This is incidentally an endemic problem of online discussion sites.
There is only one possible quality flow gradient, and that is downwards.
If a site begins well, with high-quality and relevant content, then those who wish to exploitatively extract value from that factor will be attracted to it. Eventually the clue up and leaves.
If a site begins poorly, with low-quality and irrelevant content, and quite often, abuse, disrespect, fraud, crime, and disinformation to boot ... the clue leaves early and the site rapidly becomes a cesspit.
There's a third option, of course, though one still consistent with the quality flow gradient: holding a steady state. Site quality doesn't improve, but it doesn't markedly deteriorate either. I'd put a small handful of online sites in that category, HN, LWN, and Metafilter top my own list, though I suspect there are others. What's key is that there's a sufficiently small community that norms enforcement is significantly socialised, there's effective and diligent moderation, and crucially (and possibly not the case with my examples) there's fresh blood introduced over time consistent with quality standards. Absent this last, such fora can continue for a time, even over many decades, but eventually stale, often becoming incestuous, and ultimately dying out.
Among real-world institutions which seem to manage to find similar stable points, I'd include most especially academic institutions, which balance a high flux of students with a far more stable faculty and staff cohort. Selective-admissions schools have retained high rank for many decades or centuries, in some cases millennia. Cities, larger political units (states and/or empires), some businesses (including especially professional services firms) and professional organisations (e.g., not-for-profits rather than businesses) may also succeed, at least over the decades-to-centuries span. (Charles Perrow includes a discussion of several noncommercial / nongovernmental organisations with significant changes, we'd now call them "pivots", over the 20th century, in Complex Organizations (1972, 1984).)
The media-quality-gradient is largely a result of scaling laws, the fact that elite cohorts (high degrees of expertise, sociability, and intelligence) tend to be small, and that once an interaction grows beyond the size of such a cohort it will incorporate participants less able, willing, and/or interested in maintaining original standards. I've posted occasionally on large-scale detailed studies of literacy (in the US) and computer skills (OECD) which show that at a population level only about 15% of the population has high literacy, numeracy, and/or computer skills, and that as much as half operate at poor or "worse than poor" levels. As I've discussed previously, this is both discouraging to those who consider themselves among the higher levels, and of significant concern in constructing systems which must and can be used by large portions of society, including those with low intrinsic capabilities (very young, old, sick, injured/traumatised, and the intrinsically less able). Ideally I'd prefer to see elite support where appropriate, but common accessibility where at all possible.