Alison Lee: "it doesn't make sense because UXR used to be embedded into product pillars, which then rolled up into app hierarchies (Instagram research) which then would roll up into FOA structures. but because of the leaks any and all researchers who work in sensitive spaces now technically report to central integrity or central impact"
Alison Lee: "and ONLY researchers"
Alison Lee: "so my eng and my ds are still in the old product pillar based structure, and i'm the only weirdo that reports to wellbeing, that circles up to central integrity"
Alison Lee: "idk what problem it was supposed to solve but the idea was to centralize all the teams to make it harder for people to leak stuff??"
Maxwell ████: "Honestly, the amount that we have to communicate across boundary lines is probably increased now. And furthermore since I work with a bunch of AI people, they fall under RL and have permission to see a bunch of stuff I can't. I think it just made our jobs harder to do, and they can't tell the difference between that and making things harder to leak"
Alison Lee: "thats pretty much true for all the new processes they've layered on to us"
Alison Lee: "it doesn't deliver on their intentions, and the reality is that it feels like censorship, removal of agency, and sends the message that our work (and expertise) is seen as a liability rather than an asset"
Alison Lee: "at least that's how i've been feeling about it"
Maxwell ████: "I had really hoped the leaks would demonstrate the weight of integrity research and it's importance ... I think it has for some, but others it's just motivated to cover their asses"
Alison Lee: "I think the part that has annoyed me the most in recent months is that leadership keeps trying to package it as 'making our jobs easier' when in reality I spend so much more time acquiring approvals and going through unnecessary processes. If the whole point is to lock things down more then just say that rather than selling us snake oil"
Alison Lee: "And it doesn't feel like a space that respects calling things for what it is…"
Alison Lee: "I think it's part of the reason why I was ready to move on from this space (and many others from IG as well)"
George Volichenko: "This is a celebration post to call out that the Mental Well-Being team has accomplished the primary P0 goal for the half: 'Reduce experiences of problematic use and negative appearance comparison. Goal metric: % of weekly active teens meaningfully adopting Take a Break or Alt Topic Nudge (sum of individual feature adoption figures).' As a reminder, our target was 0.25% and, as of the last full week of June, we are at 0.30%. This figure is obtained by adding the following adoption percentages: 0.137% of teen WAU clicked on one of our Alternative Topic Nudges; 0.165% of teen WAU took a 10min+ break after receiving one of our Take a Break nudges (honored the break). Great job, team! We've had a very intense period in the last few months with numerous tests and launches, but we pushed through and drove this to completion. Onwards to the next half!"
████: "Congratulations all! Special thanks to ████ and ████ who helped revisit the goal when default opt in was deprioritized, thank you for your support!"
████: "I know you've been pitching using solution adoption as proxy for reducing prevalence of a privacy problem - good to see it working in practice."
████: "We took into account severity but it was in the sense of, do we have causal evidence in the literature that these behaviors lead to negative life impact"
████: "For late night use, the negative impacts, such as on work/school performance and mood, of little or poor sleep (especially on young people) are well-documented."
████: "And product coverage was low, while competitors are making progress in this space. So those were the primary inputs to deciding late night use should be a priority area for us"
████: "We cited this stat in our H2 plan: Late Night: 20% of global WAU teens have a daily average of 1 session between 12 and 4am over one week."
"Insiders and voters hated claims that social media is not addictive. Insiders and voters generally anchor heavily on their own personal experience and don't trust research. ... Avoid making any mention of addiction. Instead, make a commitment to address concerns going forward."
"we estimate that 18.3% of IG teen users exhibit behaviors associated with PU."
"negative impact of poor sleep is well-documented (ref) and this is where we have coverage gaps for ~8% of teens"
"Previous research which linked problematic Facebook use to late-night use, has been associated with both later bedtimes and rising times and with insomnia."
"33% of teen sessions between 12 am and 4 am start due to a notif (26% receive and click after midnight)"
"An average of 40% of teen's sessions, start from a notification or an app badge."
