"1/50 pieces of content was 'age appropriate nutritious' or the sort of content we would like to promote to teens. This should also suggest that we have a big hill to climb on ensuring a higher % of a teen content consumption diet is 'nutritious'."

████: "IG and CI teams recently collaborated to outline the policy violating harm IG teens are exposed to. These results are summarized in the doc... We appreciate your feedback and inputs on this and would like to publish this as a note in closed IG/CI groups next week." "Summary 1. Generally, the countries which have enough samples to measure Teen prevalence rates have significantly higher SWOPSWOPSeverity Weighted Overall PrevalenceSWOP = Prevalence * Severity. A severity-weighted metric for the prevalence of harmful content/experiences.Full entry ↗ in the teen segment than the general population. 2. Age-based prevalence measurement on Instagram has major recall gaps, but Teens in countries with measurement are seeing significantly higher integrity harms compared to the general population. We need to upsample this User segment that is a company focus and vulnerable population. 3. Teens see 2-4x SWOP in every country we have enough samples to measure except India. SWOP heavily weights Bullying and Harassment, which appears to be a relatively uniform and prevalent harm amongst Teens. Teens are also subject to higher rates of numerous other harms, including SSI and VNI. These harms tend to manifest on Hashtags and Comments." "Recommendations 1. We need to Increase sampling for Teens in key markets once support for easier upsampling is possible. 2. Age classification is currently a limiter, and we should push strategies to improve recall of Teens. This could mean accelerating strategies to request age from existing accounts and/or revisiting our protocol to not share age with family matching. 3. Consider adopting strategies to reduce integrity Harm prevalence in the Teens user segment. SWOP reductions goals or counter-metrics should be considered in context to growth goals and experimentation. This could include showing age-appropriate content and/or adjusting ranking and recommendations based on age. 4. We should consider what levers could be applied or developed to reduce the harm experienced by users under 18, particularly for the areas of Bullying and Harassment + SSI. Given the connection of these harms to Hashtags and Comments, it aligns with other topline observations that those surfaces have higher rates of harm. 5. Push for improved IMI calculation methodology and tooling to make Teens segmentation available to end users in dashboard formats." ████: "Additional TLDR thoughts: 1/ SWOP prevelance for teens is ~0.42% of the IG ecosystem. This might seem like a small number but its still ~2x worse for teens than non-teens (0.24% vs 0.42%). It also appears to be especially bad for hashtags and comments (suggesting product opportunities around discoverability/search and proactive close friend/friend interactions). 2/ We also ran a very preliminary labeling exercise where we found that ~30% of labeled content appeared to be bad or borderline 'icky' things we would not want kids to be exposed to. We only looked at 50 pieces of content and did not use policy guidelines to do the labeling; post forthcoming from Aashna but this should give us some level of intuition that the problem is much larger than the 0.42% of things we are currently capable of measuring and actioning against with current policy. 3. 1/50 pieces of content was 'age appropriate nutritious' or the sort of content we would like to promote to teens. This should also suggest that we have a big hill to climb on ensuring a higher % of a teen content consumption diet is 'nutritious'. 4. So... we have a big challenge both with closing gaps against conventional integrety defences but also with developing new policy, measurement, detection, parental controls, and enforcement (in the form of both removal and soft actions)."