Topics · How it works
Ad ranking
Contents
Ad ranking on Instagram compares advertisements to decide which to show and in what order.12 Advertisers choose eligible audiences, but ranking delivers ads to only a subset of those people; a Meta patent separately proposes using bids to raise sponsored suggestions’ scores.23 Meta’s fairness research proposes making the demographic makeup of people receiving ads more closely resemble the eligible audience, rather than treating targeting as guaranteed reach.2 Its sparse-data research explores connecting people’s interests with advertiser and product information to support future ranking systems.4 In August 2026, Engineering at Meta described a two-stage design that combines previously processed behavior histories with fresh ad information, allowing ranking models to grow without proportional increases in serving resources.1
How it has changed
- 2023
May
Meta’s AI blog announced Meta Lattice, moving from separate prediction models toward a common model design used across different ad placements and advertiser goals to balance performance with computing efficiency.5
- 2026
August
Engineering at Meta described a two-stage ranking design that processes long behavior histories ahead of time, stores numerical representations of their patterns, and combines those stored representations with fresh information about candidate ads when ranking them, allowing the system to grow without proportional increases in the resources needed to serve ads.1
Research on ranking when response data is sparse
Meta research and patents. These may not describe how Instagram works today.
Meta’s July 2026 research explores ad ranking with little interaction data through numerical representations connecting people’s inferred interests with advertisers, products and services.4 Meta’s research combines interaction information with advertiser and product content, allowing it to learn about products or services that have rarely or never appeared in those interactions.4 Meta’s research describes these representations as a potential foundation for specialized ranking systems.4
Audience eligibility does not guarantee delivery
Meta research and patents. These may not describe how Instagram works today.
Meta’s June 2023 research examines who actually receives advertisements through ad ranking on Facebook or Instagram.2 Meta’s research distinguishes the audience allowed by an advertiser’s targeting choices from the smaller group that receives ads when opportunities to show them arise.2 Meta’s research compares the demographic makeup of those two groups and treats a substantial difference as a possible sign of unintended unfairness in ranking.2 Meta’s research proposes constraining delivery so that the demographic makeup of people receiving ads more closely resembles that of the eligible audience.2
See also
References
- 1From User Sequences to Scaling Laws: A Multi-Stage Architecture for Meta’s Ads Ranking
- 2Towards Fairness in Personalized Ads Using Impression Variance Aware Reinforcement Learning
- 3Sponsored recommendation in a social networking system
- 4Exploring Hierarchical Interest Representation For Meta Ads Deep Funnel Optimization
- 5New AI advancements drive Meta’s ads system performance and efficiency