Meta & Children's Online SafetySSI Data
Raw text
Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 1 of 44 AMENDED Exhibit 172 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 2757-4 Filed 02/20/26 Page 2 of 44 SSI Data This document is a compilation of all the SSI data for IG Harmful Content. Additional Documentation requests: Files Files (immediately below): Daiquery queries that generated these files Results/Time series (Time Series): Publish Jupyter notebook to ANP • SSI Demographics: ssi all demographics v2.xlsx • Co-occurence (people from group X are also in group Y): co-occurrence matrix.xlsx • SSI producers of admissions in California (last 90 days): California-ssi self-harn admissions-last90days.xlsx • All desired demographics: Demo Summary Table Strategy deck 1. POLICY: Should high volume admission producers be treated as promoters? SSI admissions - Among SSI Admissions, what % of their media are admissions? * 15% of accounts (4892/32920) with at least 1 SSI admission in the last 28 days have >30% of their 28d content as SSI admissions. These Distribution of SSI Admissions as a Fraction of a User's 28d Content Production All users who had an SSI admission in 28 days, how many admissions did they have in this period? bucket cnt percent 1 1:0-0.05 21773 66% 2 2:0.05-0.10 3008 9% 3 3:0.10-0.15 1580 5% 4 4: 0.15-0.20 606 2% 5 5: 0.20-0.25 532 2% 6 6:0.25-0.30 529 2% 7 7:20.3 4892 15% 8 32920 100% 66% <5% 9% 5% 2% 2% 2% >5% to 10% 10% to 15% >15% to 20% 20% 10.25% 25% 10 30% HIGHLY CONFIDENTIAL (COMPETITOR) 15% %00 META3047MDL-143-00000002 Case 4:22-md-03047-YGR [query] Document 2757-4 Filed 02/20/26 Page 3 of 44 Distribution of SSI Promotions as a Fraction of User's 28d Content Production All users who had an SSI promotion in 28 days, how many promotions did they have in this period? bucket cnt percent 1 1:0-0.05 6356 75% 2 2:0.05-0.10 642 8% 3 3:0.10-0.15 261 3% 4 4: 0.15-0.20 96 1% 5 5: 0.20-0.25 108 1% 6 6:0.25-0.30 80 1% 7 7:>0.3 930 11% 8 TOTAL 8473 100% [query] <5% 8% 17% 3% >5% to 10% >10% to 15% 15% to 20% 20% to 25%. >25% 10 30% >30% Additional query for more information: [source] Are admits the same group of people as promoters? 6.7% of people who had a SSI admission OR promotion had both. 30% of people who had an SSI promotion also had an admission A B C 1 admit promotion fraction 2 1 0 32112 0.77912 3 1 2775 0.06733 4 0 1 6329 0.15356 5 41216 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000003 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 admito promotions 32K 6K BOTH Page 4 of 44 Venn Diagram of Admissions and Promotions [query source] Is there a distinction between high and low producers (of admissions)? Those who get had at least 1 admission and 1 promotion had 337% more admissions than those who only had at least one admission. sum_admit_a sum_admit_p sum_promotion_ sum_promotion_ admit promotion να 09 avg p50 1 0 1 10 0 1.26734 1 2 1 1 12.87207 3 2.49586 1 3 10 2.94102 1 1 10 0 2. PRODUCT: Which people need help? Supportive Hashtags 13 hashtags were identified as potential support communities where people search for help or post supportive media. I looked at: 1) the total number of media with the hashtag, 2) the number of followers of each hashtag, 3) the number of followers who are DAP, and 4) the ratio of media to unique followers. These individuals represent our current best estimate of people who would be interested in receiving or giving help. An additional analysis may be to conduct a fanout from these hashtags to additional related hashtags to see how large of a community we could detect. hashtag_name recoveryisworthit itsokaytotalk 1 2 3 selfloveclub unique media with users unique DAP hashtag/unique media with this following following hashtag hashtag hashtag users following hashtag 800600 2035 1442 393.41523 16631 54 45 307.98148 186869 893 738 209.2598 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000004 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 4 anxietywarrior 64961 281 224 231.17794 5 mentalhealthwarrior 136483 1085 798 125.79078 6 mentalhealthmatters 536261 3313 2573 161.86568 7 endthestigma 420133 1847 1436 227.46779 8 edwarrior 1317708 1276 940 1032.68652 9 ptsdwarrior 23489 114 91 206.04386 10 sadgirlsclub 107053 604 449 177.24007 11 itsokaynottobeokay 118299 792 627 149.36742 12 anxiousbutcourageous 937 10 8 93.7 13 recoveryispossible 584280 1639 1153 356.48566 14 TOTAL 13943 10524 [query] [query source] Page 5 of 44 Reach: DAP Exposed to SSI Content How many DAP were exposed to known Suicide and Self-injury media? NOTE: all data in this Reach section is ONLY for Suicide and Self-Injury (ED is excluded). D10500775: