How much topics people see are linked to appearance comparison on IG

Raw text

Transcribed text as recorded, including headers and formatting fragments.

How the topics people see are linked to appearance comparison on IG · Last edited March 9, 2021 · 18 minute read [Stamp: FILED 2024 JAN 29 PM 4:52 / CLERK & MASTER / DAVIDSON CO. CHANCERY CT / [signature] D.C. & M.] How the topics people see are linked to appearance comparison on IG How does the content people see on Instagram relate to negative feelings about their bodies? In this study, we combine survey measures of appearance comparison (how much people think Instagram makes them feel worse about their bodies or appearances) with server data about the topics they had recently seen on Instagram. Takeaways - Some topics are likely to make viewers, especially teen girls, feel worse about their bodies: - Fashion and beauty (especially makeup, skincare, hair removal, nail art) - Relationships (which includes dating, romance, sex and intimacy) - Western pop stars who emphasize their bodies (e.g., Selena Gomez, Cardi B, Ariana Grande) - Images that emphasize women's bodies generally (e.g., cleavage, swimwear) - Collectively, these topics comprise 1/4 of what people see on Instagram (1/3 for teen girls). - On the other hand, seeing content that emphasizes a collective identity (e.g., sports fandom, tragedy, religion, politics) or an activity (crafts, gardening, puzzles, TV) rather than the poster's body, is associated with less appearance comparison. - Appearance comparison is being driven by top accounts and content from non-friends. When friends post about problematic topics (e.g., beauty), it doesn't trigger appearance comparison the same way content from non-friends and celebrities does. - For every piece of friend content a teen girl sees, she sees 5x as many pieces of content from top accounts. Product recommendations We should reduce exposure to celebrity content about fashion, beauty and relationships, while increasing exposure to close friend content of all kinds. Potential product directions include: - Demotions on Explore and Reels using topic and XRay features identified in this note - Not recommending celebrities to follow that post primarily fashion and beauty content - Boosting content that elicits a collective identity (e.g., sports, disaster relief) or activity (e.g., crafts, puzzles) - Making it easier for teen girls to find and consume content from their friends, especially close friends. - Encouraging people toward other content or other activities after they consume or seek out large volumes of fashion/beauty/relationship content. - Applying "sexually suggestive" non-recommendable guidelines more aggressively for teen viewers. - Providing better content controls to allow people to hide sexually suggestive photos and those that emphasize women's bodies or faces. Method The analyses below combine results from a survey with log data of survey respondents' experiences on Instagram for the two weeks prior. All analyses control for viewer demographics and triangulate with qualitative research where appropriate. Survey: We surveyed 50,590 people in 10 countries (Australia, Brazil, France, Germany, Great Britain, India, Japan, Korea, Mexico, United States) in December 2020. See our previous note for more details (full survey). Appearance comparison is the extent to which people think that Instagram makes them feel worse about their bodies or appearances. It is a specific form of negative social comparison. CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057040 We operationalize it as the average of three survey questions derived from previous research, where higher values are worse: - How often do you see posts on Instagram that make you feel worse about your body or appearance? - How often do you compare your appearance to the appearance of people on Instagram? - How much pressure do you feel to look perfect on Instagram? Note that appearance comparison as defined in this note is NOT simply feeling bad about one's body (body dissatisfaction); it's the negative impact that people ascribe to Instagram -- whether they think Instagram is making those feelings better or worse. Behavioral data: We paired survey data with the content respondents viewed in the 14 days prior to the survey on Feed, Stories, Explore, Profiles, and Reels, and identified the topic(s) associated with each piece of content using the Facebook Interest Taxonomy for Instagram (FIT), a machine learning model that takes text, image, and video features into account. We also examined some XRay/Cortex features (concepts within the images and videos themselves). Interpreting data: These findings are correlational, not causal. To make product recommendations from them, we: 1. Triangulate findings with qualitative research that helps us understand which of these correlations are likely to be causal. Recent qual work includes: Teen Girls, Body Image and Social Comparison, and Social comparison on Instagram. 