X already distinguishes between viewers who allow sensitive media and viewers who do not, while its recommendation and visibility systems also classify adult content and adult-labeled authors.
However, the current architecture appears to use adult-content classification primarily as a restriction signal rather than also using it as a positive relevance signal for eligible viewers.
I propose introducing a preference-aware adult discovery layer where correctly labeled adult content can remain restricted from users who should not or do not want to see it, while becoming eligible for recommendation to adult viewers who have explicitly enabled sensitive media and whose recommendation signals indicate an interest in that type of content.
The goal is not to weaken adult-content safety controls.
The goal is to improve matching between:
adult creators who correctly classify their content
and
adult viewers who have explicitly chosen to see sensitive/adult content.
Current behavior
The public visibility-filtering implementation already contains viewer-level context such as whether the viewer allows sensitive media.
At the same time, several NSFW post and author labels can result in:
• interstitial warnings,
• removal from recommendations to non-followers,
• or both.
This creates an asymmetry.
X knows:
- that the content is adult,
- that a viewer may have chosen to allow sensitive media,
- and that recommendations are personalized at the viewer level,
but adult classification can still function primarily as a negative eligibility signal rather than as a category that can participate in personalized recommendation among eligible viewers.
The product problem
Adult content is not inherently relevant to every user.
But it is also not inherently irrelevant to every user.
If an adult viewer explicitly enables sensitive media, repeatedly chooses to view adult posts, follows adult creators, and positively engages with that content, those signals indicate a clear preference.
For that viewer, permanently excluding correctly labeled adult posts from out-of-network discovery reduces recommendation quality.
The system is effectively saying:
“We know you are eligible to see this content, we know you have enabled it, and we may know you regularly engage with it — but we still will not use that information for discovery.”
That is not optimal personalization.
Proposed model
Treat adult-content classification as both:
- A safety constraint
For:
• minors,
• users without sufficient age information,
• logged-out users where appropriate,
• viewers who disable sensitive media,
• or users whose settings otherwise make the content ineligible,
continue applying existing restrictions.
- A relevance feature
For an authenticated adult viewer who:
• has explicitly enabled sensitive media,
• is eligible under applicable policy and jurisdiction,
• and demonstrates recommendation signals consistent with interest in adult content,
allow correctly labeled adult posts to participate in a dedicated or constrained out-of-network recommendation pool.
The content would remain labeled as adult.
The classification would not disappear.
Instead, the classification would help determine who is an appropriate viewer.
Conceptual behavior
Instead of treating:
post_is_adult -> OON Drop
as the universal outcome, use viewer-aware logic conceptually similar to:
post_is_adult && viewer_not_eligible -> Drop
post_is_adult && viewer_eligible_but_sensitive_disabled -> Drop or Interstitial
post_is_adult && viewer_is_adult && viewer_allows_sensitive_media && viewer_has_adult_interest -> eligible for adult-safe ranking
The exact implementation could remain substantially more conservative than normal recommendations.
Recommendation safeguards
Adult OON discovery could require:
• confirmed or sufficiently reliable adult viewer status,
• sensitive-media preference enabled,
• jurisdictional eligibility,
• no relevant safety/enforcement restrictions,
• high-confidence adult-content labeling,
• exclusion of illegal or prohibited sexual content,
• and existing negative-feedback protections.
Ranking could additionally use viewer-specific signals such as:
• follows of adult-labeled creators,
• voluntary opens of sensitive-media interstitials,
• dwell on adult content,
• likes,
• reposts,
• bookmarks,
• follows,
• blocks,
• mutes,
• reports,
• and Not Interested.
These signals should be used carefully and with appropriate privacy safeguards.
Why this improves safety
This proposal does not require exposing adult content more broadly.
It does the opposite: it creates a more explicit separation between audiences.
Users who do not want adult content remain protected.
Users who are not eligible remain protected.
Meanwhile, adult viewers who explicitly choose to see sensitive media can receive more relevant recommendations from properly labeled creators.
This creates a stronger incentive for adult creators to classify their content accurately, because correct classification becomes useful for matching rather than functioning exclusively as a distribution penalty.
Profile experience
The same preference-aware principle could improve profile rendering.
When an eligible adult viewer has explicitly enabled sensitive media and intentionally visits an adult-labeled profile, X could consider reducing repetitive interstitial friction for correctly labeled media while retaining clear account-level context and user controls.
A profile consisting almost entirely of individually blurred posts can create unnecessary friction for a viewer who has already explicitly opted into sensitive media.
This should remain fully controlled by viewer preferences.
Suggested experiment
Control
Current visibility and recommendation rules for adult-labeled posts.
Treatment
Correctly labeled adult posts become eligible for a constrained adult-content OON candidate pool when the viewer:
• is eligible,
• allows sensitive media,
• and demonstrates sufficient adult-content affinity.
Measure:
• reports,
• blocks,
• mutes,
• Not Interested,
• interstitial open rate,
• dwell,
• likes,
• reposts,
• bookmarks,
• follows,
• session satisfaction,
• content-policy violations,
• and accidental exposure rate.
The key safety requirement should be:
no meaningful increase in unwanted adult-content exposure among users who have not opted in.
Benefits
For viewers
More control and more relevant recommendations.
Users who do not want adult content remain protected, while consenting adult viewers receive content better aligned with their stated preferences.
For adult creators
A legitimate discovery mechanism based on audience suitability rather than blanket suppression.
For X
Better personalization, stronger incentives for accurate labeling, improved segmentation of sensitive content, and potentially less incentive for creators to evade classification.
Principle
Adult-content classification should determine the appropriate audience, not automatically eliminate the possibility of an audience.
Correctly labeled adult content should remain protected from inappropriate viewers.