all queries on SSI content Topline 5.1M DAP are exposed to suicide or self-harm admission or promotion content. That's 0.75% of DAP. That's using data from October so far. https://fburl.com/daiquery/xr2woc99 Daily Rates SM 7M SM 4M 3 M 4. Θα 10. Oct 14. Oct 16. Oct 18. Oct 20. Οα 22. Oct 24. Oct dap exposed Figure | DAP Exposed to SSI Content. Daily rate over time. There are some pronounced daily spikes. Most of these spikes are from very viral content (meme-like in quality) that have SSI content. https://fburl.com/daiquery/425mmx15 By Content Type Most of the exposures and impressions are on Admission content. Admissions account for~3/4 of impressions. DAP category Exposed Percent of Impressions Impressions /d HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000005 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 6 of 44 1 admission 4,061,145 9,775,978 73.59% 2 other 1,600,881 1,712,147 12.89% 3 promotion 1,679,647 1,795,867 13.52% 4 TOTAL 13,283,993 100.00% Table | DAP Exposed and Impressions by Type of Suicide and Self-injury Content (promotion, admission, other = suicide videos), hups://fburl.com/daiquery/yp2py738 By Content Type Viewers of SSI content are mostly seeing admission content for suicide (60%) and self-harm (13%). DAP Impressions category dub Exposed (%) 1 admission suicide admission 3,140,403 8,027,328 60.43% 2 admission self harm admission 1,015,394 1,748,650 13.16% 3 other other 988,609 1,033,166 7.78% 4 promotion self_harm_promotion 897,018 945,178 7.12% 5 promotion suicide promotion 801,893 850,689 6.40% 6 other suicide video 619,113 678,982 5.11% 7 TOTAL 13,283,993 100.00% https://fburl.com/daiquery/zzmg9zr5 By Surface and Content Type • most of the impressions are happening on Home and Explore (not surprising), and most of these are SSI admissions. • On Home, we would have to do more to understand if this is content coming from a connected person (the viewer is following that user) vs. from a connected hashtag (the viewer is following a hashtag that spreads this content to Home). surface_tag category DAP Exposed Impressions/d 1 home admission 1,872,009 4,480,932 2 explore admission 1,009,828 1,744,696 3 home other 920,546 989.215 4 home promotion 756,991 879,765 S explore promotion 403,464 420,275 6 explore other 279,776 302,502 7 profile admission 152,561 381,049 8 hashtags admission 101,769 362,220 9 profile other 59,437 63,920 10 profile promotion 56,414 62,006 11 hashtags promotion 28,785 34,358 12 admission 17,182 35,265 13 saved admission 11,961 23,114 14 hashtags other 10,886 11,935 CO Actioned/Labelled Content BY Admission/Promotion and by SSI Type (S, S-I, ED) ॐ * Most content is self-harm admission (44.5%) or suicide admission (28%), and most of the responsible users are for this same type of content (37.7% and 29.2%, respectively). Table below is the 28d totals CO labelled content for SSI (from Oct 1-Oct 28). This content is DISTINCT (meaning duplicates are only counted 1 time) Distinct Responsible Users (%) 37.72% 29.22% Distinct violation_type Group Content (%) 1 SELFINJURY self harm admission 38,159 44.50% 2 SUICIDE suicide admission 24,027 28.02% 18,630 14,431 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000006 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 7 of 44 3 SELFINJURY suicide promotion 6,109 7.12% 5,151 10.43% 4 EATINGDISORDER eating_disorder_admission 5,118 5.97% 923 1.87% S SUICIDE suicide video 4,786 5.58% 4,106 8.31% 6 SELFINJURY self_harm_promotion 4,746 5.53% 3,735 7.56% 7 EATINGDISORDER 1,606 1.87% 1,606 3.25% 8 EATINGDISORDER eating_disorder promotion 1,175 1.37% 778 1.58% 9 SUICIDE 22 0.03% 21 0.04% 10 SELFINJURY 3 0.00% 3 0.01% 11 TOTAL 85,751 100.00% 49,384 100.00% Table | Content Labelled by CO for SSI (28-d total) query: https://fburl.com/daiquery/vdp58xk7 Escalations There are 2 tiers of escalations (OS --> HYD (2 review queues) --> LERT). SSI Data Primer Demographics Groups The demographics come from users who fall into each group in the last 30 days. The only exception to this is 1) Hashtag Follower: 13 Support Hashtags (these are user who followed the hashtag at ANY time); 2) DAP (1-day); and 3) MAP (1-day). CATEGORY GROUP 1 PRODUCERS Producer: ED Disabled Profile 2 Admission: ED 3 Promotion: ED 4 Admission: Self-Harm 5 Promotion: Self-Harm 6 Admission: Suicide 7 Promotion: Suicide 8 Video: Suicide 9 Bullying Producer 10 Harassment Producer 11 DEFINITION Produced ED content that led CO to remove the user's profile Produced ED content that CO marked as an admission (and left on the sit Produced ED content that CO marked as promotion (and deleted from th Produced self-harm content that CO marked as an admission (and