How Meta's Algorithm Changed — and What It Means for Your Ad Budget
By Diana Nekrasova · SciGrowth · October 2026
If your Meta campaigns feel like a different game than they did two years ago, you're not wrong. The rules changed at the infrastructure level. Meta rebuilt how ads are retrieved, ranked, and delivered — and most DTC brands are still running a 2023 playbook. This post breaks down what actually changed, what the data says, and what I'd do with your budget right now.
What is Meta Andromeda, and why does it matter for your account?
Meta Andromeda is Meta's ads retrieval system — the infrastructure layer that narrows a very large pool of eligible ads before those ads move into ranking and auction. Meta first described Andromeda publicly in December 2024, then expanded the wider ads recommendation story with GEM in November 2025 and the Adaptive Ranking Model in March 2026.
Here's what that means in plain English: Meta runs ad selection in two stages — retrieval and ranking. Retrieval narrows the universe of eligible ads from tens of millions down to a few thousand candidates, fast enough that the auction can finish before a feed item loads. Ranking then picks the final winners from that shortlist using larger, slower models.
Before Andromeda, the system ranked broadly. Now it pre-filters using your creative as the primary signal. It evaluates your creative first, predicts which users are most likely to engage based on behavioral signals, and serves accordingly. Advertiser-defined audiences are now suggestions, not gates.
Andromeda narrows tens of millions of eligible ads to a few thousand candidates per impression (per Meta's December 2024 engineering post), and Meta reported a +6% recall improvement and +8% ads quality lift in its engineering documentation. Sitting above that is GEM (Generative Ads Recommendation Model) — as of 2026, the ranking layer is GEM, a foundation model trained at large-language-model scale that Meta credits with up to a 5% conversion lift on Instagram. And in Q2 2026, Meta launched the Generative Recommender, an LLM-powered ad ranking system. Meta's Q2 2026 earnings call attributed an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook to its advanced user-understanding models combined with GEM — a broader result than the Generative Recommender alone, which separately showed a 1% lift in Instagram app-event conversions in early pilots. Advantage+ has now surpassed a $75 billion annual run rate.
The bottom line: instead of relying on advertiser-defined audiences — interests, demographics, lookalikes — the system analyzes your ad creative to predict user match. Your creative now functions as your targeting. Broad audience settings combined with high-quality, diverse creative outperform narrow audience targeting in most account types.
Did Meta really kill manual targeting — or is that overstated?
It's not fully dead, but it's been structurally demoted. As of June 2025, Meta began consolidating many detailed interest categories into broader groupings. Specific interests such as EDM fans, SUVs, and vegan food were merged into broad groups. Detailed targeting exclusions were also removed entirely.
In Q1 2026, Meta mandated the migration of legacy campaigns into Advantage+. Because Advantage+ is even more dependent on conversion signal quality than standard campaigns, CAPI data quality directly affects how the AI allocates spend.
On October 8, 2025, Meta deprecated the legacy ASC and AAC APIs. As of Marketing API version 25.0, released in Q1 2026, new legacy campaigns can no longer be created on any API version. Existing campaigns continue to run, but Advantage+ is now the only forward path.
What I tell clients: stop lamenting the loss of interest stacks. Meta has removed many detailed targeting options over time, and its Advantage+ audience settings treat your inputs as suggestions rather than hard limits. In most accounts, broad audiences with minimal restrictions now perform as well as or better than narrow interest targeting, because the algorithm can find buyers you would never have defined. One meaningful control you did regain: Meta launched exclusion-only custom audiences in August 2026 — a dedicated audience type that permanently suppresses specific groups across Facebook, Instagram, and Meta Audience Network, closing the targeting gap that opened when Meta removed detailed targeting.
Do the Advantage+ numbers actually justify switching — or is this just Meta pushing its own product?
Fair skepticism. Here's what the third-party data actually shows, not just Meta's marketing.
In 2024, ASC accounted for 34% of ecommerce conversion spend on Meta. By 2025 that number hit 62%. The growth is happening because the performance is real: according to benchmark data from AdAmigo/MHI Media (drawn from a managed-spend dataset of ecommerce accounts), ASC averages a 4.52x ROAS compared to 3.70x for equivalent manual campaigns — a 22% relative improvement in that benchmark dataset. Results vary by vertical, catalog size, creative volume, and account maturity, and the absolute figures skew toward larger, more established advertisers. Meta's own data shows ASC cuts cost per purchase by 17% on average versus manual setup.
Third-party managed spend data backs this up directionally. Based on $50M+ in ad spend across 200+ accounts, e-commerce brands running Advantage+ Shopping see an average 17% lower CPA compared to manually structured campaigns. The caveat is that ASC works best with at least 50 conversions per week — below that threshold, manual campaigns often outperform.
There's also a newer threshold worth knowing: Value Optimization and pLTV bidding inside Advantage+ Sales campaigns can now stabilize with as few as 30–50 high-value purchase events per week; the standard learning phase threshold of 50 optimization events per week is unchanged.
My honest take: if you're spending under ~$5,000/month and running fewer than 30 weekly conversions, a well-structured manual campaign may still outperform. For newer stores with limited purchase data, manual campaigns with defined audiences often perform better initially. The recommended approach is testing ASC with 20–30% of total budget while maintaining proven manual campaigns. Above that threshold, ASC is genuinely hard to beat on efficiency.
If you want a second opinion on your specific account structure, SciGrowth's Meta Ads consulting team audits account setups and identifies where Advantage+ is being under-fed or misconfigured.
Creative is the new targeting — so what does your creative operation actually need to look like?