2. Control for demographics of respondents (gender, age reported on survey, country, L28, follower count, following count) in a regression to reduce selection effects. Findings People who see a greater proportion of content about fashion (which includes beauty), relationships (which includes dating and romance), performing arts, and music feel like Instagram makes them feel worse about their bodies and appearances (Figure 1). Collectively, these topics comprise 27.4% of all content viewed by respondents with a FIT topic (33.2% of what teen girls saw). [Figure 1: horizontal bar chart, "Appearance comparison on IG and topics viewed." Topics ranked from most associated with worse appearance comparison to most associated with better appearance comparison: Relationships Friends & Family, Fashion & Style, Performing Arts, Music & Audio, Education & Learning, Science & Tech, Health & Medical, Fitness & Workouts, Social Issues, Children & Parenting, Business Finance & Economics, Holidays & Celebrations, Books & Literature, Food & Drink, History & Philosophy, Animals & Pets, Home & Garden, TV & Movies, Travel & Leisure Activities, Games Puzzles & Play, Visual Arts Architecture & Crafts, Politics, Vehicles & Transportation, Religion & Spirituality, Crime & Tragedy, Sports. X-axis: regression coefficient controlling for demographics, roughly -0.10 to 0.10.] Figure 1. Association between topics viewed and appearance comparison on Instagram, controlling for viewer demographics. People who see more content about relationships, fashion, performing arts, and music experience worse levels of appearance comparison. Darker bars are statistically significant at p < 0.01. Results vary somewhat when narrowing the analysis to (1) only teen girls or (2) only women (see ANP), but fashion emerges consistently as a topic that is strongly associated with appearance comparison. Fashion content is problematic (see Figure 2 for examples). Previous research reinforces the idea that seeing fashion posts makes viewers feel worse about themselves because Instagram feels "realistic" but with celebrity standards of reality. Interviews with teen girls indicate that any individual celebrity's post isn't problematic, but seeing multiple pieces of content from celebrities amplifies the effect, particularly when their friends mimic celebrities' beauty and fashion standards. [Image: grid of six fashion/style photos of women in outfits/poses, examples of "Fashion & Style" content] CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057041 Figure 2. Example content from the "Fashion & Style" FIT topic. Many of these posts focus on women's bodies. People who see a greater proportion of content like this feel like Instagram makes them feel worse about their bodies or appearances. What about Music and Performing Arts topics? Why are they "bad"? That music and performing arts are associated with high levels of appearance comparison may be surprising: music, per se, has nothing to do with one's body. However the impact on appearance comparison is likely due to the emphasis on bodies by pop stars, models, and actors that are in the Music & Audio and Performing Arts FIT topics. Fashion is the top overlapping topic with Music & Audio, though the overlap is small (less than 1% of Music & Audio posts are also classified as Fashion). As we show in the subtopic analysis later, the impact of these topics is largely driven by western pop stars. Figure 3 shows examples of content from the Music & Audio and Performing Arts categories, showing why some "Music" posts may be concerning: [Image: grid of six photos from Music & Audio / Performing Arts categories -- close-up face shot, black-and-white photo of a person in a sports-bra-style top holding a camera, a woman posing in a strapless top, a mirror selfie, a woman in a bikini top, a woman in a crop top and skirt with arms raised] Figure 3. Example content from the Music & Audio and Performing Arts categories that may cause viewers to feel worse about their bodies. What about the Relationships topic? The "Relationships, Friends, & Family" topic is more complicated. It includes subtopics related to dating, romance, sex, and intimacy, all of which may be either harmful or beneficial for the viewer (or both), depending on the viewer's context (e.g., mood, relationship status), and what the post emphasizes (see Figure 4). The subtopic analysis later in the note does not identify any individual subtopics within Relationships that are, on their own, associated with greater appearance comparison. However, the "who posted it" analysis later in this note offers some clues. Relationship content is only associated with appearance comparison when posted by top accounts (celebrities) or non-friend accounts. Friends posting about relationships have no statistically significant association with appearance comparison. Thus, it may be the case that celebrities post more idealized relationship content (including photos of couples with attractive bodies), or that people know more about their friends' struggles, so relationship content from friends does not cause people to feel worse about their own bodies. Some of the content in Figure 4 seems "vulnerable" -- people expressing their difficulties finding a relationship; this would normally reduce comparison by showing viewers that the posters' lives aren't perfect. However, reaction to these posts may depend on the viewer's mood, and many of these relationship posts still emphasize physical attractiveness. [Image: grid of six relationship-themed meme/post screenshots, including a "NOW HIRING a girlfriend -Full Time -please love me" sign post and an "Organized"/"Messy" comparison post] Figure 4. Example content from the "Relationships, Friends, & Family" topic associated with higher levels of appearance comparison. Topics that elicit a collective identity or activity are "good": At the bottom of Figure 2 are the topics associated with the lowest