left on Produced self-harm content that CO marked as promotion (and deleted fr Produced suicide content that CO marked as an admission (and left on th Produced suicide content that CO marked as promotion (and deleted fro Produced a suicide video that CO deleted Produced Bullying content that CO deleted Produced Harassment content that CO deleted 12 СНЕСКΡΟΙΝΤ | Admission Checkpoint for ED 13 Admission Checkpoint for Self-Harm 14 Admission Checkpoint for Suicide Placed in the ED Admission Checkpoing Placed in the Self-Harm Admission Checkpoing Placed in the Suicide Admission Checkpoing 15 16 SUPPORT HASHTAGS Hashtag Follower: 13 Support Hashtags 17 Hashtag Visited 13 Support Hashtags Follows 1 of the 13 hashtags identified as a support community (this follo happened at any time; this group is NOT filtered by the 30-d rule listed ab Visited 1 of the 13 hashtags identified as a support community 18 19 SENSITIVE HASHTAGS Hashtag Interstitial for ED Visited one of the Sensitive ED Hashtags that have an Interstitial page 20 Hashtag Interstitial for SSI 21 22 23 Help Page for ED Help Page for Suicide Visited one of the Sensitive SSI Hashtags that have an Interstitial page Visited the Help Page for ED Visited the Help Page for Suicide 24 25 REPORTERS OF SSI AND HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000007 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 26 27 28 29 30 31 32 33 34 35 REPORTERS 10 BULLYING AND HARASSMEN Reporter (Successful) ED Disabled Profile Reporter (Successful) Admission ED Reporter (Succesful) Promotion ED Reporter (Successful) Admission Self-Harm Reporter (Successful) Promotion Self-Harm Reporter (Successful) Admission Suicide Reporter (Successful) Promotion Suicide Reporter (Successful) Video Suicide Reporter SelfHarm Reporter (Successful) Bullying Page 8 of 44 Successfully reported a user that led CO to remove the user's profile Successfully reported a user that led CO to mark as an ED admission Successfully reported a user that led CO to mark as an ED promition Successfully reported a user that led CO to mark as a self-harm admission Successfully reported a user that led CO to mark as a self-harm promotio Successfully reported a user that led CO to mark as a suicide admission Successfully reported a user that led CO to mark as a suicide promotion Successfully reported a user that led CO delete a suicide video Reported a user for Self-Harm (includes both successful and unsuccessful T 36 37 38 Reporter (Successful) Harassment Reporter CyberBullying 39 REFERENCE CO Actioned Any HC 40 41 42 CO Actioned Any OC DAP (1-day) MAP (1-day) Successfully reported a user that CO actioned for bullying Successfully reported a user that CO actioned for harassment Reported a user for CyberBullying (includes both successful and unsucces Produced content that CO actioned for any HC violation (30 days) Produced content that CO actioned for any OC violation (30 days) DAP on 1 day MAP on 1 day Population Overview $ * ॐ There are 30% as many users responsible for HC as compared to OC in a month (30 days). There are 26% as many Reporters of SelfHarm as there are of Cyberbullying in a month (30 days). There are 20% as many users implicated in any production of SSI + ED content versus those implicated in Cyberbullying production. 279K unique users interact with the interstitial for SSI Hashtag pages in a month (30 days), and 181K interact with the interstitial for ED pages. • More people have Admissions than Promotions • Admissions for Suicide are 2.8 X higher than Promotions for Suicide • Admissions for Self-Injury are 5.8 X higher than Promotions for Self-Injury • ED admissions and promotions are very similar in size; there are actually more promotion users (703) than admissions (613). HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000008 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 CO Actioned Any OC CO Actioned Any C Reporter CyberBullying Reporter Sal Harm Reporter (Ruccessful Video Suicide Reporter (Successful) Promotion Suicide sabe Reporter (Successful) Admission Suicide zas Reporter (Successfull Promotion Self-Harm 3,357 Reporter (Successful) Admission Self-Harm Reporter (Succesful Promatice XD ses Reporter (Successful) Admission 20 sje Reporter (Successful) FO Disabled Proßle 1,122 Help Page for 10 sds Hashtag Interstrial for SS Hashtag intersticial for Hashtag Visted 13 Support Hashtags 1721 Heslitag Follower: 13 Support Hashtags Admission Oreckpalat for Suicide Admission Checkpoint for Self-Harm Admission Checkpoint for 101,968 Harassment Producersex Bullying Producer Video: Suicide a Promotion Suicides o Admission: Suicide 15 Promotion: Selfisarm 1,500 