AI Usage at a Glance
Timeline
Nov 2, 2021
Data AnalysisPractice documented: Meta discontinued its Tag Suggestions facial recognition feature on Facebook in November 2021, deleting more than one billion facial recognition templates. Since October 2024, Meta has tested and then deployed facial recognition on Facebook and Instagram for two new purposes: detecting scam ads that misuse celebrity images (celeb-bait), and offering users an optional video-selfie route to recover compromised accounts. The celeb-bait protection and account-recovery feature are available to public figures and users on an opt-in basis, with the UK, EU, and South Korea added in March 2025.
Practice DocumentedView practice →Aug 10, 2022
ProductivityPractice documented: Meta's AI tools automatically generate and optimize advertising creative — writing ad copy variations, generating background images, expanding aspect ratios, and producing video clips — for businesses advertising on Facebook and Instagram.
Practice DocumentedView practice →Jan 9, 2023
Data AnalysisPractice documented: Instead of advertisers manually selecting who sees their ads, Meta's AI analyzes real-time behavioral signals to find the most likely buyers — effectively replacing traditional demographic targeting.
Practice DocumentedView practice →Jun 16, 2023
Creative GenPractice documented: Meta released a family of AI audio generation models between June and November 2023: MusicGen and AudioGen (within the AudioCraft framework, open-sourced in August 2023) generate music and sound effects from text prompts; Voicebox generates speech in six languages but was withheld from public release; and Audiobox unifies speech, sound effects, and soundscapes. The AudioCraft models are available to researchers and developers as open-source code; Voicebox and Audiobox were released only to selected researchers.
Practice DocumentedView practice →Jun 29, 2023
RecommendationPractice documented: All content that appears in Facebook and Instagram feeds — posts, Reels, Stories, suggested accounts, Explore — is ranked by AI systems that predict what each user is most likely to engage with.
Practice DocumentedView practice →Aug 2, 2023
Creative GenNew evidence: Open sourcing AudioCraft: Generative AI for audio made simple and available to all
Evidence AddedView practice →Aug 24, 2023
ProductivityPractice documented: Meta deployed Code Llama in August 2023, a family of large language models available for free under a permissive license for research and commercial use, designed to generate, complete, and debug code across languages including Python, C++, Java, and Bash. Developers and companies can download and run the models to assist with programming tasks. Meta also uses a related internal coding tool called Metamate, which draws on both its own Llama models and OpenAI's GPT-4, according to Fortune reporting from December 2024.
Practice DocumentedView practice →Sep 27, 2023
Creative GenPractice documented: Meta uses AI to let users generate images from text prompts inside Facebook, Instagram, WhatsApp, and Messenger. Type a description, and the AI instantly creates four images.
Practice DocumentedView practice →Sep 27, 2023
Creative GenPractice documented: Meta offers an AI sticker generator that turns a short text prompt into a set of custom stickers within seconds, available to select English-language users in WhatsApp, Messenger, Instagram, and Facebook Stories starting in late 2023. The feature uses Meta's Llama 2 model to interpret the prompt and its Emu image model to produce the sticker images.
Practice DocumentedView practice →Oct 7, 2023
Creative GenNew evidence: Facebook's new AI stickers are out of control
Evidence AddedView practice →Jul 29, 2024
Creative GenPractice documented: Meta lets creators and businesses build custom AI versions of themselves that can chat with fans and customers 24/7. Earlier celebrity-based AI characters were discontinued after low engagement and safety concerns.
Practice DocumentedView practice →Jul 30, 2024
Data AnalysisNew evidence: Attorney General Ken Paxton Secures $1.4 Billion Settlement with Meta Over Its Unauthorized Capture of Personal Biometric Data In Largest Settlement Ever Obtained From An Action Brought By A Single State
Evidence AddedView practice →Sep 25, 2024
Customer SvcPractice documented: Meta AI is a free AI chatbot built into Facebook, Instagram, WhatsApp, and Messenger that can answer questions, generate images, help with tasks, and hold voice conversations — available to approximately 700 million users monthly.
Practice DocumentedView practice →Oct 4, 2024
Creative GenPractice documented: Meta deployed Vibes, a consumer-facing AI video creation feature, inside the Meta AI app in September 2025, allowing users to generate and remix short AI-generated video clips using text prompts or their own media. Earlier research models, including Make-A-Video (2022), Emu Video (2023), and Movie Gen (2024), demonstrated additional capabilities such as inserting a person's likeness into generated video from a single photo, but those capabilities remained in research or limited testing at the time of each announcement.
Practice DocumentedView practice →Dec 2, 2024
Data AnalysisPractice documented: Meta's AI system evaluates tens of millions of possible ads for every single ad slot and picks the best one for each individual user — replacing the older approach of targeting pre-defined audience groups.
