Facebook Ads in 2026: The Rules Have Changed
Mykyta Hryhorenko
CEO & Co-founder
The Algorithm Is Not What It Was
In late 2024, Meta rebuilt its ad delivery engine from the ground up. The new system is called Andromeda, and by now it runs across all objectives and placements globally. Most advertisers felt the shift before they understood what caused it — rising CPMs, inconsistent results, campaigns that had been stable for months suddenly underperforming without any obvious explanation.
The logic of the old system was straightforward: you define the audience, Meta finds people who match it, and shows them your ad. Andromeda reverses this logic entirely. The system now starts with your creative, reads it visually and contextually, and decides which users are most likely to respond based on what the ad communicates. Your creative is no longer just the message you send to an audience you selected. It is the primary signal that determines who the audience becomes.
Since February 2026, Meta officially documented that detailed targeting inputs within Advantage+ are advisory. The algorithm will reach outside your defined interest groups whenever it predicts better performance elsewhere. Your carefully built audience stacks are suggestions, not constraints. This is not an incremental update. It is a different model of how paid social works, and campaigns built on the old logic are paying for it in wasted spend every day.
What This Means for Creative
If the algorithm targets through creative, then creative volume and diversity become the primary performance variable — more important than audience selection, placement choices, or bid strategy. Running the same hero image with five headline variations — which worked reliably in 2022 — actively hurts you now. Meta's visual recognition models treat minor variations of the same visual as a single ad entity. If your library looks repetitive to the system, it raises your CPMs as a direct penalty for what it reads as creative fatigue.
Top-performing advertisers are now running 15 to 50 genuinely different ads per ad set. Not 15 variations of one concept — 15 approaches that differ in format, hook, angle, and stage of the customer journey. Static images still drive 60 to 70% of conversions on Meta, so video-only strategies aren't the answer. The algorithm needs variety across formats: static, video, carousel, UGC-style, catalog. Each format is a different retrieval ticket in Andromeda's system.
Meta has added two new metrics to Ads Manager to help manage this. Creative Fatigue flags when your library is going stale and CPMs are about to climb. Creative Similarity scores how diverse your active ads actually are — a high score means the algorithm sees repetition and will charge you for it. There's also a format requirement that matters: 90% of Meta's ad inventory is now vertical. Ads not produced in 9:16 or 4:5 get auto-cropped, often badly. If your creative team isn't producing for vertical first, the algorithm is working against you before the auction even starts.
Targeting Controls Are Being Removed
Two significant changes happened on the targeting side in the past year, and both point in the same direction: manual control is being systematically reduced in favor of AI-driven delivery. The first change consolidates many detailed targeting interest options into broader categories. Sports-related interests are now grouped together, as are film and music genres, car models, and food preferences. Ads Manager suggests alternatives automatically, but campaigns using outdated interest selections display warning banners and will eventually stop delivering entirely.
The second removes the ability to use detailed targeting exclusions in both Ads Manager and boosted posts. Meta's internal testing data shows that campaigns running without exclusions often outperform those with them, reporting a median cost per conversion 22.6% lower. The reasoning is consistent with Andromeda's design: when you exclude audiences, you're restricting the surfaces the algorithm can use to find conversions, and the algorithm's view of who will convert is more accurate than your manual assumptions.
The practical shift for most advertisers is to stop investing time in interest architecture and start investing it in first-party data infrastructure. Custom audiences built from your own sources — email lists, website visitors, customers, engaged social followers — are now the primary precision tool. Lookalike audiences built from those lists remain effective. What's no longer reliable is layered interest targeting built on the assumption that Facebook knows who your customer is because they liked a certain brand page three years ago.
Advantage+ Is Now the Default
Meta restructured campaign creation to make Advantage+ the default path for new campaigns, and this reflects the direction of the entire product roadmap. Advantage+ Shopping has been renamed Advantage+ Sales. Several manual controls have been simplified or removed. The underlying logic is the same as Andromeda: feed the system quality inputs — strong creative, clean conversion data, solid landing pages — and let it optimize distribution and delivery. Trying to outmaneuver the automation with tight placement restrictions or narrow audience caps consistently produces worse results than letting the system work.
The consolidated account structure that performs best under Andromeda looks nothing like what most advertisers built in 2022. One main Advantage+ campaign with your full creative library, broad or no audience targeting, and all placements enabled. A secondary retargeting campaign at reduced budget. That's the core. Brands testing this simplified structure have reported 15 to 17% lifts in conversions from the structural change alone — before touching creative.
The hardest adjustment is patience. Andromeda needs 7 to 14 days to move through the learning phase and begin optimizing meaningfully. Pulling campaigns after 48 hours of weak early results is the single most common reason performance data looks bad. Frequent edits reset the learning phase entirely. In the Andromeda era, stability is a strategy — not laziness.
The Measurement Shift
Pixel-only tracking is no longer a viable foundation. The Conversions API sends server-side events that survive browser privacy restrictions, iOS tracking changes, and ad blockers. Running Pixel and CAPI together with proper deduplication is the current baseline, not an advanced configuration. Meta's Events Manager shows you an Event Match Quality score — below 7 means the system is struggling to connect your conversion events to real users, which directly degrades Andromeda's ability to learn and optimize. Low match quality means you're paying for an algorithm that's navigating blind.
Two attribution changes in early 2026 are also affecting how numbers appear in dashboards. Click-through attribution now counts only actual link clicks, not all post interactions. The engaged-view threshold for video dropped from 10 seconds to 5. If your reported conversions look lower than expected, check whether these attribution model changes account for the gap before drawing conclusions about performance. For most accounts, 7-day click attribution remains the right setting. For impulse-purchase products with fast decision cycles, 1-day click gives a cleaner signal.
The Shift That Actually Matters
Strip away the feature names and the product updates, and the through-line is clear: Meta has moved decisively from a system that rewarded audience precision to one that rewards creative quality and data integrity. The advertisers who are struggling right now are largely running the same playbook they built three years ago. The ones doing well have accepted a different role — less audience architect, more creative director and signal supplier.
The tools for doing this well are better than they've ever been. Andromeda is genuinely more capable than the system it replaced. But it only performs at the level of what you feed it. Strong creative, clean conversion data, stable campaign structure, and patience during the learning phase — these aren't new principles. They're the same fundamentals that have always defined good paid social. The algorithm has just made ignoring them significantly more expensive.
