The main change of 2026: broad audiences outperform narrow ones. Meta's AI algorithm finds buyers within a broad audience more accurately than manual interest targeting, so narrow targeting now often gets in the way. Start with a broad audience (gender + age + location) or Advantage+ Audience. Build lookalikes from your best customers (top 25% by LTV), not all leads, for a 20-40% lower CPL; use 1-3% on small budgets and 5-10% on large ones. Retargeting follows a warmth ladder with a different offer for each level (ROAS 3-6 times higher than cold). Don't split into 15 small ad sets: the algorithm needs 50 conversions a week for each. Exclusions save 5-15% of the budget. The rule: fewer structures, more data, more accurate optimization.
Meta targeting has changed radically in the last 3 years. What worked in 2022 (detailed interests, narrow segments, 20 ad sets to compare) is now an anti-pattern. There's one reason: the algorithm has become smarter than the marketer at finding buyers. In our experience across 300+ projects, those still “setting interests by hand” lose to those who've learned to feed the algorithm the right data. Here's how audiences work in 2026.
The big shift: broad has beaten narrow
Just a couple of years ago the logic was: the more precise the targeting, the better. Now it's the opposite. Meta's AI algorithm finds the right people within a broad audience better than you can with manual targeting.
Why this happened: Meta has accumulated so much data on behavior and intent that it predicts purchase readiness more accurately than any interests you set. When you narrow an audience with interests, you cut the algorithm off from some of the buyers it would have found on its own.
| Approach | When it worked | Status in 2026 |
|---|---|---|
| Narrow interest targeting | 2018-2022 | More often hinders the algorithm |
| Broad (gender + age + location) | Gaining ground since 2023 | The main choice for cold traffic |
| Advantage+ Audience | Launched in 2023 | Where the whole platform is heading |
| Splitting into 15+ ad sets | 2019-2021 | An anti-pattern that chokes learning |
In practice: for cold acquisition, start with a broad audience or Advantage+ Audience. In our data, this delivers a 15-30% lower CPL than narrow targeting on budgets from $30/day.
When narrow audiences are still needed
Broad isn't dogma. Narrow targeting makes sense in two cases:
- A very specific niche with a small audience: premium B2B, a niche hobby, a rare profession
- Retargeting, where you know the segment exactly and work with a warm audience
In all other cases (mass market, e-commerce, services, online courses), broad wins.
Lookalikes: source quality matters more than percentage
A 1% lookalike is the audience most similar to the source (1% of the country's population, as close to your customers as possible). 5-10% is broader and colder.
The percentage rule:
- Small budgets (up to $2,000/month): 1-3%, higher precision
- Large budgets (from $5,000/month): 5-10%, because 1% burns out quickly
But the source matters more than the percentage. A lookalike of all leads = average results. A lookalike of buyers with a high LTV (top 25% by order value) = a 20-40% lower CPL. Build lookalikes from your BEST customers, not everyone.
The minimum source size is 100 people; the optimum is 1,000-5,000. The cleaner and more valuable the source, the more accurately the algorithm finds similar people.
Advantage+ Audience: the happy medium
Advantage+ Audience is a mode where you don't define the audience rigidly but give the algorithm suggestions on interests, age and location. It uses them as a guide but can go beyond them if it finds buyers outside.
It's a middle ground between strict manual targeting and a fully broad audience.
When to use it:
- Broad brings traffic that's too expensive or off-target, but strict interests choke the algorithm
- At launch, to give direction without restricting
- For e-commerce and mass-market products
In our data, Advantage+ Audience in 2026 matches or beats manual targeting in 70% of cases, and it's the direction Meta is taking the whole platform.
“The most common argument with new clients is that they want to ‘set up a precise interest-based audience’, because that's how it's always been done. We have to prove it with numbers: we run two identical campaigns, one on narrow interests, the other broad. Two weeks later, the broad one almost always delivers cheaper leads. People don't believe that ‘showing it to everyone’ works better than ‘showing it to interested people’. But in 2026 the algorithm knows more about purchase readiness than any set of interest checkboxes.” — Natalia Maslakova, founder of MAS Project
Custom audiences: the foundation of retargeting
Custom audiences are the foundation for both retargeting and lookalikes. What to collect:
- Customer lists: upload emails/phone numbers (with consent to processing)
- Website visitors: 30 / 90 / 180 days by depth
- Instagram/Facebook engagers: likes, comments, saves, profile visits
- Video viewers at 25% / 50% / 75%, by depth of interest
- Funnel actions: added to cart, started checkout, submitted a request
The richer your set of custom audiences, the more accurate both the retargeting and the lookalikes you build from them.