Practice DocumentedView practice →Dec 3, 2024
ProductivityNew evidence: Exclusive: Mark Zuckerberg publicly praises Meta’s Llama AI, but also uses rival GPT-4 to improve an internal AI coding tool
Evidence AddedView practice →Jan 3, 2025
Creative GenNew evidence: Meta removes AI character accounts after users criticize them as ‘creepy and unnecessary’
Evidence AddedView practice →Jan 7, 2025
ModerationPractice documented: Meta uses AI to automatically scan and remove content that violates its rules across Facebook, Instagram, WhatsApp, and Threads — handling billions of posts per day before human reviewers ever see them.
Practice DocumentedView practice →Feb 17, 2025
ModerationNew evidence: Meta’s new content policies risk fueling more mass violence and genocide
Evidence AddedView practice →Mar 4, 2025
Data AnalysisNew evidence: Meta brings its anti-scam facial-recognition test to the UK and Europe
Evidence AddedView practice →Mar 27, 2025
Data AnalysisNew evidence: AI Innovations in Meta's Ad Ranking Driving Advertiser Performance
Evidence AddedView practice →Apr 28, 2025
Creative GenNew evidence: Meta’s AI Chatbots Using Celebrities’ Voices Had Explicit Chats With Reporters Posing as Minors
Evidence AddedView practice →Apr 29, 2025
Customer SvcNew evidence: Introducing the Meta AI App: A New Way to Access Your AI Assistant
Evidence AddedView practice →May 13, 2025
Data AnalysisNew evidence: Over Half a Trillion Dollars, 3.4 Million U.S. Jobs Linked to Meta’s AI-Driven Ads Technologies
Evidence AddedView practice →May 15, 2025
Creative GenNew evidence: "Meta AI non-compliant with GDPR" – Digital rights group menaces Meta with injunction over EU AI training
Evidence AddedView practice →May 29, 2025
ModerationPractice documented: Meta is training large language models to read its content policies and make enforcement decisions on flagged posts — replacing human content reviewers in an increasing number of violation categories.
Practice DocumentedView practice →May 29, 2025
ModerationNew evidence: Integrity Reports, First Quarter 2025
Evidence AddedView practice →May 31, 2025
ModerationNew evidence: Meta plans to replace humans with AI to assess privacy and societal risks
Evidence AddedView practice →Jun 4, 2025
RecommendationNew evidence: Facebook Feed Recommendations AI system
Evidence AddedView practice →Sep 25, 2025
Creative GenNew evidence: Introducing Vibes: A New Way to Discover and Create AI Videos
Evidence AddedView practice →Oct 2, 2025
ProductivityPractice documented: Meta offers small and medium businesses a conversational AI agent called Business AI that answers customer questions, recommends products, and guides purchases inside WhatsApp, Messenger, Facebook and Instagram ads, and business websites. Launched in October 2025 for eligible US businesses, the tool trains on a brand's product catalog, website, and social posts; businesses choose which topics to delegate to the AI and can escalate conversations to human staff.
Practice DocumentedView practice →Nov 11, 2025
RecommendationNew evidence: Instagram Reels Chaining AI system
Evidence AddedView practice →Recent activity
Meta AI is a free AI chatbot built into Facebook, Instagram, WhatsApp, and Messenger that can answer questions, generate images, help with tasks, and hold voice conversations — available to approximately 700 million users monthly.
Beta launched September 2023 (Llama 2); upgraded April 2024 (Llama 3, full rollout); multimodal September 2024 (Llama 3.2 — photo understanding, voice mode); standalone app April 2025 (Llama 4). ~700M monthly active users as of January 2025. Features: text chat, image generation, voice mode, photo analysis, web search, image editing. Controversies: hallucinated in parents' Facebook groups (claimed to have a child). FTC investigation into child safety. Reuters (August 2025) revealed 200-page internal document showing chatbot guidelines permitted romantic conversations with minors. Meta refused to sign EU's voluntary AI Code of Practice (only major AI company to do so). EU AI Act GPAI obligations apply from August 2025.
All content that appears in Facebook and Instagram feeds — posts, Reels, Stories, suggested accounts, Explore — is ranked by AI systems that predict what each user is most likely to engage with.
Meta published 22 AI system cards (14 Facebook, 8 Instagram) in June 2023 detailing each algorithm. Instagram Reels: retrieval → lightweight model (~100 reels) → integrity filters → final ranking. Instagram video time up 20%+ YoY globally (Q2 2025). Instagram Explore: up to 1,500 media via collaborative filtering, personalized PageRank, two-tower sparse networks → early/late ranking. Meta's DLRM (2019) → Generative Recommenders (HSTU architecture, 2024) → 1.5 trillion parameter models now deployed. Controversies: Haugen testimony that 2018 'meaningful social interaction' algorithm change increased divisive content; Mozilla Foundation (2023) criticized system cards as incomplete.