Admission: Self-Harm 2017 Promotion 80 703 Admission E siz Producer: 80 Disabled Profile 1,822 ۵ 2X30X Number of Linique Users Number of Unique isers 340000 Page 9 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000009 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Video: Suicide Promotiom: Self-Harms Admissions Self-Harm Promotion ED Admission E Producer: ED Disabled Prottle Number of Linique Lisers 2:0000 Number of Unique isers Page 10 of 44 SSI by Age Main takeaways While teens only make up 18% of DAP, the vast majority of users who admit and promote self-harm and suicide are teens (45%-60%). The visitors to the SSI Help Page are similar: only 46% are teens. • Reporters of self-harm and suicide are also disproportionately represented by teens (52%-56%) The visitors and followers of the 13 supportive hashtag pages looked more like DAP and MAP users: 18% of the visitors to these pages and 15% of the hashtag followers were teens. This suggests that this list of hashtags may be skewing toward older people, though it may also indicate that older people are best able to seek out support communities online. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000010 28 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 11 of 44 100% 40% Z0X 10% Actioned Arny HC Reporter (Successful) Harassmen Reporter (Successful) Bullying Reporter Successfull Video Side Reporter (Successfus) Pramantine Suickle Reporter (Successful) Admission Suicide Reporter (Successfulj Admission Saff-Harm Reporter (Succesful Prasnation 20 Reporter (Successful) Admission Nele Page for Suicide Help Page for Hashtag interstalar for ٤٥ 00000000000 Hashtag Visited 13. Support Hashtags 200 Admission Checkpoint for Self arm Video: Suicide Promotion, S Admission: Succe Adrian: 000000000000000:人心。 2008 Her Page 15 Hanthong.tit AGE Checkprost (Suicide Chachpulkst Reporter ISIK P ***** Early High School (13-15) Late High School (16-18) Colege (19-22) Early Work (28-35) & Unknowa Repor Gender Main takeaways (All results should consider that there are a large number of people with unknown genders) SH and suicide admission, and suicide promotion is more common among females; however, suicide promotion is more common among males. • Proportionally more females, compared to males, seek help for SSI (41% versus 16%). HIGHLY CONFIDENTIAL (COMPETITOR) Vitinden Larte work 36+0 Scary work )23-35 ص( Culinge 09-22 yoi Larte META3047MDL-143-00000011 18 15 go) Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 12 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000012 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 13 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000013 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 14 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000014 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 15 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000015 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 16 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000016 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 17 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000017 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 18 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000018 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 19 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000019 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 20 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000020 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 MAP (1-day) CO Actioned Any C Actioned Any C Reporter Cyberbullying Repartar (Successful Harassment Reporter Successful) Bullying Reporter Successful) Video Suicide Reporter (Successful) Promotion Suicide Reporter (Successful) Promotion Self-Harm Raperter (Successful) Admission Self-Harm Reporter (Successful) Admission 50 Reporter (Succestul) ED sabied Profile Help Page for Suicide Help Page for 80 Hashtag Interstitial for SSi flashtag Interstitial for 10 Hashtag Visited 18 Support Hashtags Hesiktag Follower: 13 Support Moskitogs Admission Checkprint for Suicide Admission Checkpoint for Self-Harm RRRRRRRRRRRRRRRRRRRRRRRY Harassment Producer >> Bullying Producer Promotion: Suicide Admission: Suicide Promosian: Self-Harm Admission: Self-Harm Promotion 10 Admission CD 17/17 DAP 0.30 Page 21 of 44 Account Reach Account reach is the number of distinct accounts that a user reach in 7 days. We take the max over 90 days to determine their maximum reach over this time period. NOTE: the y-axis is on the log scale, so subtle differences in this chart can mean A LOT. Suicide Video Producers and Suicide Promoters have some