Research Report: How the Instagram Algorithm Works
(2023–2026)
Prepared for: Drishan Puri
Prepared by: ChatGPT (GPT-5 Thinking mini)
Date: February 18, 2026
Executive Summary
This report synthesizes official statements from Instagram/Meta, academic analyses of recommender
systems, and industry reporting (2023–2026) to explain how Instagram's ranking algorithms decide
what users see in Feed, Stories, Reels, and Explore. Key findings: (1) Instagram uses a family of
machine learning recommender systems that weigh many signals (user activity, content information,
relationships, and engagement metrics); (2) short-form video (Reels) and watch time are prioritized as
of 2024–2026; (3) Instagram has introduced user-facing controls to tune recommendations; and (4) the
platform's optimization tradeoffs raise ethical concerns about attention, youth safety, and transparency.
Citations for primary claims appear in the References section.
Contents
1. Introduction 1
2. Methodology 2
3. Architecture & Signals 3
4. Ranking by Surface (Feed, Reels, Explore, Stories) 4
5. Recent product changes (2023–2026) 5
6. User controls & transparency 6
7. Ethical, societal, and business implications 7
8. Recommendations for creators & researchers 8
9. Limitations 9
10. References 10
1. Introduction
This report compiles public-facing explanations from Instagram (Meta), academic literature on
recommender systems, and investigative reporting to describe how Instagram's algorithms determine
content ranking. The goal is to provide a clear, referenceable overview for creators, product teams, and
researchers.
2. Methodology
Sources: official Instagram/Meta blog posts and creators pages, peer-reviewed and preprint research
on recommender systems and platform effects, and recent technology reporting. We prioritized primary
Meta statements and peer-reviewed or institutional analyses where available. Key web sources cited:
Instagram's own ranking explainer, creators resources, a 2024 independent analysis 'Debunking
Instagram's Algorithm', Columbia/academic overviews on recommender systems, and recent reporting
on 'Your Algorithm' user controls.
3. Architecture & Signals (high-level)
Instagram operates as a set of specialized recommender systems (often separate models for Feed,
Reels, Explore, Stories). Common signal categories used across surfaces include: • User activity (what
you like, save, share, comment, and watch). • Content information (captions, hashtags, audio, video
frames, visual features). • Relationships (accounts you follow, people you message or search for). •
Popularity & engagement (speed and volume of likes/comments/shares; watch time for videos). •
Recency and freshness (newer posts are favored for Feed/Stories; relevance vs strict chronology varies
by surface). • Device/location & session context (session length, device type, locale can modulate
ranking). These signals are combined by machine-learned ranking models that predict objectives such
as probability of engagement, watch time, or retention. (See Instagram's own explainer and creator
guidance for similar signal categories.).
4. Ranking by Surface
Feed & Stories: Historically weighted toward 'relationship' signals — content from accounts you interact
with more appears higher. Recency remains important but is not strictly chronological. Engagement
signals (likes/comments) and direct interactions (DMs, saves) increase the likelihood of seeing content
from that account. (Instagram official explanation). [Ref: turn0search1]
Reels: Since Instagram's strategic pivot to short-form video, Reels uses watch time, completion rate,
and shares heavily. Instagram has stated watch time and meaningful interactions are top factors;
industry analyses (2024–2026) emphasize watch time, shares per reach, and likes per reach as core
metrics for amplification. Reels' recommendation net is broader (it surfaces content from accounts you
don't follow more commonly than Feed). [Refs: turn0search1, turn0search15]
Explore: A discovery surface driven by content-level similarity and collaborative filtering; ranking
emphasizes content information and early engagement signals to surface trending content. Instagram
has described 4 main signals for Explore: info about the post, your activity, creator, and popularity. [Ref:
turn0search1]
5. Recent product changes (2023–2026)
• 'Your Algorithm' and user-facing tuning: Instagram introduced tools that allow users to view and edit
the topics it thinks they like for Reels and other surfaces, aiming to give users control over
recommendations. (Tech reporting May–Dec 2025/2026). [Ref: turn0news35] • Prioritization of
short-form video & watch-time signals: from 2023 onward Meta doubled down on Reels as the primary
growth surface; many 2024–2026 guides and Meta statements highlight watch time and shares as key
ranking signals for reach. [Refs: turn0search1, turn0search15] • Safety & teen protections: Meta rolled
out PG-13 style filters and other teen safeguards in 2025–2026 to limit exposure to mature content for
younger users (affects what content algorithm can surface to teen accounts). [Refs: turn0news36,
turn0news37]
6. User controls & transparency
Meta has published explainer posts and creator resources discussing ranking but remains opaque on
exact model architectures and parameter weights. Recently, Instagram added controls to let users tune
interests ('Your Algorithm') and added more feedback options for Reels (like 'Not Interested').
Independent research (2024) critiques the gap between Instagram's public messaging and observed
behavior, highlighting potential 'sugarcoating' of how manipulative recommendations can be. [Refs:
turn0search10, turn0news35]
7. Ethical, societal, and business implications
• Attention optimization vs wellbeing: recommender objectives such as maximizing watch time or
engagement can create feedback loops that prioritize sensational or polarizing content. • Effects on
creators: algorithmic shifts (e.g., favoring Reels) force creators to adapt formats and may disadvantage
creators who focus on other content types. • Youth safety & regulation: new safeguards respond to
criticism and potential regulation; algorithmic changes can limit or expose teens to different content. •
Transparency & contestability: academic work urges platforms to offer clearer contestation/appeal
mechanisms for ranking decisions. [Ref: turn0search16]
8. Recommendations for creators & researchers
Creators: • Prioritize watch-time and early engagement for Reels (strong hook + full watch). •
Encourage shares and saves; both are high-value signals for reach. • Maintain consistent interactions
with your community (DMs, replies, stories) to strengthen 'relationship' signals for Feed placement.
Researchers & product teams: • Use mixed-objective evaluation (engagement + user wellbeing
metrics). • Audit recommendation outcomes across demographics and content categories. • Explore
user-facing controls and transparent explanations as mitigation strategies.
9. Limitations
This report synthesizes public materials and secondary analyses; platform internals (exact model
architectures, feature engineering pipelines, and training objectives) are proprietary and not fully
disclosed. The algorithm evolves rapidly; the reader should consult primary Meta updates for the latest
changes.
10. References (selected)
[1] Instagram — 'Ranking explained' (Instagram/Meta official blog). ([Link] ref: turn0search1) [2]
Instagram creators resources — 'Algorithms and ranking' ([Link]). ([Link] ref:
turn0search6) [3] 'Debunking Instagram's Algorithm — Sugarcoating' (research paper, 2024). ([Link]
ref: turn0search10) [4] The Verge — 'Instagram gives you more control over your Reels algorithm'
(reporting on 'Your Algorithm' feature). ([Link] ref: turn0news35) [5] 'Understanding Social Media
Recommendation Algorithms' — Columbia/Knight overview (2023). ([Link] ref: turn0search16)
Additional industry guides: Hootsuite (turn0search8), FanpageKarma (turn0search15).