"Long periods of use" mitigations listed for the "Instagram coverage" column: "Daily usage limits", "Take a break (upsell and tool)", "Take a break (tool only, no upsell)"
"FACEBOOK'S INSIGHTS WORK: PROBLEMATIC USE
External research has [ HYPERLINK "https://akjournals.com/view/journals/2006/8/1/article-p169.xml" \h ] that excessive use of social media can result in riskier decision-making patterns associated with substance use and addictive behavior disorders. While Facebook [ HYPERLINK "https://fb.workplace.com/groups/ResearchInsights/posts/2042922635756341/" \h ] evidence that it could be capable of causing true addiction, there are still several aspects of problematic use that are negatively impacting users' lives which Facebook is solving for.
Facebook [ HYPERLINK "https://cortex.thefacebook.com/Share/a7xwn7pcguf1f8b805owuw13yd76s2f1?TP=UXR" \h ] this problematic use as having serious problems with sleep, work, or relationships that a user attributes to Facebook - and concerns with/preoccupations about how they use Facebook (e.g., a lack of control or a persistent fear of missing out) - and finds that 31% of US users meet this definition."
"33% of teen sessions between 12 am and 4 am start due to a notif (26% receive and click after midnight)"
For users with late night PU behaviors: "An average of 40% of teen's sessions, start from a notification or an app badge." and "18% of dau start a session from a notification that was sent and clicked within 12 am and 4 am"
Gargi Apte: "To clarify, 7.5% of teens exhibit proxy PU behaviors. 66% of these open the app at night due to a notif."
"43% (of the 18%) of the sessions between 12 am and 4 am are caused by Direct notifs"
"18.3% of WAU IG teens (over one week) exhibit behaviors associated with problematic use overall"
"PU is among the top WB issues teens report experiencing, and a driver of churn"
"33% of teens WAU Have at least one 20min+ session per week"
"Teen needs are a function of biological development… Social media apps can fill these evergreen teen jobs"
"Instagram has an existential crisis with Teens who are turning to other platforms. Retaining teens is key to the long term health of our business and our metaverse aspirations."
████: "Can I please see the Mercury project results? ████ led me to ████ who is... not being responsive."
████: "I actually DON'T have access to those results myself, as they were kept pretty confidential... all that is to say, happy to track this down for you, but you have to tell me the business reason for needing them because i'll need that to make the case"
████: "lol no way, this is like the most secretive thing i've ever heard of"
████: "my business reasons is that i'm a research manager in the wellbeing space so like, I generally find it helpful to know what wellbeing research results say for... not doing duplicative work, and generally doing my job of understanding the impact of our platforms so i can make it better"
████: "and honestly I can guarantee I have worse shit than whatever it says lol Feed Relevance consistently distributes SSI vpvs to teenagers and nobody has noticed, like how much am I gonna care that deprivation studies show wellbeing gains from not using our shitty products"
████: "I once saw Policy outright state that they wouldn't take down the account of a Cameroonian expatriate with a massive in-country following who was explicitly calling for the murder of Ambazonians who participated in the Cameroonian elections – which then absolutely happened, we saw the videos – because of the PR implications. I think folks are overestimating how bad the research evidence is vs the actual practice of working at this company."
████: "I hear you. Let me dig in and see if I can figure out a) why this is so sensitve (my sense it's less about the findings and more about leadership's reactions); and b) if I can get around this, where it is"
"56% of IG teens surveyed say it's difficult to manage how much time they spend on social media and 14% say IG makes it worse"
"Teens and young adults are more likely to report experiencing problematic use, consistent with developmental differences in self-control"
"Teens say they want help… 41% of weekly teen users say they try to limit their IG use and 22% say they want help controlling their usage"
Child sexual exploitationFriend/Account Suggestions
"Results Summary: Significant Improvements Needed"
"The IG Well-being Teen Safety team does not always limit teen accounts from being recommended to the IIC potential violators* and vice versa. For example, on a sampled day tested, on the Accounts You May Follow (AYMF) surface, IA identified: 1) 7.2% (~1.4M out of ~19M) of potential violator accounts in AYMF were recommended to at least one teen. 2) 4.0% (~3.4M out of ~85M) of AYMF recommendations to potential violators were recommending teen accounts."
"For example, on 2022-11-25, IA observed that out of ~2.5M IIC potential violators who viewed the follow list surface, 37% ( 924K) were shown at least one unconnected teen account."
"For a sampled week tested: 1) 2.2% (~27K out of ~1.2M) of IIC potential violator created content on the Explore surface was recommended to at least one teen, accumulating ~466K teen impressions. 2) 1.6% (~83K out of ~5.2M) of contents recommended to IIC potential violators on the Explore surface were created by teens, accumulating ~200K potential violator impressions."
"Users' perceived control can be validly measured across age cohorts with questions assessing low attentiveness around platform use (i.e., spending more time, checking, and consuming content without realizing it) and low capability to change platform use (i.e., low confidence in ability to reduce time, checking, or interactions)"
"Rates of perceived low control are similar for youth and adults, with low attentiveness (~34-40%) more common than low capability (13-20%)"
"Perceived low control nearly fully explains the small association between on-platform behaviors and perceiving that one's platform use prevents fulfilling other responsibilities in life ('life interference')."
"Global behaviors associated with perceived low control reflect high usage, evening + late night time spent, and specific behaviors, like no engagement repeat sessions and video chaining."
"Rates of perceived low control are similar for youth and adults, with low attentiveness (~31-35%) more common than low capability (16-20%)."
"Behaviors associated with perceived low control reflect high + frequent usage, response to notifications, evening + passive time spent, and specific behaviors like no engagement repeat sessions and surface switching."
"Specific product usage associated with perceived low control include content consumption + engagement (Stories, News Feed, Reels) as well as notifications."
"Stories usage is the strongest and most consistent predictor of low control. This may be due to content consumption affordances (like autoplay) but also about ephemerality, which may contribute to not wanting to miss updates and desire to respond to others in a timely way (consistent with social pressures, FOMO; expert input supports these hypotheses 1, 2; user research 1)."
"Notifications have been specifically pointed out as a possible trigger of or contributor to low control usage (1, 2, 3, 4); this may not only be due to initiating new sessions, but diversion within a session."
"Frequently viewing chained content, regardless of which surface, is associated particularly with low attentiveness. This makes sense as a product affordance that may contribute to people spending time or consuming content without realizing it (expert report). Importantly, depth of chaining is less associated. This would support enacting more limits on autoplay of next videos that occurs across surfaces, and/or adding pauses or friction."
"Stories creation is the strongest predictor of perceived low capability to reduce use, and also associated with low attentiveness - though has been found to be a high intent behavior (1). Content creation was also identified in previous PU research (2), as highly active users are also more likely to create content. Creation likely leads to more responses and thus likelihood of returning to use the app, all in a time-bound ephemeral way."
Child sexual exploitationFinancial sextortionDirect Messaging
"[Bad actor number 2] had a total of 900 message threads within a few months With 338 reactive reports against both the account and message threads. Many of these reports contained signals of sextortion. However, you'll see none of them were actioned for sextortion or NCIT leading to disabling. We will call out the user was striked for bullying and harassment and banked NCII, but these strikes did not enforce on that bad actor in any way."
"The first proximate for this problem is that we, as in Meta, did not proactively detect these bad actors or their violating content and conversations in IGDM. And that is because we do not have proactive detection and enforcement running for NCII or sextortion in IGDM"
"We also do not have any classifiers running for sextortion specifically."
"The third root cause for this proximate is that the users did not hit the IIC, or inappropriate interactions with children, behavioral classifier threshold."
"0% of these IGDM reports against these bad actors were given human review. We did this because our ranking system did not rank these jobs high enough for human review."
"Also, 83% of these jobs were routed to the sub-Sahara Africa queue that has low human review capacity This is because our jobs are routed to a queue based on the location of the responsible party, in this case Nigeria. And we also only have enough human review in this market to review 1-2% of reported jobs."
"And finally, we deprecated the blackmail contact form that formerly allowed users to specifically report sextortion And we did that because of a lack of reviewer bandwidth."
Child sexual exploitationBody image & eating disordersSuicide & self-injuryContent Algorithm
"The Internal Audit (IA) team has completed the IG Youth Well-being Assurance Audit - Phase 1. This report is rated as 'Significant Improvements Needed' with 3 High, and 14 Medium risk observations."
"Key observation themes are summarized below:
- Inconsistent filtering on potential Inappropriate Interaction with Children (IIC) accounts and content.
- Incomplete coverage of recommendation filtering on Minor-Restricted Regulated Goods and Services (RGS) content (e.g., diet products) for teens.
- Lack of documented rationale for exemption of verified accounts and Suicide & Self Injury (SSI) signal retention period.
- Missing comprehensive monitoring and alert of filtering volume, leakage, and machine learning performance."
Child sexual exploitationFriend/Account SuggestionsDirect Messaging
"One method we're using is to remove minors in inbox search for unconnected adults… You can typically search for people you are not connected w/ in Inbox search (i.e. start a new message w/ someone you're not connected to in this way); this blocks that feature… If minors are discovered through other mechanisms, FB and Messenger have special protections in place for minors… At the moment, we only have risky adult <> minor filtering on People You May Know for FB Blue - so the above does not apply to blue"
"Youth is one of the more acute social issue challenges we face right now with a level of external scrutiny that has increased substantially over the last 6 or so months."
"We are facing over 1000 lawsuits from AGs, school districts and parents alleging that social media companies purposely make their apps addictive to grow their teen audiences, at the detriment of teen safety and mental health."
"We've seen investigative reports about alleged child harm (pedophile networks)."
"However, the level of scrutiny has increased exponentially and we've been too slow to respond."
0.45% of users predicted to be Youth and predicted to reside in the U.S. had Take a Break enabled on Instagram between January 8, 2023 and September 24, 2024.
"My primary goal was to ensure the app remains culturally relevant as measurement by sessions, time, and sharing, particularly with teens. Overall this went well. Reels have driven time to all time highs at ~41.9m per DAU globally and ~37.4m in the US. Sessions are up X% globally and roughly flat year on year in the US. The sum of Feed, Stories, and producer participation rate is up above 20% and has grown more than 6% globally and above 18% and up 2.6% in the US."
"These goals are a second iteration for 2023… Stabilize Instagram with teens (US teen MAP). This is both the most important and most difficult of my goals for 2023. We have a 10% chance of hitting this goal at best, but we'll do well to think big and swing big."
"The p90 threshold for late night (10 PM - 4 AM) time spent over the last 28 days corresponds to 25 minutes per day on IG."
"35% of late night time spent for p90+ teens comes
from sessions that had a notification delivered within 5 minutes of starting a session on the app."
Gail ████: "Hi both, I see you are both on the call for 18-20 talking to unconnected minors."
Gail ████: "For 2) the safety risk appears to be from messaging per se, possibly exacerbated by age liars. Anand's data indicated that 32% of IIC reports relates to 16-17 year olds. Given we - largely - go with stated age, it's impossible to know if this is actual 16-17 year olds or age liars. Allowing 18 year olds to message 17 year olds wouldn't change this."
Gail ████: "option 2 would appear to break the product, at a time when we are looking to engage with teens and offer them a positive, safe experience. We're basically saying that high school seniors need to work out a complicated way to message each other, if they meet at a party."
Ravi Sinha: ""Thanks, Gail and Senka. I think that, if we're called to explain the shift to Option 3, we'll likely have to get into the details of current policy, which I think is extremely hard to defend. Tho Antigone (and Comms) has concerns around reception of Option 3 in the US environment more broadly, my most acute external concerns center around what questions we'll get asked when we start talking to regulators/media about these protections differently, as we'll have to do with Option 3."
Gail ████: "I'm not worried about the general policy environment. I think in the regulatory environment we will have to explain why we currently allow 20 year olds to message 13 year olds. That seems like a far greater risk."
Sources include internal documents, public filings and press reports. “Alleged by plaintiffs” denotes the plaintiffs’ characterization, not Meta’s own words. Source labels identify provenance, not a finding of fact.
Meta & Children's Online Safety