andromeda learns from that, and goes to find more of them. So you don't tell Meta who to find; you show it, one ad at a time. That's why volume and variety of creative angles now matters more than the polish of any single ad.
How many creative variations should I run on Meta?
Far more than the old 'two polished films' approach — we typically ship 8–12 distinct angles per round: different hooks, pain points and formats, much of it raw, UGC-style and phone-shot, because that's what performs now. Each angle is a probe into a different segment, and the winners reveal who your real audience is and sharpen the next round of creative and the offer. Variety of angles, not production value, is what feeds the machine.
What is the Conversions API and do I need it?
The Conversions API (CAPI) sends conversion events to Meta server-side rather than relying on the browser pixel alone, so you lose far fewer events to browser restrictions — and it lets you send back which leads were actually qualified or paid, ideally with a value attached. You need it because Advantage+ is only as good as the signal you feed it: without it, Meta optimises toward whatever's cheapest to collect, which is usually low-quality leads that never buy.
Does Advantage+ automation replace the need for a good offer and funnel?
No — and that's where the durable advantage now lives. Andromeda won't save a weak offer, and Advantage+ won't rescue a page that loads in nine seconds or a business that takes two days to call a hot lead back. The AI got dramatically better at one narrow job: matching an ad to a person likely to respond. Everything on either side of that click — the offer, the landing page, the follow-up — is still yours and still manual, and it's what separates accounts now that everyone has the same automation button.
Tools & next steps
Put this into practice, go deeper, or see how we'd do it for you.
Written by
Mr. Chandan Kumar
Founder & Performance Marketing Director, Global Info Edge
Founder of Global Info Edge and a performance-marketing specialist with 17+ years in the digital marketing world — Google & Meta ads, conversion funnels and growth.
View full profile