They’re not exact. You can’t go “Waluigi was ranked number 1! He’s the most highly requested character!” Because obviously it has many flaws and holes and whatnot which keep it from being accurate. But if the character is in the top 15, it’s pretty much safe to say that they’re highly requested (unless there is an actual flaw, which would be obvious).
The problem is Voluntary Response Bias is at its worst with Smash fan polls. You're starting population is only ever going to be the most hardcore dedicated fans and their immediate circles, and of those groups, it has to be the people who are most vocally supporting characters
and willing to spend whatever amount of time needed to respond to the poll. The people engaging with forums such as SmashBoards or even on something like Reddit or ResetERA's more specific Smash communities are already one massive degree of voluntary response bias (and often a certain degree of additional bias that comes from being the types of people who have excessive time to talk Smash of all things) and then we introduce another by specifically creating a voluntary response bias in a brand new poll, and then furthermore the most vocal of communities are the ones who generally get placed the highest since they actively work to rally troops for such polls. How many times do you see a support thread link a poll on an unrelated site to try and encourage the visibility of their character? Yes, organizing in any sort of "Democratic" context is objectively a good thing, but that's not traditionally what polls are meant to be as a statistical test when there being treated fundamentally as a true ballot.
What you end up doing is testing extremely specific sectors of the Smash fandom over and over with most of those polls is the ultimate result, and it more clues you in from a data perspective to which people you're polling at a certain point as opposed to any sort of true representations of the Smash "most requested fighters." The "polls" Smash communities refer to are most frequently the absolute worst polls from a data collection perspective because there is basically no control over the parameters of said polls. Many are fundamentally open ended affairs in terms of the responses you can give (a number would allow me to literally input a response of literally thousands of different options, which is rarely what you want with any statistical test/study/data collection method) and rarely make parameters for who can participate, and they're also subject to minor forms of tampering and/or simply being promoted in more specifically vocal or online communities.
Smash fan polls aren't useless, but they're not particularly useful items either.