But where the viewer is an eligible adult who has explicitly chosen to see sensitive content, the recommendation system should be able to use that preference to make better matches.
In short:
consent-based discovery instead of blanket suppression.
X already distinguishes between viewers who allow sensitive media and viewers who do not, while its recommendation and visibility systems also classify adult content and adult-labeled authors.
However, the current architecture appears to use adult-content classification primarily as a restriction signal rather than also using it as a positive relevance signal for eligible viewers.
I propose introducing a preference-aware adult discovery layer where correctly labeled adult content can remain restricted from users who should not or do not want to see it, while becoming eligible for recommendation to adult viewers who have explicitly enabled sensitive media and whose recommendation signals indicate an interest in that type of content.
The goal is not to weaken adult-content safety controls.
The goal is to improve matching between:
adult creators who correctly classify their content
and
adult viewers who have explicitly chosen to see sensitive/adult content.
Current behavior
The public visibility-filtering implementation already contains viewer-level context such as whether the viewer allows sensitive media.
At the same time, several NSFW post and author labels can result in:
• interstitial warnings,
• removal from recommendations to non-followers,
• or both.
This creates an asymmetry.
X knows:
but adult classification can still function primarily as a negative eligibility signal rather than as a category that can participate in personalized recommendation among eligible viewers.
The product problem
Adult content is not inherently relevant to every user.
But it is also not inherently irrelevant to every user.
If an adult viewer explicitly enables sensitive media, repeatedly chooses to view adult posts, follows adult creators, and positively engages with that content, those signals indicate a clear preference.
For that viewer, permanently excluding correctly labeled adult posts from out-of-network discovery reduces recommendation quality.
The system is effectively saying:
That is not optimal personalization.
Proposed model
Treat adult-content classification as both:
For:
• minors,
• users without sufficient age information,
• logged-out users where appropriate,
• viewers who disable sensitive media,
• or users whose settings otherwise make the content ineligible,
continue applying existing restrictions.
For an authenticated adult viewer who:
• has explicitly enabled sensitive media,
• is eligible under applicable policy and jurisdiction,
• and demonstrates recommendation signals consistent with interest in adult content,
allow correctly labeled adult posts to participate in a dedicated or constrained out-of-network recommendation pool.
The content would remain labeled as adult.
The classification would not disappear.
Instead, the classification would help determine who is an appropriate viewer.
Conceptual behavior
Instead of treating:
post_is_adult -> OON Drop
as the universal outcome, use viewer-aware logic conceptually similar to:
post_is_adult && viewer_not_eligible -> Drop
post_is_adult && viewer_eligible_but_sensitive_disabled -> Drop or Interstitial
post_is_adult && viewer_is_adult && viewer_allows_sensitive_media && viewer_has_adult_interest -> eligible for adult-safe ranking
The exact implementation could remain substantially more conservative than normal recommendations.
Recommendation safeguards
Adult OON discovery could require:
• confirmed or sufficiently reliable adult viewer status,
• sensitive-media preference enabled,
• jurisdictional eligibility,
• no relevant safety/enforcement restrictions,
• high-confidence adult-content labeling,
• exclusion of illegal or prohibited sexual content,
• and existing negative-feedback protections.
Ranking could additionally use viewer-specific signals such as:
• follows of adult-labeled creators,
• voluntary opens of sensitive-media interstitials,
• dwell on adult content,
• likes,
• reposts,
• bookmarks,
• follows,
• blocks,
• mutes,
• reports,
• and Not Interested.
These signals should be used carefully and with appropriate privacy safeguards.
Why this improves safety
This proposal does not require exposing adult content more broadly.
It does the opposite: it creates a more explicit separation between audiences.
Users who do not want adult content remain protected.
Users who are not eligible remain protected.
Meanwhile, adult viewers who explicitly choose to see sensitive media can receive more relevant recommendations from properly labeled creators.
This creates a stronger incentive for adult creators to classify their content accurately, because correct classification becomes useful for matching rather than functioning exclusively as a distribution penalty.
Profile experience
The same preference-aware principle could improve profile rendering.
When an eligible adult viewer has explicitly enabled sensitive media and intentionally visits an adult-labeled profile, X could consider reducing repetitive interstitial friction for correctly labeled media while retaining clear account-level context and user controls.
A profile consisting almost entirely of individually blurred posts can create unnecessary friction for a viewer who has already explicitly opted into sensitive media.
This should remain fully controlled by viewer preferences.
Suggested experiment
Control
Current visibility and recommendation rules for adult-labeled posts.
Treatment
Correctly labeled adult posts become eligible for a constrained adult-content OON candidate pool when the viewer:
• is eligible,
• allows sensitive media,
• and demonstrates sufficient adult-content affinity.
Measure:
• reports,
• blocks,
• mutes,
• Not Interested,
• interstitial open rate,
• dwell,
• likes,
• reposts,
• bookmarks,
• follows,
• session satisfaction,
• content-policy violations,
• and accidental exposure rate.
The key safety requirement should be:
no meaningful increase in unwanted adult-content exposure among users who have not opted in.
Benefits
For viewers
More control and more relevant recommendations.
Users who do not want adult content remain protected, while consenting adult viewers receive content better aligned with their stated preferences.
For adult creators
A legitimate discovery mechanism based on audience suitability rather than blanket suppression.
For X
Better personalization, stronger incentives for accurate labeling, improved segmentation of sensitive content, and potentially less incentive for creators to evade classification.
Principle
Adult-content classification should determine the appropriate audience, not automatically eliminate the possibility of an audience.
Correctly labeled adult content should remain protected from inappropriate viewers.
But where the viewer is an eligible adult who has explicitly chosen to see sensitive content, the recommendation system should be able to use that preference to make better matches.
In short:
consent-based discovery instead of blanket suppression.