There are a lot of ways to handle this issue, but from my experience they tend to fall into three schools of thought. My preferred one is actually the last one I will present.
One school of thought says, "Hey, this is a role playing game, we like player involvement and engagement, so let's reward creativity!"
The upshot of this approach is that if you want to generate creativity and engagement from your players, then you reward that. How do you reward that? By letting good, creative, or fun ideas succeed, which in this case translates to how you've been running things: If Harald Halitosis's player comes up with a good and compelling line of persuasive patter to the young princess away from the royal guard, well that's creative and engaged, so go ahead, let that work.
The good part of this is that it will tend to generate player activity. The bad part of this is that it leads to those stats and skills being used as dump stats and isn't very realistic-- it's very easy for this to generate a group min-maxed characters, where the 'min' doesn't have much effect.
Emphasize Accurate Role-Playing
Another school of thought says, "Hey, this is a Role-Playing Game, shouldn't we be actually playing these roles?"
The key idea here is that in general, even if the player of Ulgo the Unlikable thinks fast on his feet, and is quite the charmer in real life, the character whose role he is playing is not. So the player shouldn't be playing him that way. He should be playing him as... whatever he decides is the way low charisma and untrained social skills manifest for his character.
Notice I said 'emphasize' rather than 'reward' here. My experience is that this is a situation where excellence is its own reward, and it's a sort of a table-culture that needs to be nourished and encouraged rather than directly rewarded.
The good part here, if you can achieve it, is that you minimize the tendency to treat these as dump stats, and when you do, you have players that will at least act like their characters are lacking something. The bad part is that I find it difficult to figure out appropriate and meaningful rewards for the behavior I want to encourage.
Enforce the Skills/Stats as a Mechanical Interface
Finally, a third approach is to emphasize and enforce the skills and stats mechanically. This approach says to itself, "Look, this is an RPG with mechanics. I don't let Ned the Noodle-armed perform feats of strength just because his character is a body builder in real life. So I'm not going to let Ted the Tongue-Tied act as the party barrister just because his player is on the debate team."
A crude way (and not one I recommend) is just to make a snap judgment: "Your character is too low Charisma to do that." Whether it comes from the GM directly or as the result of a die roll, that's just a slap in the face.
A subtler way, though, is to try to treat social skills and stats the same way as the physical ones: As an interface that informs the results of the players' desired actions. What this means is that if a player decides he wants his character to move a heavy object, then the player's strength is not relevant. His character's strength is relevant and decide what actually happens, whether the character moves the object or just strains and grunts ineffectively.
Likewise for social stats: The character really might have a compelling line of argument, but if his social skills are too low he might not be able to get the NPC to even listen. Or might not be able to articulate that idea as well as the player could. Or it might be a case of, "That sounded better in your head than it did in your mouth."
This is where the die rolls come in, and where they intersect with GM judgment. Even someone perpetually tongue-tied gets to have a good day when the dice are with him. And if the player really does come up with a good idea, you as GM can reward that with a bonus to the die roll.
I find this very rewarding when I can make it work, because it tends to, again, reduce the degree of minmaxing, and puts the bite back in those minimizations.
But, in fairness, this is NOT an easy technique to master, and it is very much best done up front and transparently. Meaning, it is best to tell your players what it is you're doing. Otherwise you can end up confusing them, irritating them, looking inconsistent, looking like you're playing favorites, etc.
There Is No Absolutely Correct Method
The best method for you is the one that works best for you and your table. None of those schools of thought are objectively correct, or necessarily better than the others. They are different solutions to a problem that yield different results. It's on you and your players to work out which seems like the most fun to you, or to work out yet another approach if none of those appeal.
Don't let anyone tell you differently.