levels of appearance comparison after controlling for viewer demographics: people who see more of these topics are less likely to think that Instagram makes them feel bad about their bodies. These topics include Sports, Crime & Tragedy (which includes disaster relief and humanitarian aid), Religion & Spirituality, and Politics. What do they have in common? Most invoke a greater sense of community (e.g., sports team fandom, global humanity, religion, political ideology) rather than focusing on an individual. Other "good" topics in this set include Visual Arts, Architecture & Crafts, Games, Puzzles & Play, TV & Movies, Home & Garden, Animals & Pets -- all topics that focus on an interest or an activity rather than the body. Figure 5 shows examples of posts from these categories: CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057042 [Image: grid of nine photos -- Minecraft video game screenshot, macrame wall hanging, two soccer players celebrating, a hospital/medical scene with a Portuguese-language news chyron reading "Hospitais devem disponibilizar 70 novos leitos para tratar da Covid-19 em Maceió no mês de março," a silhouetted figure against a stormy sky, a bouquet of pink carnations, a modern living room interior, a woman holding a Portuguese-language Bible ("BÍBLIA") with an "SBB" logo, a close-up of a dog's face] Figure 5. Example posts from the topics associated with the lowest levels of appearance comparison on Instagram, including Sports, Crime & Tragedy, Religion & Spirituality, Visual Arts, Games. These topics de-emphasize the body and instead focus on a collective identity, shared purpose, or interest. Getting more specific: Subtopics and image features "Fashion & Style" is likely too broad a topic to be useful for product changes (e.g., demotions, especially since so much of Instagram is about fashion), and more innocuous-seeming topics like "Music & Audio" are unlikely to be justifiably demotable from a policy perspective. Here we take two approaches to identify more specific topics for product changes: We examine the sub-topics within FIT as well as some of the granular image features from Cortex/XRay most strongly associated with appearance comparison. Makeup and other beauty posts are problematic. List 1 shows the FIT subtopics within Fashion & Style that are most strongly associated with appearance comparison after controlling for viewer demographics in a regression. The majority are related to beauty. People who see a greater proportion of posts about makeup, skincare, hair, and hair removal feel like Instagram makes them feel worse about their appearances, even after accounting for demographics (teen girls follow more of these topics). Figure 6 includes example posts from these subtopics. Similarly (not shown), western pop is the most common subtopic within Music & Audio that elicits appearance comparison. People who see more content about western pop artists like Selena Gomez, Miley Cyrus, Ariana Grande, and Cardi B feel worse about their appearances. These musicians regularly focus on their bodies in their posts (all are included in Figure 2). However, we don't know specifically what makes content about these celebrities problematic. Do they match a eurocentric beauty ideal? Do they post a higher proportion of beauty content? Or is it simply because they have a large teen girl following (and thus still stand out in a regression, even after controlling for the age and gender of the viewership)? Though the list is mostly female celebrities, Justin Bieber and the wrestler John Cena are also associated with greater appearance comparison. Justin Bieber's content often includes his wife, a model, though his primarily female following could still explain his inclusion. Content about Cena (though not content that the wrestler posts himself) regularly features women's bodies. The only pop star with a FIT subtopic who was associated with *less* (better) appearance comparison was Michael Jackson; this may reflect his international fandom (not limited to western countries where appearance comparison is higher) and that teen girls are likely a smaller proportion of his fanbase. Content from these problematic subtopics comprise 26.3% of respondents' VPVs (36.5% for teen girls). Subtopics within Fashion & Style most associated with appearance comparison (after controlling for viewer demographics): - Makeup & Cosmetics - Lipstick - Makeup Artists - Beauty - Eye Liners - Jeans & Denim - Eye Makeup & Cosmetics - Skin Care - Eye Shadows - Outerwear - False Eyelashes & Adhesives CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057043 [List 1 continued:] - Hair Coloring - Fashion Modeling - Foundation (Beauty) - Costume Makeup & Makeup Art - Face Care - Makeup Brands - Shirts & Tops - Fashion Photography - Hair Styling Tools & Products - Face Makeup & Cosmetics - Hair Removal - Clothing & Accessories - Boots - Celebrity Style - Lip Makeup & Cosmetics - Hair - Sleepwear & Loungewear - Nail Art - Purses - Hair Extensions - Modeling & Photography - Piercings - Nails (Beauty) - Bags & Purses - Wearable Tech - Women's Shoes - Body Modification - Pants & Shorts List 1. Subtopics within Fashion & Style that are most strongly associated with problems with appearance comparison on Instagram. [Image: grid of nine beauty/skincare/makeup photos, including a video still reading "This is what pan-facial toxin looks like / Frown lines & Crow's feet" showing a cosmetic injection procedure, and another video