Why broad audiences are dead on Meta in 2026

I've spent 25 years buying media. Enough to have watched three eras die. Demographic targeting went first, murdered by lookalikes. Then lookalikes got hollowed out by App Tracking Transparency. And on January 15, 2026, Meta quietly flipped another switch: dozens of detailed interest categories were retired, and every ad set still running on them simply stopped delivering. No email. No fanfare. Just silence in the dashboard.

Half the tier-1 brand accounts I know woke up that morning with ad sets that had been pacing fine on Friday and were frozen on Monday. The ones who didn't wake up with frozen ad sets were already on Advantage+. Meta didn't tell anyone broad audiences were dead. It just made them mandatory.

What actually happened on January 15

Meta removed a large chunk of the granular interest universe that media buyers had been stacking for a decade — the cheat codes of "interested in X AND behaved like Y AND lives within Z miles of a Whole Foods." Campaigns built before June 23, 2025 had been allowed to coast on the old options; on January 15, 2026, the coasting ended. Ad sets relying on the retired interests stopped delivering outright.

Meta's public justification is the same one they've been building toward for three years: Advantage+ Audience — the AI-driven targeting system that takes your conversion event and your creative and decides who sees the ad — is simply better. Their own numbers claim Advantage+ campaigns deliver around a 22% increase in ROAS compared to manual setups. You can question the benchmark (I do), but you can't argue with the lever: it's the one Meta is pulling, so that's the game.

"Broad" is a misleading word now

Here's the quiet thing nobody is saying out loud. What used to be called a "broad audience" — no interests, no behaviors, just age and geo — was already doing most of the lifting on well-trained accounts by late 2025. The real shift isn't that broad is winning. The real shift is that targeting has been absorbed into the model, and your input to that model is no longer a checkbox tree. It's your creative.

When a media buyer used to write "women 28-45, interested in premium skincare, travel to luxury hotels" in the targeting panel, they were giving Meta a human-legible description of their ICP. In 2026 you don't give Meta that description in the targeting panel. You give it to Meta in the creative itself: the casting, the setting, the voiceover, the copy, the hook, the pacing. The model reads who reacts and who doesn't, and targets from there. The creative is the brief, the creative is the audience, the creative is the budget-allocator.

The creative is the targeting. Everything else is plumbing.

Where the work actually moved

If the model is doing the targeting, the job of a paid media team in 2026 is not to out-target Meta. That's a fight you lose every month. The job is to feed the model better than the competition does. Three things feed the model: signal quality, creative volume, and cadence.

Signal quality — the part most accounts quietly fail

A Conversions API that isn't firing clean server-side events is a model eating junk food. When we audit a new tier-1 account, seven out of ten have duplicate events, missing deduplication keys, or a Purchase event firing on the thank-you page and somehow also on a view-item. The model trains on that noise for weeks before anyone notices. Before you touch creative, fix the plumbing: CAPI health check, event dedupe, proper hashing, a clean taxonomy between ad account and CRM. Unglamorous work. Pays for itself in a week.

Creative volume — the brutal new floor

Meta has publicly shared that Advantage+ Shopping Campaigns running 20 or more creatives delivered a 29% lower median incremental cost per purchase versus campaigns with fewer assets. Let that land for a second. Twenty creatives per campaign. Not per quarter. Per active concept. The agencies still shipping four polished hero cuts per month are buying the equivalent of Facebook ads from 2018 at 2026 prices.

This is where AI stops being a buzzword and becomes the unfair advantage of whoever actually uses it. We produce the first-frame hooks in Flux, the motion variants in Seedance and Veo, the voiceovers cloned from approved talent, the static alt-cuts in Midjourney, and the copy variants in a small in-house model trained on the brand. Twenty cuts from one concept in a day, not a month. A brief that would have been $15,000 and three weeks in 2022 is a morning and a coffee now. That is the Creategic floor.

Cadence — 7 / 14 / 21

The framework we run on tier-1 accounts is simple enough to tape to the wall. Seven days to learn: every new creative gets a $50-150/day guaranteed burn to escape learning phase. Fourteen days to scale: the winners get duplicated into a scaling campaign with lifted budget. Twenty-one days to kill: anything that hasn't found its audience by day 21 is paused, dissected, and recycled into the next testing batch as ingredients. No concept is precious. The only KPI that survives is incremental CPA against the account's blended benchmark.

Then vs now — the side-by-side

Dimension	Meta 2021	Meta 2026
Main lever	Audience targeting	Creative volume & quality
Who controls delivery	Media buyer	Meta's Advantage+ model
Unit of testing	Ad set (audience)	Creative variant
Healthy variants / week	2–4	15–20
Signal backbone	Pixel + 3rd-party data	Conversions API server events
Fatal mistake	Audience too broad	Not enough creative to feed the model

What about budget allocation across platforms?

Short answer: the same logic. Google's Performance Max and TikTok's Smart+ are chasing the same endgame — absorbed targeting, creative-as-input, server-side conversion signal. If you run a multi-platform brand account, allocate to the platform that has the most complete signal stack and the most creative you can hand it. The old instinct of "spread budget by audience strength" is inverted: allocate by creative readiness. A perfectly tuned Meta account is worth nothing if your team can only ship four video cuts a month; a scrappy TikTok account with 40 weekly variants will outperform it on relevant verticals.

In 2026 your paid media P&L is decided in the creative kitchen, not the targeting panel. If your agency is still selling you on audience optimization, they're selling a lever Meta already owns.

What we actually do at Cipion

For the brand accounts we take on, the first two weeks are plumbing — CAPI audit, event hygiene, catalog sanity, duplicate campaign cleanup, nuking any ad set still trying to hand-target. Week three we restructure into Advantage+ with a small number of concept-level test campaigns. From week four, we ship 15–20 variants per active concept per week from the AI production stack, and run the 7 / 14 / 21 cadence. Diego is on every account personally. No junior team disguised as a senior one. No 40-page media plan nobody reads. The work is creative and signal, in that order, every week.