very high reach accounts: the 95th percentile has a maximum reach of more than 780K and 344K, respectively. o this is significantly higher than the 95th percentile for Bullying (17k) and Harassment (14k). • There are more high reach accounts (defined as having a max 7d reach of >=30K) than DAP among Suicide Admissions (40% higher vs. DAP) and Suicide Videos (20% higher vs. DAP). • Promotion accounts (for Suicide and Self-Harm) are in line with DAP in terms of the number of high reach accounts • there are very few high-reach ED accounts (promotion, admission, disable profile) • Reporters have higher reach than DAP, but the median is not substantively different (a 2X difference, but only a difference of~100 users reached). * HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000021 Max Reach Over 30 Days (Bandh 100 Рефании. 10 Diabel Priße Admission 10000 ২০১ 0000001 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 22 of 44 ২৪১ 525250 Aditation Bu Viden Sidd www え Admission Chur 288 ৫০৫ RRRRRRRRRRRRRRR GLAMOND BHA 京 २७१ Max Reach Over 90 Days (Reach = 7d) .......... 20 ২১ ৫১ 39 ২৫ ১০ ४० adam Wa 13 Ruppert sistidage besoning insompikini be Help fragatur Et Sterior Sumanda) ED Cisabad Free Supotter (Succesult Admiralon - Пиратья Брюки Proton Sal-Hem Roll Acussion Buick ২ Reporter Stinopathuat Pentatless Suichla Πώματα (δυσηιεnika Vities Suibitte Flopader Shar ४ .......... 2 pil tud uses this definition b/c it corresponds to the p9995). Having a maximum 7d reach of more than 30K users (over a 90d period) qualifies someone as a high reach account (Election Integrity HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000022 ووووو وووووووون २८ 240 २०৫ Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Number of Users with >=30K Max Organic 7d Reach MAP (1-day) CO Actioned Any OC CO Actioned Any Reporter Cyberbullying Repartar (Successful Harassmen Reporter (Successful) Bullying- Reporter Successful) Video Suicide Reporter (Successful) Promotion Suicide Reporter (Successful) Promotion Self-Harm Reporter (Successful) Admission Self-Harm Reporter (Successful) Admission & Reporter (Succestul) ED sabied Profile Help Page for Suicide Help Page for 80 Hashtag Interstitial for SS flashtag Interstitial for 0 Hashtag Visited 18 Support Hashtags Hesiktag Follower: 13 Support Moskitogs Admission Checkpoint for Suicide Admission Checkpot fer Self Harm Harassment Producer Bullying Producer Promotion, Suicide Admission: Suicide Promotion: Self-arm3 Admission: Self-Harm Promotion 10 Admissioen 313000 Page 23 of 44 Users With 30x Max Organic 70 Reach This chart scales the number of high reach accounts by the number of people in each group. • As a fraction of group size, the producers have many more high reach accounts as compared to DAP. • 24% of Video Suicide accounts are high reach accounts • 13% of Suicide Promotion accounts are high reach HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000023 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Reporter (Suonentul Video Sui Reporter (Bacsess Rapeter (Successful) ED Disabled Pre Kala Page box Suicide Help Page tox & Hashtag Visted 25 Suobert Hashtag Hashtag Flower: 13 Support H Admission: 20 Producer 50 Dabled Pete Percent of users with v30K Max Organic Reach FB Connected 8 Those involved with SSI are about 1/4 to 1/2 as likely to be FB connected vs. IG DAP Visitors and Followers of the 13 Support Hashtags are again in line with DAP and MAP • Among producers, ED producers are the least likely to be FB connected Page 24 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000024 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 25 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000025 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 26 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000026 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 27 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000027 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 28 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000028 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 29 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000029 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 30 of 44 involves taking down all the associated media with an account. Our actions actually affect more unique users for self-injury and suicide (each, separately) than eating disorder. • Entities actioned: 13 ED :: 2 self-injury :: 1 suicide • Unique users: 4 self-injury :: 2 suicide :: 1 ED Daily average - data from last 30 days entities_actioned_a unique_users_a violation_type entity_type_name 1 EATINGDISORDER IG LIVE VIDEO BROADCAST vg vg 2.5 2 