still reading "...LASTS UP TO A YEAR" showing a cosmetic injection near the mouth] Figure 6. Example posts from the Makeup & Cosmetics, Skin Care, Eye Liner, and other beauty subtopics. People who see more content like this feel higher levels of appearance comparison (after accounting for viewer demographics). Furthermore, List 2 [shows] some of the image features (XRay/Cortex concepts) that are most useful for distinguishing which fashion posts are likely to be problematic. To identify these features, we trained an FBLearner model on photos from the Fashion & Style FIT topic to predict whether viewers of these photos would have an appearance comparison score >= 4 (indicating high levels of appearance comparison) or not. The model isn't very good (test AUC = 0.54 on a balanced dataset), because our labels are at the person-level (survey-taker) rather than the content-level CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057044 (the image itself). However, it does identify features consistent with qualitative research, in that many are about the body (particularly women's bodies), such as cleavage. Many are related to the "sexually suggestive" part of our non-recommendable / objectionable content guidelines, so we could apply these guidelines in more places or more aggressively with teens. That said, we need to be careful not to over-enforce these guidelines or shadow ban underprivileged groups; false positives like "breastfeeding" reinforce the need to exercise caution here. Image features within Fashion posts most strongly associated with appearance comparison: - Cleavage - Sexual eating - Smiling - Water - Face - Blond - Hat - Jewelry - Breastfeeding - CEI reportable - Black and white - Leather - CEI_2 - Breastfeeding - Fineart - Women's swimwear - Shoes - CEI - Non-sexual nudity - Making out - Bent over - Crotch grab - Suggestive - Weight loss - Sexual nudity - Nipple through clothing - Bedroom - Phallic objects - Eyewear - Hair List 2. Top image concepts (XRay/Cortex) predicting which Fashion-topic photos are most likely to elicit appearance comparison. Most relate to the body or sexuality. "CEI" stands for "child exploitation imagery." Who posted it Here, we break down the topics based on who posted them. In summary: Content from top accounts and non-friends about fashion/beauty/romance is likely problematic, while content from friends about these topics is not. Top accounts: Top accounts (celebrities) are defined as those with the top 0.1% of followers globally or by country. More than half (54%) of the content that people see comes from top accounts (median of 180 pieces of content across all IG surfaces per day among survey respondents; 223 for teen girls). As Figure 7 shows, appearance comparison is connected to the content that people see from top accounts, rather than non-top accounts (red bars are statistically significant, while blue bars are not). This suggests that product interventions we consider (e.g., nudges or pivots in Feed) should aim to reduce exposure to top-account content related to these topics, rather than non-top-account content. CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057045 [Figure 7: horizontal bar chart, "Appearance comparison on IG is more strongly associated with what people see from top accounts." Same topic list/order as Figure 1, paired red ("Top account") / blue ("Non-top account") bars per topic.] Figure 7. Appearance comparison on Instagram is associated with the content that people see from top accounts (red bars) rather than non-top accounts (blue bars). Darker bars are significant at p < 0.01. Friends: "Friends" are defined based on three criteria from the Instagram Home team (learn more). - The target is a normal (non-interest) account - Both accounts follow each other - They have at least three mutual follows As Figure 8 shows, the problems associated with content about relationships (e.g., dating, coupledom, sex) and fashion (including beauty) are coming from non-friends. Seeing more friends post about these topics is not associated with worse levels of appearance comparison on Instagram. The role of friend content is complex. In some cases, friend content may be "protective," associated with lower levels of appearance comparison. For example, consider the case of "Fitness & Workout" content in Figure 8. Seeing fitness content from friends is associated with less appearance comparison (people don't think Instagram makes them feel worse about their bodies). Seeing fitness content from non-friends does not have that potential benefit. Similarly, seeing friends post about celebrations, children, and the home is associated with less appearance comparison on Instagram, while non-friend content doesn't have these effects. Why might friend content on these topics be good and non-friend content on them be bad? One reason is that people know the struggles and context behind their close friends' lives, but they don't have the same information about non-friends (an idea supported by interviews with teens). Thus, beautiful content about a happy relationship or successful weight loss journey may feel great when it comes from a friend whom the viewer knows has struggled in the past, but feel awful when it comes from a non-friend (especially top accounts, as shown above). However, friends also make more relevant comparison targets. People's middle-tier connections (not their close friends and not celebrities) are the most common sources of comparison, because