The retargeting ladder: a different offer for each level
Retargeting is built as a warmth ladder: the hotter the audience, the more aggressive the offer:
- Website visitors over 30/90 days: a product reminder, a soft offer
- Product/service viewers: a specific offer for that product
- Added to cart but didn't buy: a promo code or answers to objections (delivery, warranty)
- Started checkout: the most aggressive offer, a 48-hour discount; they almost bought
- Buyers: upsells, repeat purchases, loyalty (a separate campaign, excluded from acquisition)
The rule: to cold audiences you sell the idea; for hot ones you remove the last barrier. Retargeting delivers a ROAS 3-6 times higher than cold traffic, but volume is limited: it's for closing, not your main channel.
Exclusions: the hygiene everyone skips
Mandatory exclusions in 2026:
- Existing buyers: exclude them from acquisition; they get a separate upsell campaign
- Employees and their devices, so test visits don't pollute pixel data
- Recent converters (a request in the last 7-14 days): the sales team is already working with them
- Overlaps in retargeting: exclude the hot level from the cold one so audiences don't overlap
Proper exclusions save 5-15% of the budget and prevent impressions from overlapping between campaigns. We find this gap in 60% of the accounts we audit.
The biggest mistake of 2026: over-fragmentation
People used to create 15-20 narrow interest-based ad sets and compare them. Now that kills results: the algorithm needs at least 50 conversions per ad set per week to learn. On a small audience with a $10 budget it never gets there and stays stuck in “Learning limited” forever.
The 2026 rule: 2-3 broad campaigns with lots of data beat 15 narrow ones at $10. Consolidate: merge similar audiences and put more budget into fewer ad sets.
Fewer structures → more data per structure → more accurate optimization. Splitting only makes sense with genuinely different products or markets with different economics.
A working account structure for 2026
| Funnel stage | Audience | Goal |
|---|---|---|
| Cold (acquisition) | Broad / Advantage+ Audience | Maximum new leads |
| Cold (scale) | 3-5% lookalike of top customers | People like your best customers |
| Warm | Website visitors, engagers | Reminders, nurturing |
| Hot | Cart, checkout started | Remove the barrier, 48-hour offer |
| Customers | Buyers (separately) | Upsells, loyalty |
Audience setup checklist
- Set up the Conversions API and key events, and reach an EMQ of 8+
- For a cold start, a broad audience or Advantage+ Audience
- Build custom audiences: website 30/90/180, engagers, video, cart
- Build lookalikes from the top 25% of customers by LTV, not all leads
- Match the lookalike percentage to the budget: 1-3% small, 5-10% large
- Build a retargeting ladder with offers for each level
- Set up exclusions: buyers, employees, recent converters
- Don't fragment: 2-3 broad campaigns instead of 15 narrow ones
- Give each ad set at least 50 conversions a week
- Don't change audiences more than once every 1-2 weeks; let the data build up
Want properly configured audiences for your project?
We'll audit your account structure, find where budget leaks through overlaps and narrow targeting, and rebuild it for 2026 algorithms, drawing on 300+ projects. Free 30-minute consultation.
Get a free consultationFrequently asked questions
Are broad or narrow audiences better on Meta in 2026?
In most cases, broad. The AI algorithm finds buyers within a broad audience better than manual interest targeting. Start with broad (gender + age + location) or Advantage+ Audience. Narrow makes sense for specific niches and retargeting. For 80% of projects, broad delivers a 15-30% lower CPL on budgets from $30/day.
Which lookalike percentage should you choose: 1% or more?
Small budgets (up to $2,000/month): 1-3% (higher precision); large budgets (from $5,000/month): 5-10% (1% burns out quickly). But the source matters more than the percentage. A lookalike of the top 25% of customers by LTV delivers a 20-40% lower CPL than one of all leads. Minimum source size: 100 people; optimum: 1,000-5,000.
What is Advantage+ Audience, and when should you use it?
A mode where you give the algorithm suggestions on interests/age/location but it can go beyond them. A middle ground between strict targeting and broad. Use it when broad brings off-target traffic but interests choke the algorithm, at launch to set a direction, and for e-commerce. It matches or beats manual targeting in 70% of cases.
How do you build proper retargeting?
As a warmth ladder: 1) website visitors: a soft offer; 2) product viewers: a specific offer; 3) cart: a promo code or answers to objections; 4) checkout started: a 48-hour discount; 5) buyers: upsells (separately). The hotter the audience, the more aggressive the offer. ROAS is 3-6 times higher than cold, but volume is limited: it's for closing.
Should you split audiences into lots of small segments?
No, it's a 2026 anti-pattern. The algorithm needs 50 conversions per ad set per week, and on a small audience at $10 it gets stuck in “Learning limited.” 2-3 broad, high-volume campaigns beat 15 narrow ones at $10. Fewer structures, more data, more accurate optimization.
Which exclusions must you set up?
1) Existing buyers (from acquisition); 2) employees and their devices; 3) recent converters (a request in the last 7-14 days); 4) overlaps in retargeting (the hot level from the cold one). It saves 5-15% of the budget and removes overlapping impressions. Skipped in 60% of the accounts we audit.