Meta Andromeda Broke a Lot of DTC Accounts. Here's What's Still Working.

When we published our original Andromeda breakdown, we were working off the first wave of account data. With more accounts and more time behind it, the picture is a lot clearer now.

We've been watching patterns across CPG and ecommerce accounts closely. What Andromeda changed isn't subtle. It shows up in delivery data, comment thread performance, and creative burnout rates. Here's what the data actually shows now.

What the Data Actually Shows Now

Confect analyzed 3,014 eCommerce advertisers across $834 million in ad spend and 115.7 billion impressions during Andromeda's full rollout period. The topline numbers are telling.

The average decline masks a wider split. Mid-sized advertisers and accounts built around catalog-first creative improved. Large accounts running primarily static creative dropped from 8.7 to 8.0 average ROAS. Affordable price tier products saw a 35% ROAS collapse. Mid-priced product accounts actually went up.

The accounts absorbing the biggest losses share the same profile: heavy static single-image spend, thin creative variety, and no catalog feed active. The accounts winning share the opposite.

What Spam Comments Are Actually Doing to Your ROAS

This one surprised us the first time we saw it surface across accounts. Spam in your ad comment threads isn't just noise. It drags on the metric Meta's delivery optimizes hardest against: click-through rate.

When a prospect hits your ad and sees spam in the first three visible comments, they scroll past. CTR drops. And because click-through rate is a core delivery signal, that drop gets read as weak creative-audience fit, even when the creative is performing fine everywhere else.

FeedGuardians' 2026 benchmarks, drawn from 48 million moderated comments across 2,400+ brands, put the average harm rate at 18% of all ad comments, spread across scam links (34%), competitor bait (21%), toxicity (19%), bot spam (14%), and off-topic (12%). For fashion, fitness, and beauty brands that rate runs higher at 21-26%.

The fix is not manually replying to each spam comment. A reply doesn't hide anything from other users, and the spam stays right where a new visitor sees it. The right move: hide comments (hiding pulls them from public view without deleting them or notifying the commenter) and configure keyword filters in Ads Manager to auto-hide common phrases before they reach your threads. Automated moderation that hides or auto-replies at scale shows a measurable conversion lift, so this is worth systematizing rather than doing by hand.

Brands running systematic comment moderation have seen ROAS lifts north of 50% in the highest-harm categories (beauty and the like, at 26%+ harm rates). Even lower-baseline accounts around 12% harm, often with some moderation already in place, showed 14-22% ROAS improvement. Comment hygiene is now a paid performance function, not a community management task.

What Happens When You Reply to Spam Comments on Meta Ads

Replying to spam comments seems like the obvious move. You see spam, you call it out, push it down, maybe hope the platform notices. In practice, none of those outcomes happen.

When you reply to a spam comment, the spam stays fully visible to every user who encounters your ad. Your reply appears beneath it, making the thread look more active, not cleaner. To a new visitor, the spam is still the first thing they see, and your reply has done nothing to remove the CTR drag. It has only added activity to the same polluted thread.

Hiding is the default move. A hidden comment disappears from the public thread but isn't deleted: it stays visible to the person who wrote it and their friends, so they aren't tipped off, and you can reverse it. Pulling it from public view does trim your visible comment count slightly, but that's a fair trade for removing content that's actively suppressing your CTR. Deleting is more aggressive and permanent, so save it for genuinely abusive material.

For accounts seeing persistent spam, the more durable fix is keyword filters in Ads Manager under Page Settings. Add phrases that commonly appear in your spam threads, and Meta auto-hides any matching comments before they become visible. Worth filtering: crypto and investment redirect phrases, competitor brand name baiting, and generic bot patterns. For high-volume campaigns, building this filter list is one of the more underrated levers in your comment hygiene setup.

Why Catalog Ads Are Pulling Away from the Pack

Catalog ads give Andromeda's retrieval model exactly what it wants: product images, price, availability, and feed-level signals it can match to high-intent users based on their browsing behavior. Static creative gives it far less to work with.

Source: Confect.io analysis of 1 million ad creatives, $834M in spend across 3,014 advertisers.

The brands that restructured toward catalog-first in Q1 are seeing it in account-level ROAS. The ones still running primarily static single-image ads are absorbing the 17% average decline. This is one of the clearest structural format shifts we've seen in paid social in years.

Creative Fatigue Is Hitting Sooner Than It Used To

One consistent pattern across accounts: creative fatigue is arriving earlier. Andromeda cycles through an ad set's creative more aggressively as it learns. In Confect's dataset, effective ad lifespan compressed from six to eight weeks before Andromeda down to two to four weeks now.

Top performers in Confect's dataset are 38% more likely than the rest to run five or more creative variants. Not five identical formats with different colors. Five meaningfully different approaches that give Andromeda something to test and differentiate, so it can find the best creative-audience pairing for each.

The second differentiator from top performers: multiple placement formats. Single-placement campaigns see Andromeda running the same creative into the same surfaces repeatedly. Multi-placement setups give the algorithm more distribution optionality, which shows up as higher ROAS in the aggregate data.

How to Adapt: Five Steps That Are Working Right Now

What Andromeda Hasn't Changed

A few things to be clear about. Andromeda is not a performance multiplier that works regardless of fundamentals. It makes good creative perform better and accelerates the death of weak creative. It doesn't fix broken landing pages, weak offers, or attribution setups already producing bad data.

The basics are the same: product-market fit, a relevant offer, a fast-loading destination page. If your underlying ecommerce marketing strategy isn't solid, Andromeda evaluates and punishes that faster than the old system did. The penalty for ignoring fundamentals is steeper now.

If ROAS compression continues after addressing catalog format, comment hygiene, and creative depth, the next layer to look at is Meta retargeting setup and your CPG shopper marketing strategy. The algorithm is one layer of the equation.

For a full breakdown of how Andromeda changed the ads retrieval process and the structural account implications, see our original Meta Andromeda explainer. For CPG channel strategy beyond Meta, the CPG retail marketing playbook covers the full cross-channel picture.