EATINGDISORDER INSTAGRAM COMMENT V2 3.9 3 EATINGDISORDER INSTAGRAM MEDIA V2 3789.6 1 3.2 82.1 4 EATINGDISORDER INSTAGRAM USER V2 53.3 53.3 5 SELFINJURY IG LIVE VIDEO BROADCAST 2 1 6 SELFINJURY INSTAGRAM COMMENT V2 34.2 32.2 7 SELFINJURY INSTAGRAM MEDIA V2 615 344.8 8 SELFINJURY INSTAGRAM USER V2 2 2 9 SUICIDE IG LIVE VIDEO BROADCAST 1.4 1.4 10 SUICIDE INSTAGRAM COMMENT V2 1 1 11 SUICIDE INSTAGRAM MEDIA V2 275 181.3 People All in the last 30 days (except for DAP AND MAP, which are in the last day) Producers are less likely to have private accounts than reporters, DAP, MAP, users checkpointed for SSI or eating disorders, sensitive hashtag page visitors, and help page visitors. • Everyone involved with SSI and eating disorders (producers, reporters, help page, hashtag, checkpoint) are more likely to * be in early high school or late high school than DAP, MAP, and producers of other OC/HC Producers of suicide content are more male; producers of ED much less likely to be male • Reporters of suicide, selfinjury, and ED are more fernale than DAP, MAP * Visitors to sensitive hashtags and help pages are more female than DAP, MAP * the US or DE are good places to start b/c of the ratio of those involved with SSI and ED vs. DAP/MAP A C D E G 1 is_business_pr early_high_sch late violation_type STATUS ent is private ofile [is_employee ool ol 2 eating_disorder checkpoint 2023 42% 11% 0% 17% 29 3 self harm checkpoint 18118 33% 24% 0% 25% 28 4 suicide checkpoint 47652 39% 18% 10% 27% 31 5 eating_disorder hashtag page 171441 46% 10% 0% 14% 18 6 ssi hashtag page 282137 51% 13% 0% 25% 24 7 eating_disorders_visitor help_page 9592 46% 14% 0% 18% 22 8 suicide_prevention_visitor help_page 57878 47% 14% 0% 24% 25 9 EATINGDISORDER producer 2461 25% 8% 0% 15% 20 10 SELFINJURY producer 10202 33% 26% 0% 26% 28 11 SUICIDE producer 5127 28% 28% 0% 17% 23 12 EATINGDISORDER reporter 2695 51% 13% 0% 19% 25 13 SELFINJURY reporter 47394 48% 19% 0% 28% 27 14 SUICIDE reporter 31082 53% 16% 0% 28% 27 15 REFERENCE any actioned 6482349 21% 5% 0% 4% 7% 16 REFERENCE dap 640313 50% 6% 0% 7% 11 17 REFERENCE map 1030421 41% 5% 0% 7% 11 1. Producers HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000030 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 31 of 44 2. Viewers 3. Reporters 2 kinds of reporters: all reports vs. actioned reports @ Interstitial: The interstitial appears as a pop-up over top of hashtags that have been marked as violating from 2 separate sources: hard-coded list of hashtags from CO: https://admin.instagram.com/admin/tags/safety/, and into the "overwritten list" option, choose the "sensitive" tier and it would be all tags under ssi and eating disorders daily-changing list of hashtags: https://fburl.com/jldvzjt8 NEED TO GET THIS Note: the interstitial does not prevent users from visiting the hashtag page; it simply prompts them to be redirected to a separate help page for suicide, self-injury, or eating disorders (NEED TO GET INTERSTITIAL FOR EACH OF THESE 3 HELP PAGES) • e.g., search for #suicide Link: and then click "Overwritten List" and select "sensitive" from the tier dropdown Detalle Overwritten Tags Browser Choose a tier sensitive Filter by tag prefix Input penarios ( e inmi Tag Name cut4bieber cutforbieber cutter cuttingtorbieber Tier Category sensitive ssi sensitive ssi sensitive sensitive ssi 4. Impressions on Hashtag Page Number of unique people who hit interstitials for each category on 1 day (https://fburl.com/daiquery/0ovxr2v7) A 1 average distinct Row Labels users/d 2 eating disorder 8413.92857 3 SSI 13498 daiquery 258190988295729 Number of hashtags that re-route to the interstitial (https://fburl.com/daiquery/qld624io): category hashtags 1 regulated firearms 14 2 eating disorder 162 4 ssi 2 A 1 3 > HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000031 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 4 5. Interstitial Interaction NOT LOGGED: the actual interstitial is not logged (pending 6. Visits to the Help Center The interstitial is actually logged in two places: ॐ "Can we help?": Page 32 of 44 creating logger config) BLOCK; Meeting with on Th 7/11 needs to set up a logger config- T31145197 • help.instagram.com • compassion_resource_events o need to make sure that ig_source is set, then since that param probably isn't whitelisted, need to join datr_cookies from first query to all other events from the rest of the table https://help.instagram.com/resources/63617265/viewer 4500 4000 3500 3000 suicide prevention Help Pg ast visit eating disorder Help Pg ast visit 1500 2000 1500 2000 May In the last month (2018-06-11 through 2018-07-10): 1 2 resource_type eating_disorders_visitor