they're close enough to have impact (and seem like "normal people" and thus reasonable targets for comparison), but far away enough that viewers don't know the backstory. In fact, people who see more friend content overall experience more appearance comparison (ANP). Only 13% of the content that people see comes from friends (15% for teen girls), a median of 31 pieces of friend content per day (46 pieces of friend content for teen girls). Thus, for every piece of friend content a teen girl sees, she sees 5x as many pieces of content from top accounts. We should make it easier for teen girls to find and consume content from their friends, especially close friends. Topics that elicit a sense of greater community are good from non-friends, including posts about sports, crime & tragedy (which includes disaster relief and humanitarian aid), and religion. (Above, we showed that these topics are "good" from top accounts, too.) CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057046 [Figure 8: horizontal bar chart, "Appearance comparison on IG and views of content from friends/non-friends." Same topic list as Figure 1, paired blue ("Friends") / red ("Non-Friends") bars per topic.] Figure 8. Seeing content from friends is generally good, while seeing fashion and relationship (dating) content from non-friends is associated with feeling worse about one's appearance. Content from non-friends about a collective identity (sports, tragedy, religion, politics) or interests (games, arts, animals) are generally good, too. Darker bars are significant at p < 0.01. Connected vs. unconnected accounts: Unconnected content is likely problematic. About three-quarters (73%) of the content that people see is produced by accounts they've chosen to follow (a.k.a. "Connected accounts"). Pooled across all topics, seeing more unconnected content is associated with worse appearance comparison. Women who spend proportionally more time on Explore (where we promote unconnected content) also have higher levels of appearance comparison (ANP). In a recent listening session, one creator described Explore as "a landmine for everything I want to avoid on IG" (@kendramorous) because it triggers appearance comparison. In this analysis, no specific topic stands out as being problematic when posted by an unconnected account (Figure 9); this may be due to low statistical power per topic on unconnected accounts. As Figure 8 shows, content from unconnected accounts (those that people didn't choose to follow, red bars) all skew to the right. Though no individual topic is statistically significantly "bad," cumulatively, they trend that way. [Figure 9: horizontal bar chart, "Appearance comparison on IG and views of content from connected/non-connected accounts." Same topic list as Figure 1, paired blue ("Following") / red ("Not Following") bars per topic.] Figure 9. Appearance comparison is more strongly associated with content that people are following ("connected to"), though this may be due to low statistical power per-topic for unconnected content. Nearly all red bars (unconnected content) trend to the right, suggesting that seeing unconnected content may still be problematic. Darker bars are significant at p < 0.01. Summary and Recommendations People who see greater proportions of content about fashion and beauty (e.g., makeup, skincare, hair removal) or more content from globally western pop stars emphasizing their bodies say that Instagram makes them feel worse about themselves: they feel more pressure to be perfect, they feel worse about their own bodies, and they feel like they compare themselves to others on Instagram more often. Collectively, we refer to this concept as appearance comparison, which is a form of negative social comparison. On the other hand, when people see greater proportions of content that emphasizes a collective identity or broader humanity (e.g., sports, tragedy, religion, politics), or an activity (crafts, gardening, puzzles, TV) rather than the poster's body, they feel lower levels of appearance comparison. The topics that elicit appearance comparison comprise 1/4 of the content people see on Instagram, and 1/3 for teen girls. This is a high amount and not all of it is problematic, but even reducing the prevalence of these topics down to 1/5 or 1/6 would go a long way toward addressing appearance comparison on Instagram, especially if we boost content about collective identities or activities instead. These findings take into account viewer demographics: teen girls who see proportionally more makeup content experience more appearance comparison on Instagram than do teen girls who see less makeup content. There are still selection effects that we can't account for (e.g., self- CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057047 esteem, which accounts they choose to follow), so additional research is needed to draw firmer causal conclusions (see below). It's worse when the content comes from celebrities (top accounts) and non-friends, rather than from friends, and it's likely worse from unconnected accounts (those that the viewers didn't choose to follow). We should reduce exposure to celebrity content about fashion, beauty, and relationships, while increasing exposure to close friend content of all kinds. Potential product directions include: - Demotions on Explore and Reels using topic and image/video features identified in this note - Not recommending celebrities to follow that post primarily