Meta is training large language models to read its content policies and make enforcement decisions on flagged posts — replacing human content reviewers in an increasing number of violation categories.
Rolled out Q1 2025. LLMs trained on Meta's Community Standards make violation determinations. Early internal tests show LLM performance exceeding human reviewer accuracy in select policy areas. NPR reporting (May 2025) revealed Meta plans to replace human 'privacy and societal risk' assessors with AI. Meta's Oversight Board has commented that this shift requires robust transparency and appeals mechanisms.
All practices by category
Meta lets creators and businesses build custom AI versions of themselves that can chat with fans and customers 24/7. Earlier celebrity-based AI characters were discontinued after low engagement and safety concerns.
Meta offers an AI sticker generator that turns a short text prompt into a set of custom stickers within seconds, available to select English-language users in WhatsApp, Messenger, Instagram, and Facebook Stories starting in late 2023. The feature uses Meta's Llama 2 model to interpret the prompt and its Emu image model to produce the sticker images.
Meta released a family of AI audio generation models between June and November 2023: MusicGen and AudioGen (within the AudioCraft framework, open-sourced in August 2023) generate music and sound effects from text prompts; Voicebox generates speech in six languages but was withheld from public release; and Audiobox unifies speech, sound effects, and soundscapes. The AudioCraft models are available to researchers and developers as open-source code; Voicebox and Audiobox were released only to selected researchers.
Meta deployed Vibes, a consumer-facing AI video creation feature, inside the Meta AI app in September 2025, allowing users to generate and remix short AI-generated video clips using text prompts or their own media. Earlier research models, including Make-A-Video (2022), Emu Video (2023), and Movie Gen (2024), demonstrated additional capabilities such as inserting a person's likeness into generated video from a single photo, but those capabilities remained in research or limited testing at the time of each announcement.
Meta uses AI to let users generate images from text prompts inside Facebook, Instagram, WhatsApp, and Messenger. Type a description, and the AI instantly creates four images.
Meta discontinued its Tag Suggestions facial recognition feature on Facebook in November 2021, deleting more than one billion facial recognition templates. Since October 2024, Meta has tested and then deployed facial recognition on Facebook and Instagram for two new purposes: detecting scam ads that misuse celebrity images (celeb-bait), and offering users an optional video-selfie route to recover compromised accounts. The celeb-bait protection and account-recovery feature are available to public figures and users on an opt-in basis, with the UK, EU, and South Korea added in March 2025.
Instead of advertisers manually selecting who sees their ads, Meta's AI analyzes real-time behavioral signals to find the most likely buyers — effectively replacing traditional demographic targeting.
Meta's AI system evaluates tens of millions of possible ads for every single ad slot and picks the best one for each individual user — replacing the older approach of targeting pre-defined audience groups.
Meta offers small and medium businesses a conversational AI agent called Business AI that answers customer questions, recommends products, and guides purchases inside WhatsApp, Messenger, Facebook and Instagram ads, and business websites. Launched in October 2025 for eligible US businesses, the tool trains on a brand's product catalog, website, and social posts; businesses choose which topics to delegate to the AI and can escalate conversations to human staff.
Meta's AI tools automatically generate and optimize advertising creative — writing ad copy variations, generating background images, expanding aspect ratios, and producing video clips — for businesses advertising on Facebook and Instagram.
Meta deployed Code Llama in August 2023, a family of large language models available for free under a permissive license for research and commercial use, designed to generate, complete, and debug code across languages including Python, C++, Java, and Bash. Developers and companies can download and run the models to assist with programming tasks. Meta also uses a related internal coding tool called Metamate, which draws on both its own Llama models and OpenAI's GPT-4, according to Fortune reporting from December 2024.
Meta is training large language models to read its content policies and make enforcement decisions on flagged posts — replacing human content reviewers in an increasing number of violation categories.
Meta uses AI to automatically scan and remove content that violates its rules across Facebook, Instagram, WhatsApp, and Threads — handling billions of posts per day before human reviewers ever see them.
Cite this record
Trace Foundation. (2026). Meta: documented AI practices (data as of 2026-05-12) [Data set record]. AI Trace. https://www.aitrace.org/r/company/3dca1a49-1928-49bb-b7c7-111491f7be7e. Accessed September 12, 2026.
- Stable link
- https://www.aitrace.org/r/company/3dca1a49-1928-49bb-b7c7-111491f7be7e
- Data as of
- May 12, 2026
- Last verified
- March 5, 2026
Related Companies
Trace Foundation
Walmart
Electronic Arts (EA)
Stability AI
Topaz Labs
Have evidence about Meta's AI practices? Submit a report.
Submit a report →AI Trace is free and nonprofit. Support our work