suicide_prevention_visitor Visits to Help Page/d June miy num_visitor S percent 1st first_visit visit 411 308 75 2491 1831 73 1st visit = we didn't see a visit to the Help Page for this person and for this particular Help Pg (within eating disorder or suicide) in the last 90 days. Repeat Visits How many users had a repeat visit to the Help Page in 30 days? (2018-08-06 through DS-30) [query] num visits number num_visit*num ber 1 1 2464 2464 2 2 192 384 3 3 17 51 4 4 9 36 5 5 2 10 6 6 2 12 7 8 3 24 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000032 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 33 of 44 8 10 3 30 9 15 1 15 10 Total 2693 3026 Help Page Funnel query for sensitive hashtag visits: just count the number of impressions on the 219 sensitive screen hashtag pages. That query is listed above. query for Help Page Visits: https://fburl.com/daiquery/pfi5yel4 WHO NEEDS HELP?: VISIT SENSITIVE HASHTAGS Daily Rate 219 sensitive hashtags 570/d Partner Page: (4% of total) 500000000000000000000 493/d Viewer Page: (3.6% of total) 13.5k/d Unique People (cially) 2.0k/d Unique People (daily): (16% retentio (1.6k/d, 82% 1st time visitors) 333/d Visit Friend Page: (2% of total) "analysis does not include dally-changing hashtags Checkpoints: suicide checkpoint (does not come from user reports) deleted content checkpoint (relies on user reports) DATA SOURCE: instagram checkpoint Admission Checkpoint This is distinct users. median_enr e 1 2 suicide 3 self harm checkpoint_type olled_peopl mean_enroll median_clea mean_clear eating_disorder 81 ed_people red_people ed_people 83.96429 80 78.82143 2157 2184.32143 2107 2122.14286 665 681.57143 649 655.32143 https://fburl.com/daiquery/3t3volmu For checkpoints, there are more enrollments for suicide than for self-harm and eating disorders. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000033 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 34 of 44 2.5 8 28 358 500 0 4. jul 6. jul 8. jul suicide enrolled_people 12. jun 14. jun 16. jua 18. jun 20. jun 22. jun 24. jun 26. jun 28. jun 30. jun 2. jul self harm enrolled_people eating disorder enrolled_people https://our.intern.facebook.com/intern/datainsights/1877389722562853/ https://fburi.com/daiquery/w5x6vgbv Stages of Checkpoint https://fburl.com/daiquery/n4krsdio This is NOT distinct users. ACTION step_name checkpoint_type avg_cnt 1 enroll eating disorder 93.78571 2 clear 3 proceed eating disorder get tips eating_disorder_get_tips eating disorder 2.27273 eating disorder 2.36364 4 reset eating_disorder_get_tips eating_disorder 1.35714 5 clear eating_disorder_offer_help eating_disorder 76.78571 6 proceed eating disorder_offer_help eating disorder 103.03571 7 select eating disorder_offer_help eating disorder 98 8 start eating disorder_offer_help eating disorder 98 9 clear eating_disorder_select_help_method eating_disorder 7.60714 10 proceed eating disorder select help method eating disorder 13.25 11 reset eating_disorder_select_help_method eating disorder 1.82609 12 clear eating_disorder_talk_with_friend eating_disorder 1.47059 13 proceed eating disorder talk with friend eating disorder 1.47059 14 reset eating disorder talk with friend eating disorder 1.18182 15 clear eating disorder talk with volunteer eating_disorder 1.7 16 proceed eating disorder talk with volunteer eating disorder 1.9 17 reset eating_disorder_talk_with_volunteer eating_disorder 1.27273 18 enroll self harm 735.39286 19 clear self_harm_get_tips self harm 9.39286 20 proceed self harm get tips self harm 9.42857 21 reset self harm get tips self harm 6.78571 22 clear self harm offer_help self harm 560.28571 23 proceed self harm_offer_help self harm 1026.96429 24 reset self_harm_offer_help self harm 1.5 25 select self harm offer help self harm 835.53571 26 start self harm offer_help self harm 835.53571 27 clear self harm select_help_method self harm 111.92857 28 proceed self harm select_help_method self harm 197.03571 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000034 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 35 of 44 29 reset self harm select help_method self harn 23.5 30 clear self harm talk with friend self harm 13.46429 31 proceed self harm talk with friend self harm 13.57143 32 reset self harm talk with friend self harm 4.96429 33 clear self harm talk with volunteer self harm 13.25 34 proceed self harm talk with volunteer self harm 13.39286 35 reset self harm talk with volunteer self harm 7.85714 36 enroll suicide 2689.46429 37 clear suicide get tips suicide 44.89286 38 proceed suicide_get_tips suicide 45.75 39 reset suicide get tips suicide 35.39286 40 clear suicide_offer_help suicide 2160.82143 41 proceed suicide offer