This is the shift most DTC brands are slowest to operationalize. Practitioners running accounts structured for Andromeda rather than fighting it frequently report 20% to 35% higher ROAS than legacy account structures, though results vary by category and execution. Across the wider 2026 benchmark set, creative quality is frequently cited as accounting for more than half of Meta ad performance.
The retrieval system has a specific mechanical quirk you need to understand: Andromeda doesn't look at your ad count — it looks at conceptual uniqueness. Meta assigns each creative an internal identifier called an Entity ID based on its visual fingerprint. If you upload 30 ads that share the same template, same background, same visual structure with different text overlays, Andromeda collapses them into one Entity ID.
That's the creative similarity trap in practice. Practitioners report that scores above 60% on Meta's Creative Similarity Score can trigger retrieval suppression, and recommend keeping the index below 40% across active assets so the system treats each ad as a distinct candidate. These thresholds are based on practitioner observation and have not been independently confirmed by Meta's primary engineering documentation.
Fatigue timelines have also compressed significantly. Ad fatigue windows have shrunk from 6+ weeks to 2–3 weeks under Andromeda, according to practitioner data. What I'd do: launch with 8–15 genuinely different creative assets — different angles, formats, and messaging approaches, not minor variations of one concept — and plan to add 3–5 new assets every 1–2 weeks to prevent creative fatigue.
What should your budget structure look like in Q4 2026?
Based on what's working across accounts right now, here's a starting framework — adjust for your conversion volume and catalog size:
- 60–70% to Advantage+ Sales campaigns (formerly ASC): for optimal results, allocate 60–70% of your budget to Advantage+ Shopping campaigns, 15–25% to retargeting, and 15–20% to testing new creatives.
- Conversions API is non-negotiable. Advantage+ is even more dependent on conversion signal quality than standard campaigns — CAPI data quality directly affects how the AI allocates spend. Clean server-side events are the single highest-leverage infrastructure investment right now.
- Scale budgets slowly. When scaling, increase budgets gradually by 20–30% every 3–4 days to avoid triggering a new learning phase.
- Measure beyond platform ROAS. Broad audiences and Advantage+ settings let the algorithm find buyers, so your main levers are creative, conversion data, and budget structure. Judge results with blended business metrics rather than platform ROAS alone. MER (marketing efficiency ratio) and new customer acquisition cost are cleaner signals.
One structural note: the new setup heavily leverages the Opportunity Score — an algorithmic rating of your campaign structure. Keeping Meta's recommended Advantage+ settings intact helps maintain a high score, whereas adding restrictions will pull it down. However, hitting a perfect 100 points should never be the ultimate goal. The real metric of success remains actual business outcomes — sales, qualified leads, and profitable acquisition costs.
If you want experienced eyes on your account before Q4 peak — budget allocation, CAPI audit, creative volume assessment — book a Meta Ads consulting engagement with SciGrowth. We run real accounts, not just strategy decks.
FAQ
- What is Meta Andromeda in plain English?
- Andromeda is Meta's ad retrieval engine — the system that filters tens of millions of eligible ads down to a few thousand candidates per impression before ranking and the auction even begin. It uses your creative content as the primary signal to decide which ads are eligible to compete, replacing the older model where advertiser-defined audience segments did that filtering work.
- Do I still need a Conversions API (CAPI) setup if I'm running Advantage+ campaigns?
- Yes — arguably more than ever. Advantage+ campaigns are more reliant on clean, deduplicated conversion signals than legacy manual campaigns because the algorithm uses that data to allocate spend across audiences, placements, and creatives automatically. A misconfigured or missing CAPI connection means the system is flying partially blind, which degrades both efficiency and learning speed. Check your Event Match Quality score in Events Manager; anything below 6.0 out of 10 warrants attention.
- How many conversions per week do I need before Advantage+ campaigns make sense?
- The general practitioner threshold is 50 weekly purchase conversions for standard Advantage+ Sales campaigns to exit the learning phase reliably. For Value Optimization and pLTV bidding, Meta updated the floor to 30–50 high-value events per week in 2026. Below 25–30 weekly conversions, a well-structured manual campaign with a defined warm audience often produces more predictable results while you build volume.
- What is the Creative Similarity Score, and why does it matter?
- The Creative Similarity Score is a metric in Meta Ads Manager that measures how visually and conceptually similar your active ad assets are to each other. Under Andromeda's retrieval logic, ads that are too similar get collapsed into a single Entity ID — meaning 15 ads that look nearly identical compete as if they were one ad. Practitioners recommend keeping your similarity score below 40% to ensure each asset is treated as a genuinely distinct candidate in the retrieval stage. The practical implication: creative diversity is a performance lever, not just a creative preference.
- Is it still worth running any manual Meta campaigns in 2026?
- In some cases, yes. Manual campaigns with warm custom audiences (recent site visitors, email lists, purchasers) can still outperform Advantage+ for retargeting at lower budget levels, because the signal pool is small and well-defined enough that the algorithm's exploration overhead isn't warranted. The recommended structure for most DTC brands right now is Advantage+ Sales as the primary prospecting vehicle, with 15–25% of budget reserved for manual retargeting campaigns and 15–20% for new creative testing.
- My ROAS dropped after migrating to Advantage+. What should I check first?
- Three things, in order: (1) Signal quality — check your CAPI Event Match Quality and deduplication rate. Poor signals make the algorithm bid on the wrong people. (2) Creative volume and diversity — if you migrated with fewer than 8 meaningfully different creative assets, the system doesn't have enough to work with. (3) Attribution window alignment — Advantage+ campaigns often show lower click-attributed ROAS than manual campaigns because the algorithm serves more across Reels and placements where view-through conversions matter. Compare on a 7-day click / 1-day view window consistently before drawing conclusions.
Sources: