Instagram Appearance Comparison Decision -

Raw text

Transcribed text as recorded, including headers and formatting fragments.

Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 1 of 16 AMENDED Exhibit 121 PLAINTIFFS' OMNIBUS OPPOSITION TO DEFENDANTS' MOTIONS FOR SUMMARY JUDGMENT Case No.: 4:22-md-03047-YGR MDL No. 3047 In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 2 of 16 Instagram Appearance Comparison Decision A/C Priv - Decisions Needed: 1. Can IG move forward in development of Appearance Comparison metrics, including the deployment of user surveys, for potential future goaling provided that IG addresses CWB feedback on survey development, risk mitigations, and metric validation plans? Deciders: Central Social Impact / Research: Pratiti Raychoudhury IG / Youth: Miki Rothschild Full RACI[ HYPERLINK "https://docs.google.com/document/d/1E 1ZUnyPzriP5U3CLHO6H GaRE/7hAKbjQvLuFJrqxha8Q /edit" \h ] Timeline: e Needed for a decision: mid-March (email review to start March XYZ, decision triggers urgent H1 execution or roadmap replanning) Summary: e Motivations/Expected Benefits: Given current IG Well-being strategy focus on Appearance Comparison, IG would like to continue investment in development of an appearance comparison metric to assess success of in development products and potentially for future goaling. e Risks: Central Well-being (CWB) believes that the risks associated with continuing this sensitive-topic metric development outweigh the benefits given lack of clarity on theory of change for current products, lack of clear alignment with company well-being strategy, and in development alternative metrics for problematic use. e Mitigations: If IG is able to continue pursuing AC metric development, CWB strongly recommends multiple survey quality and risk mitigation strategies. Decision Details AC Measurement Motivailions and Expected Benefits 1) Instagram decided to focus its mental well-being strategy on appearance comparison because: a) Appearance comparison is experienced by many in our communities. An estimated 33% of Instagram DAU across 10 countries (FCI + Mexico) compare their appearances to others' on Instagram "often" or "always"; 48% "occasionally" or "never" compare their appearances. An estimated 48% of teen girls compare HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005132  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 3 of 16 their appearances to others "often" or "always", and 30% "occasionally" or "never" compare their appearances ([ HYPERLINK "https://fb.workplace.com/notes/1065141417573177" \h ]). b) Academic research suggests a causal link between Instagram use and appearance comparison and/or body image issues. Longitudinal studies indicate a bidirectional link between social media use and body image issues ([ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S1740144521001327" \h ], [ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S0747563217302157" \h ], [ HYPERLINK "https://doi.apa.org/doiLanding?doi=10.1037%2Fppm0000206" \h ], [ HYPERLINK "https://onlinelibrary.wiley.com/doi/10.1002/eat.22640" \h ], [ HYPERLINK "https://journals.sagepub.com/doi/10.1177/1461444819894346" \h ]); lab experiments in which participants were assigned to browse appearance-relevant content also demonstrated increases in body-related concerns compared to participants assigned to browse appearance-neutral material ([ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S1740144519303754?via%3Dihub" \h ], [ HYPERLINK "https://psycnet.apa.org/record/2019-14939-001" \h ], [ HYPERLINK "https://pubmed.ncbi.nlm.nih.gov/25035882/" \h ], [ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S1740144517305326" \h ], [ HYPERLINK "http://www2.psy.unsw.edu.au/Users/lvartanian/Publications/Fardouly,%20Diedrichs,%20Vartanian,%20&%20Halliwell%20(2015).pdf" \h ]). Appearance comparison is typically believed to be the link between social media use and body image ([ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S1740144514001375" \h ], [ HYPERLINK "https://isiarticles.com/bundles/Article/pre/pdf/36346.pdf" \h ], [ HYPERLINK "http://www2.psy.unsw.edu.au/Users/lvartanian/Publications/Fardouly,%20Diedrichs,%20Vartanian,%20&%20Halliwell%20(2015).pdf" \h ]), particularly on image-based social media platforms ([ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S0736585319308111" \h ], [ HYPERLINK "https://www.sciencedirect.com/science/article/pii/S0747563220300832" \h ]). c) This is one of the most significant external criticisms we're facing with regard to teen well-being. Regulators and press are pushing us to address social comparison and negative appearance comparison specifically ([ HYPERLINK "https://www.blumenthal.senate.gov/newsroom/press/release/blumenthal-and-blackburn-introduce-comprehensive-kids-online-safety-legislation" \h ], [ HYPERLINK "https://www.whitehouse.gov/briefing-room/statements-releases/2022/03/01/fact-sheet-president-biden-to-announce-strategy-to-address-our-national-mental-health-crisis-as-part-of-unity-agenda-in-his-first- HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005133 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 4 of 16 state-of-the-union/" \h ]). We've also made our commitment to addressing social comparison a primary proof point in our recent external narratives ([ HYPERLINK "nttps://about.instagram.com/blog/announcements/using-research-to-improve- your-experience" \h ], HYPERLINK [ "https://about.instagram.com/blog/announcements/pausing-instagram-kids" \h ]). 2) There was leadership HYPERLINK [ "nttps:/Avww.dropbox.com/s/ql6eicl4kygszzh/Re_%20Answering%20your%20question% 20from%20Q%26A%200n%20Social%20Comparison.em|?dl=0" \h ] between Adam and Pratiti in June 2021 on IG proceeding with negative social comparison IG proactive work through the lens of agency and control, one of the core mechanisms for how Meta can help reduce Problematic Use. 3) IG is defining success for its teen mental well-being investments to focus on reducing appearance comparison, with IG leadership alignment. 4) IG's bets were framed around generalized agency & control solutions due to uncertainty around its use of AC. These bets are: (1) Take a Break (long-session reminder) and (2) Topic Nudge (in-feed nudge on long scrolling in recs to alternate topics). The AC measurement that is being discussed here is partly intended to inform H1 decision making by enabling IG to weigh likely engagement regressions with user benefit (as measured by AC sentiment). 5) IG aims to unblock new product bets beyond H1, where AC is more deliberately targeted with appropriate theories of change. Having clear approval to measure AC will enable IG to design more targeted interventions both within the current H1 bets and to pursue other AC reducing pathways beyond agency & control (current HYPERLINK [ "nttps://docs.google.com/presentation/d/19JpzbdOFo6PwDnRF-AGLOL- cbtyDbNGgXQ8UJSL vpjc/edit" \I "slide=id.g1048090c7e9 8 37" AC Measurement Risks 1) Sensitive Topic Risks. Asking users on platform about sensitive topics (like appearance comparison) come with a number of risks that must be balanced against likely benefits. These include (1) negative user experiences (i.e., users do not come to our platforms to be asked about high sensitivity topics like AC); (2) concerns about having data on such experiences retained in our systems; and (3) comms risks (e.g., screenshot risks). (4) Unclear or unknown consequences of optimizing for this type of metric (e.g., it may reduce important and healthy content too, like motivational and inspirational posts). 2) Benefits may not outweigh these risks: The CWB team has concerns that the benefits of AC survey measurement do not outweigh these risks. These concerns include: a) Appropriate theory of change: There is not clear theoretical or empirical basis to expect that Take a Break should impact AC. While there may be more of a Commented [1]: If we are pursuing more longitudinal path for Topic Nudges, this depends highly on product design specifics and | maybe we canremove tis? Measurement, would require that the product target and reduce the prevalence of appearance- HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005134  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 5 of 16 comparison content. If we are taking on the inherent risks here, there should be a very clear expected linkage between intervention and measured outcomes. If IG has a path forward to building substantially more targeted interventions, and will not be relying on the AC metrics for core decision-making on current generally scoped products, concerns are substantially lessened. b) (Mis-) Alignment with Core Company Wellbeing Strategy. Unless anchored in Problematic Use, appearance comparison is not an active core topic priontized in the x-Meta strategy. This does not mean that teams cannot work on this topic or that it is a bad topic to work on (as per previous leadership alignment), but it does mean that expected benefits to the company are lower in the overall risk seasateeeasatneasateeeasaneteasateeeasaneesasaenesasaenesasaeensasaeensasesasatseensasseeaeasseeneatsneneasseeneaeas, calculation. | Commented [2]: | think we should remove this bullet. Just because it's not in the «mela WB strategy does 3) Alternative Measurements. Given these concerns, CWB recommends that IG pursue not mean expected benefits to the company are lower. different metrics for currently proposed products. a) [Short term] Usage Metrics: IG switches to focus on usage metrics (ex. how many people click Topic Nudge to end long-scrolling on one topic). This tells us whether people find value in product interventions but not if the interventions reduce outcomes like appearance comparison. This is the status quo. b) [Medium term] Switch to Problematic Use (PU) measure. IG switches to actively invest sooner than planned in Central Well-being's PU measure. CWB has started a new initiative to develop company-level well-being metrics around |A the five prioritized well-being topics, including Problematic Use (PU). PU v0 | measure will be identified at the end of H1 or saon after. Commented [3]; @fb.com this is an i) Benefits of using PU instead of AC include (1) clearer theory of change important detail that can terially change the outcome. H1 work well planned out, so if we want to is for current product portfolio and (2) clearer alignment with cross-company be more PU focused in H2, we'll need fo start planning well-being strategy. in Q2. Design planning will work backwards from how PU is defined, What would it fake to develop a clear enough definition of PU so enable H2 planning within IG? \ Assigned to @fo.com Commented [4]: CWB is standing this up now from 2 defintional perspective and puiting together the associated measurement framework, which we expect within next several weeks. This will enable standing Appendix = up of metrics in Q2 (pending some staffing arrangements) Overview of AC Metric Development Plan 1. The measurement plan consists of two metrics: NAC ground truth (survey) and pNAC (statistically estimated ground truth proxy). We expect to have a vO version ready for validation and understand work in Q2. We will follow [ HYPERLINK "nttps://fb.workplace.com/groups/1540209922820872/permalink/1785747918267070/" \h ] to determine the appropriateness of the proxy. o The key criteria we'll anchor success to are: m= Construct validity demonstrate that operational metric correlates well with the Ground Truth metric. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005135  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 6 of 16 Sensitivity demonstrate that the metric can be moved in individual - experiments and the holdout ie) Demonstrating construct validity requires re-launching the survey in experimental holdouts 2. IG was also planning to iterate to improve the measurement apparatus used for goal setting. In particular, we intend to launch post-session surveys so that we can better correlate key session events, content and context to the ground truth collected by the survey. If we continue this plan, |G can amend our total plan as follows: o Use vO pNAC primarily for understand work in H1 products © Incorporate changes required by CWB into the next survey iteration, and base the v1 pNAC on this ground truth. Commented [5]: Our discussions with policy pariners. Follow above validation steps on v1 pNAC, incorporating additional around doing work of this kind strongly advise AGAINST doing work on such sensitive topics with validations as requested by CWB. ii | minors UNLESS parental consent is obtained. The message has been: this type work can be very of ' | valuable, but only done well/ethically, and parental if consent is a cornerstone of research with minors, especially if we're asking about health/well-being CWB Execution Recommendations and iG Responses Commented [6]: Appearance comparison is common : among women of most ages, it would sti! be possible to measure the phenomenon among that population Risk Mitigations. If |G does move ahead with AC measurement development, CWB | | Vand extrapolate learnings. recommends multiple strategies to improve survey quality and minimize risks / maximize benefits. Below are the specific recommendations and planned mitigations/responses from IG: Viewpoints {one methodology under consideration) i actually doesn't allow minors to be surveyed since they require a payment method. Commented 4 Are you suggesting extrapolating findings from adult women to teen girls? | wouldn't "| Recommendations Description IG's response suggest that, from @ developmental perspective, Do you have evidence suggesting the phenomenon manifests similarly on our platforms across the Consent Clear and explicit consent; ideally parental include explicit consent at the begi lifespan? And it seems like the strategic motivation is consent from minors. of the survey; discuss with policy drawn from a desire to address this phenomenon i whether to include parental conse specifically among teens (also what we've committed to | doing publicly). i Consider implementing on Viewpoints to minors. likely bea (This should surveys ce enable robust consent and data protection Policy that includes _| other on. PU Right, Viewpoints isn't currently enabled for minors (but practices. well-being-related topics.) [think they'd like to bel). 4 | Commented [8]; Just sharing this as a potential Survey Remove emotional terms from survey Done approach to mitigate risks. There are certainly tradeoffs. and more work would need tobe done to: questions (and start with validated items). validate that extrapolation would be useful, (This could also be young adult women, say 18-25, to get closer In: Audit survey questions; process is TBD, but Included in [ HYPERLINK terms of develooment stage.) What we do seeis that: something like [ HYPERLINK "nttps://docs.google.com/docume! there Is plenty of prevalence for women across ages who deal with upward appearance comparison as it "nttps:/Ab.quip.com/1DrgA2DT Ghmk" \h J, 5z5y3IR2zjkdOp1zZhWzWIQlYA) pertains to images seen on fm not sure about how including, but not limited to validating cdXulzPJc/edit" \h ] directly this experience translates across development instrument is catching the problematic stages- but if the sharp increase in botox usage is any content (e.g., if a classifier exists, pull indication it's likely that intemalization of beauty ideals content being demoted and then see if it is something that persists across age ranges. correlates with desired [ HYPERLINK The approach of improving the platform to enable teens "nttps:/Ab.workplace.com/100051002047003/p (perhaps mediated through parents) is another : potenti : HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005136  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 7 of 16 osts/432212635155468/" \h J; or include an open-ended question around why they gave their response to a survey question). Seek external consulting on use of items for Continue to develop our relationships survey input (either via IRB and/or academic with external experts including Phillippa advisory board) Diedrichs. ~| Commented [9]: Would want fo see and align upon a stronger plan here. Strong Data Protection Practices Document a data storage plan that includes: only sharing data with essential personnel; Continue to follow existing storing de-identified data pratti after, 90 Commented Agreed. This should be more than just a relationship. This should be a commitment to ensuring data is not connected to any getting academic advisory about the approach overalt Land the specific survey items proposed identifiable information (name, contact ba . es information, IP address). | Commented You probably know this but we've (1G) been working with Dr. Dietrichs one of our primary Consider implementing on Viewpoints to expert partners. As has been mentioned before, she's enable robust consent and data protection i: | been a key help in developing our strategy. https /b. | workplace com/notes/7 76131753309 This was eary feedback but she also was practices. consulted during our Take Break design and concept A testing. | néed to check in with our policy partner (Vaishnav V) whom has been working with her and upfront how you'll determine 'whether the pAC validation; will likely rely:on mi Soliciting feedback throughout to see our current survey items validate pNAC). surveys. PNAC i proxy is a Teck(formalized. working relationship with her. | believe it was a advisory boards we're building out for youth or well- being that we might be able to tap into here? | do think it would be helpful to think more deeply about the Communication Any research outputs should be caveated relationships we have, beyond Dr. Dietrichs and figure and framed appropriately, based on the to sharecuts. out the pecifics around how we will engage them. robustness of the measure; findings properly @fo.com do you have a more expansive list: contextualized. of experts we might want to consider for something like i i this from your lit review? Foundational work Keep work in the foundational bucket, not to IG has not selected using this me The IG teams haven't yet used IRBs for research, to be used for goaling more for understand- — a goal yet, but it is under consid' my nowledge. | think it's worth serious consideration related research to help inform product Continue with foundational work @fo.com and @fo.com do you have strategy. any examples of what this has looked like for teams Appearance Comparison to infor here that we can learn from and any benefit trade-offs product strategy. Re-evaluate goal the academic advisor/panel route? to the academic advisor/panel route? es metrics as we gather more data, | || Commented [13}: not sure this sufficiently including sensitivity to product chal addresses concems here. Commented [14]: if we're going to do this on-platform, recommend the team speaks with the Health RL | i Gamer central well-being leadership to goal on this Yes, for deciding giog t team, who've developed strong data storage practices (including use of a data custodian) for sensitive data. approval we'd invite central well-being If goaling (not stakeholders to the IG |. Commented [15]: @ib.com can the sel of currently advised by review. validation concerns be more carefully laid out? privacy | CWB) Incorporate guardrails around interpretation of metric, and what product(s) it should be Yes, as part of the decision revie adopt a goal on this metric, we'd s can en > used for and not use cases and standard guardra {16}: dro@fb.com | thought those use cases. Commented What kinds of criteria would be used | | to determine if it should be used for goaling? i HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005137  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 8 of 16 Voseceneettnnnennannnceentannnnenntanneeaentnnneennmne Elucidate counter-metrics (i.e., how you will We'd cover this in the decision. i Commented [18]: See comment above about determine that high pNAC content doesn't to adopt a goal on this metric. elucidating criteria for goaling. And, what are thoughts ALSO play an important other role for users) about what such counter-metrics prdbnibivasnining might look like? faunirbnashardsnsnipanntnurizainissashirssatnisdsatiirasapestasasiiessatnicssatesecsasisenasicdon Explore and report on what type of content Continue ongoing research to and/or experiences is being increased or understand (1) what subpopulations may decreased as well as differential impacts by be disproportionately affected by product sub-population. interventions and (2) what kinds of content are increased/decreased by our metric choice. Seek external consulting on use of items for See above on external consulting. goaling (either via IRB and/or academic advisory board) Theory of Change (High Level) 3. The MWB H1 product bets (TaB & TN) were designed to be intentionally general with an emphasis on agency & control. Appearance Comparison has always been the North Star but due to measurement & allowable intervention uncertainty the team opted to start with the more general framing. o The theory of change for H1 bets is centered on awareness. Although many young people can recall NAC experiences when asked about them retroactively, it's typically not top of mind for them while they're using IG. They[ HYPERLINK "https://docs.google.com/presentation/d/1PHGJS6jOw-uCKT8J_u-2z- FpeHKyTdvQ2vrGcel8cgl/edit" "slide=id.gd9b8e4fdae_0_ 288" \h HYPERLINK "https://docs.google.com/presentation/d/1 PHGJS6jOw-uCKT8J_u- Zz-FpeHKyTdvQ2vrGcel8cgl/edit" "slide=id.gd9b8e4fdae_O 288" \h] on external interruptions or internal conflict to spark awareness of when they're feeling bad about themselves, and it can be challenging for them to proactively identify the behaviors and experiences that lead to these feelings.. o Reducing NAC specifically through this mechanism will likely be achieved through targeting. The initial target designs are more content agnostic, so the expected % impact on NAC in H1 may be diluted because the TaMis all active teens, not just those with high NAC potential. By helping the right targeted set to reflect on their current experiences on Instagram and how those experiences make them feel, we can empower them to make informed choices, change their behavior, and take actions that reduce NAC Background: 1. In early December, Instagram publicly HYPERLINK [ "https://about.instagram.com/blog/announcements/raising-the-standard-for-protecting- teens-and-supporting-parents-online" \h ] that we're "Providing teens with more tools to better manage their Instagram experience". HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005138  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 9 of 16 a. This is so that "people feel good about the time they spend on Instagram", in particular for cases like when someone has been scrolling for a certain amount of time (tool: Take a Break) or dwelling on one topic for a while (tool: Topic Nudge). b. To continue to provide public proof points of our progress, Instagram plans to launch Topic Nudge in May and roll out Take a Break improvements to follow the Feb global launch. 2. In early December, Instagram decided that the go-forward [ HYPERLINK "https://docs.google.com/presentation/d/19JpzbdOFo6PwDnRF-AGLOL- cbtyDbNGgXQ8UJSLvpjc/edit" \I "slide=id.g1048090c7e9_8 37" \h] in this area is to "reduce experiences of problematic use (PU) and negative appearance comparison (NAC), and land a steady drumbeat of expert-backed visible products that support teen well-being". a. This definition of success assumes that over the long-term (2023-24), we progressively land measurement that covers: i. Reducing NAC ii. Reducing PU iii. | Supporting teen well-being (some subset of positive outcomes) 3. In November, Instagram decided on a [ HYPERLINK "nttps://docs.google.com/presentation/d/1vmLxi¥dD_UohE5jmq06_H6SYBtzeZ5wHNw9 tzd7wmal/edit" "slide=id.gfc3ea09a61_0 0" \h] to complete initial measurement of Negative Appearance Comparison for specific HYPERLINK [ "nttps://docs.google.com/presentation/d/1vmLxi¥dD_UohE5jmq06_H6SYBtzeZ5wHNw9 tzd7wmal/edit" "slide=id.g1016f4519c4 0 0" \h]. This was informed by the Core Data Science <> IG partnership on pTRIPS and CDS's recommendation on the [ HYPERLINK "nttps:/Ab.workplace.com/notes/1025043 198241838" \h ] (xMVP). a. In this phased validation process, the "pNAC" operational metric would be used by mid-2022 for identifying new product opportunities and for evaluating impact for the long-term "reduce experiences of negative appearance comparison (NAC)" goal. b. If all goes well, PNAC would be used for team-level goaling in 2023. c. pNAC is not proposed as a metric to be used outside Instagram or to be shared externally. 4. In November, the NAC survey at the core of the phased validation process was blocked by central survey reviewers for needing additional review given Facebook Files. 5. In mid-December, Research for societal topics HYPERLINK [ "nttps:/Ab.workplace.com/groups/userexperience/posts/5253454 121348735" \h ] to report centrally. Leaders in this new reporting structure were not present in decisions made in the prior month. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005139  Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 10 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005140 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 11 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005141 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 12 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005142 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 13 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005143 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 14 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005144 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 15 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005145 Case 4:22-md-03047-YGR Document 2756-53 Filed 02/20/26 Page 16 of 16 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-035-00005146