As someone who's done polling/sentiment analysis work, while I have to say polling helps to give a more quantifiable data source and can be a useful source of feedback when structured/phrased appropriately, it should not be the only feedback path. Poll data always has to be considered alongside other sources of data, including in-game data and qualitative data from forums and interviews.

Poll data comes with its own issues, including sampling bias (e.g. online poll response rates also tend to be fairly low and riddled with participation bias, although this can be mitigated with targeted reminders; you have to decide if/how are you going to reach out to those who have unsubscribed temporarily from the game, etc). Poll data is relatively easily analysed and quantified, but in the end it is still a judgement call on what results are "good enough". Is 51% of the job player base (and here we run into the "to filter or not to filter" issue) being satisfied with the job good enough? What is the "KPI" if any that the dev team should aim for? 70%? 80%? Because it is quantifiable, you potentially lose out on the nuances and weak signals (not majority view but minority views that can help identify what are or may become big issues) brought by more qualitative data like forum posts.

At the end of the day, they have to look at all of these different sources of data, decide what to weight more heavily in their considerations, and make a judgement call on job direction looking forward. I think the gap here is that they may not fully understand the context in which certain jobs are played (e.g. healer complexity and playstyle of OGCD healing) to make a convincing case to players, or are otherwise not able to fully justify the reasons for such changes (as a NIN main, I was disappointed with their explanation for changing the TA buff during the 6.1 patch notes reading).