help suicide 3512.42857 42 reset suicide_offer help suicide 3.41667 43 select suicide_offer_help suicide 3281.03571 44 start suicide_offer help suicide 3281.03571 45 clear suicide_select_help method suicide 307.78571 46 proceed suicide_select_help method suicide 633.60714 47 reset suicide select help method suicide 58.35714 48 clear suicide talk with friend suicide 58.71429 49 proceed suicide talk with friend suicide 59.39286 50 reset suicide talk with friend suicide 29.35714 51 clear suicide talk with volunteer suicide 46.5 52 proceed suicide talk with volunteer suicide 48 53 reset suicide talk with volunteer suicide 35.64286 Promotion Checkpoint This is the deleted content checkpoint and should be no different than the number of actioned media (distinct by user_id) from above. SKIP FOR NOW B/C SHOULD NOT PROVIDE NEW INFORMATION. Finstas and Rinstas Finstas are fake Instagram accounts and Rinstas are real Instagram accounts. To Do distribution of SSI admission (histogram of number of SSI admissions vs. users) for users marked as SSI admission, what % of their content is SSI admission? look for 30% threshold of admissions to make policy argument is the account dedicated to SSI? or regular person going through some stuff? for high volume SSI admits, do they visit the Help Page? vs. other users? for high volume SSI admits, what is their pattern over time? time until second repeat (among those with 2+ SSI admissions, what is the time until second SSI admission post?) are they more likely to be actioned for other bad content? (e.g., bullying) finstas vs. rinstas shared device ID is the finsta promoting or does it need help? XX% of people marked as having a Self Injury admission also have a Self Injury Promotion. XX% with a Self-Injury admission also have a Suicide Admission PRODUCT: Should low volume producers to be supported? (targets for help and a new product) finsta vs. rinsta low volume producers also viewing this content? distributional argument what does the population of low volume producers look like? HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000035 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 are they mixed topics account? do they get engagement on their admission content? what's their DM behavior? profile and activity and time spent information engaging with posts is bullying the real problem? growth over time are there helpers? Page 36 of 44 'DM me for support" in the bio consumers are there high volume consumers vs. low volume? are they commenting and helping? are they reporting? are they posting anew? are they also abusers (admission, promotion)? consume posts of followers (friend) vs. active searching for content am I helping? reporting? connected vs. not actions once they see the content reporting viewing more posting are there candidates for helping in this pool of users? Post about a trigger warning (Jess) Building a community Tools/Ops time spent reviewing content number who are repeat offenders volume, action rate, time to resolution all compared to other areas can we use the FF dashboard for this? Definition of Finsta vs. Rinsta device ID (use as the first crack for directional) shared FB login shared cookie pool defintion of high volume power users: people responsible for most deletions (number of deletions/reviews) number of reviews vs. number of actions wasting review time impressions on their content reach of these accounts high searchers -are they also high volume producers? do they DM with posters? Leftovers If the goal is to help people, xxx affects the most. If the goal is to remove bad content and minimize harmful impressions, then xxx affects the most Ideas HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000036 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 37 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000037 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 38 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000038 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 39 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000039 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 40 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000041 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 41 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000043 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 42 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000045 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 43 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000047 Case 4:22-md-03047-YGR Document 2757-4 Filed 02/20/26 Page 44 of 44 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-143-00000048