fashion/beauty content. People can find these accounts on their own, but we shouldn't amplify their influence through recommendations. - Boosting content that elicits a collective identity (e.g., sports, tragedy) or activity (e.g., crafts, puzzles) - Making it easier for teen girls to find and consume content from their friends, especially close friends. - Nudging or pivoting people toward other content or other activities after they consume or seek out large volumes of fashion/beauty/relationship content. - Applying "sexually suggestive" non-recommendable guidelines more aggressively for teen viewers. - Providing better content controls to allow people to hide sexually suggestive photos and those that emphasize women's bodies or faces. Areas for more research - What is it about Western pop star content? Are there pop stars that are "good" or subsets of pop star content that reduce appearance comparison (e.g., photos of people enjoying a concert)? Or is this finding just reflecting a selection effect of teen girls following female musicians who emphasize their bodies? - Makeup and beauty content is associated with worse appearance comparison, yet many teen girls choose to see this content. What product interventions would be most effective for them? Would they use content controls to hide content that's problematic for them, or would demotion be more effective? - Are there subsets of makeup and beauty content that are inspirational (e.g., when creators show their journeys and struggles, or at least not harmful)? Are there subsets of beauty content that are more inspirational than harmful (e.g., when creators show their journeys and struggles)? One rudimentary option might be to investigate makeup content with and without people in them (e.g., just showing the product); we should be able to use a combination of FIT topics and XRay features to identify those images. - What are the circumstances in which friend content is beneficial, and when is it harmful for appearance comparison? The present research did not identify specific topics that may be problematic when posted by friends, but friends are more relevant comparison targets than celebrities, and teen girls report feeling worse when friends mimic celebrity beauty standards. When would promoting friend content help and when could it hurt? - The present study was at the person level (survey respondent). We need data at the content level to reduce bias from selection effects. Ideally, we'd have viewers (especially teen girls) answer a survey about pieces of content they view, letting us know how that content makes them feel about themselves. This would allow us to train content-level models and identify more accurate features (including combinations of XRay concepts, embeddings, caption keywords, and FIT subtopics) than are possible with the present person-level data. You can explore images by FIT topic and subtopic that were associated with higher and lower levels of appearance comparison among survey respondents in the FIT Topic Explorer dashboard. [14 reactions, 3 Comments, 4 Shares] CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057048 This is wonderful and important work! I am curious about the recommendation for "reduce exposure to celebrity content" - isn't that what IG mostly about? getting a peak to the (very photogenic) life of the top 0.1%? isn't this the reason why teens are on the platform? And I'd guess it goes beyond fashion celebrities - the people I follow are the top chefs, runners, photographers and cyclists in the world. It's fun and inspiring (because they are soooo amazing) and a bit depressing at the same time. I'd also guess that not all people are impacted the same (plug for Drebbel...), so identifying susceptible populations might be a better solution than reducing all celebrity content to everyone. Furthermore, I'm guessing (cc Shankar) that there's just not enough friend inventory and even if we demote top-accounts content, it will still be very prominent. Like Reply 1y : Great points! Definitely agree on all fronts (including lack of friend inventory) and I doubt we'd do any kind of uniform cut across all celebrity content, but rather something subtler like demoting body-focused beauty celebrity content on surfaces like Explore. (And as you know we're using Drebbel to measure whether our algorithms amplify that kind of content now.) Supporting vulnerable populations (e.g., teen girls) with different offerings is more challenging on IG than FB since we don't have demographics and inferences aren't great in most countries. We've talked about (1) offering people controls so that people who feel more susceptible to comparison can hide that kind of content (though more research is needed to see whether teens would even opt in), (2) nudging people away after excessive exposure and (3) separating top-account feed from close-friend feed so that regular people don't have to "compete" side-by-side with celebrities. Like Reply 1y [3 reactions] : Maybe the best way to reduce exposure to celebrity content is to increase friend inventory. It's hard, of course, but would probably do more to mitigate these problems than trying to target specific comparison-inducing content for demotion. Like Reply 51w Recommended Reading "Be Different" -- Naomi Gleit "Preview of Return-To-Office room booking experience..." -- "All Systems Go for RTO: VC Feature Friendly Guide" -- CONFIDENTIAL - PROTECTED BY CONFIDENTIALITY AGREEMENT MT-IG-AG-00057049