Stackalone
Stackalone
Books
BlogMissionFAQContact
Stackalone

Links

  • Books
  • Mission
  • FAQ
  • Contact

Contact

  • hello@stackalone.com
  • 651 N Broad St, Suite 201
    Middletown, Delaware 19709, United States
Privacy PolicyTerms of ServiceRefund Policy

© 2026 Stackalone. All rights reserved.

← Meta Blog
Meta·Update·Jan 20, 2026

Facebook Reels RecSys — User Feedback Is Rewriting AI Training

Meta is retraining the Reels recommendation AI faster using user feedback. How "Not Interested" and dwell time actually shape your ad delivery.

Facebook Reels RecSys AI Model
Facebook Reels RecSys AI Model

"User feedback changes ad delivery"

Meta Engineering blog, January 2026. Details on how the Facebook Reels recommendation AI learns from user feedback are now public. The advertiser takeaway: the actions users take under your ad determine its next impression.

Source: Meta Engineering — Adapting the Facebook Reels RecSys AI Model

Signals the Reels AI learns from

Meta's official taxonomy of user feedback signals:

Explicit feedback:

  • Like / Reaction
  • Comment
  • Share
  • Save
  • Follow
  • "Not Interested" button
  • "Hide Ad" click

Implicit feedback:

  • Dwell time
  • Completion rate
  • Unmute
  • Profile visit
  • Exit timing mid-play

Explicit feedback is a strong signal, implicit is a volume signal. The AI combines both to decide the next delivery.

What this means for advertisers in practice

1. The cost of a "Not Interested" click

When users repeatedly tap Not Interested or Hide Ad on your creative:

  • Your ads stop serving to that user
  • Relevance drops for similar user clusters too
  • In severe cases, overall ad delivery is restricted

So the "Hide Ad" advertisers fear is literally the core signal Meta's AI uses for ad quality. It directly hits your Ad Quality score.

2. Dwell time matters more than you'd think

Reels ad dwell time (average watch time):

  • High first-3-second drop-off → flagged as low relevance
  • High completion rate → strong relevance and quality scoring

Practical impact: the first 3 seconds of a Reels hook isn't a cosmetic issue — it's an AI training signal. The mobile 3-second rule just got re-validated.

3. Minimize negative feedback

Avoid in creative:

  • Over-edited "ad-looking" polish — triggers Not Interested
  • Irrelevant targeting — users perceive "this topic shouldn't be here" and hit Hide
  • High repeat frequency — frequency 5+ causes annoyance

So what should you do?

Checklist:

  • [ ] Check Ad Quality score (Events Manager) — low = negative feedback accumulation likely
  • [ ] Cap Reels ad frequency (keep under 3)
  • [ ] 50%+ of your creatives should have high completion rates
  • [ ] Targeting isn't so narrow that users get annoyed

Recommended A/B observation:

  • Monitor dwell time in the first 48 hours after launching new Reels ads
  • If early Hide Ad rate is high, pull it early
  • Creatives with heavy positive feedback can scale

New success conditions for Reels ads

Before: CTR and CPA-centric → Now: dwell time + CTR + feedback balance.

  • High CTR but short dwell time → weak learning
  • Long dwell time but no conversions → wasted spend
  • Dwell + positive feedback + conversions = long-term winner

Creative direction

Traits shared by Reels creatives that have gained ground since the RecSys update:

  • Entertainment tone — it's an ad, but fun
  • UGC style — feels creator-made
  • Short storytelling — clear beginning-middle-end
  • Natural audio — smooth music and SFX (account for sound-off too)

Avoid:

  • Amateur PPT-style slides — users bounce fast
  • Hard-sell tone ("Buy now!") — reads as pushy
  • 15+ second single shot — boring

Reels creative planning, hook formulas, and AI creatives are covered in Meta Ads Book 3.

Creative Is Performance

The Ads That Get Clicked Are Different

Stackalone

Covered in depth in
The Ads That Get Clicked Are Different
Creative Is Performance
→
Tags
#reels#recsys#feedback-loop#ranking
← Previous
AI-Written Ad Copy — 10 Policy Violation Patterns to Catch
Jan 5, 2026
Next →
New Open-Source LLMs Released — What Advertisers Should Care About
Feb 2, 2026

Related Topics

  • Update · Mar 23, 2026
    Friend Bubbles — What Reels' "Friend-Based" Discovery Means for Ads
    #reels#social-discovery#friend-signals
  • Update · Mar 20, 2026
    REA — An AI Agent Is Replacing Meta's Ranking Engineers
    #rea#algorithm#ai-agent
  • Update · Jun 10, 2025
    Instagram Explore Ranking — Understanding the Context Behind Ad Placements
    #instagram#explore#ranking
  • Update · Jun 2, 2025
    Instagram's 1000-Model Era — How Fast the Recommender Learns
    #instagram#ml-models#algorithm