Which AI design principles for social media have research support?

The paper closes with four principles, "aimed at social media platforms, policymakers, and stakeholders," meant to "guide an ethical and effective integration of generative AI into social media." They are: "Appeal of optional use and transparent disclosure"; user-focused personalization; context-sensitivity "to account for both topic and user intent"; and "prioritizing intuitive user interfaces." They are the prescriptive answer to the duality reported in Do AI writing tools improve online discussion or degrade it?. The excerpt does not say which of the costs each principle is meant to address, so the link between principles and findings is the authors' own.

Only two of the four rest on anything the excerpt reports. Optional use has the most direct support. Chat users submitted "an average of 3.36 to 3.55 prompts per round," about 13–16% of their final comments overlapped with the AI's text, and in over 89% of cases where participants used Feedback while composing, they submitted the comment. The conclusion reports that Chat and Suggestions users found those tools "more supportive overall," easier to use, more intuitive and better for their contributions than Feedback and Conversation Starter users. That grounds the interface principle, though the supporting figures sit in supplementary material the excerpt does not include. Disclosure has no test here. The experiment itself withheld which model was used and whether content was AI-generated, so the disclosure principle is an argument, not a result. Personalization and context-sensitivity are conditioned on tools being "developed with greater personalization, contextual awareness, and stylistic nuances," and the excerpt has no test of either. The one related analysis, the demographic check, found effects that did not reach significance.

The personalization principle assumes personalization helps. Does chatbot personalization build trust or expose privacy risks? reports that personalization also raises privacy concern and that each interaction ratchets expectations, a cost the excerpt never mentions. The interface principle favors "simplicity and familiarity." Do generated interfaces outperform text-based chat for most tasks? reports a preference for generated, task-specific interfaces over chat in other settings. The two concern different tasks, and the excerpt does not test generated interfaces, so they do not contradict each other. A platform that followed the familiarity principle would still be choosing the conversational form that neighbor reports losing to generated interfaces in its setting.

These are the authors' recommendations, not the results of any tested intervention. The excerpt reports no outcome for disclosure, personalization or context-sensitivity, and no evidence of any platform or policymaker adopting them. At the strength the evidence allows, optional use and interface simplicity are the two principles with reported usage or questionnaire support. The other two are design hypotheses, and a platform that adopts them would need its own measurements to know whether they hold.