Option 2 (the proposed system):
The biggest conceptual change is that the tagged lines would now represent an attempt without any guarantees of success. It’s up to the player to read people based on the available info and consider what would work best. You can have two different streetwise lines, for example, one would result in a positive reaction, the other in a negative.
That brings us to the second biggest change. Most lines would no longer lead to success or failures but result in positive and negative reactions, represented numerically. Your skill level would act as modifiers, magnifying positive reactions and reducing the effect of blunders. The final check would tally up the reactions, which will determine whether you’ve succeeded or failed.
Let’s say your Persuasion is 3. You’re offered 3 arguments. The NPC will respond very favorably (+2) to argument #1, favorably to argument #2 (+1) and very negatively to argument #3 (-2). Your skill will modify these reactions to 4, 2, and 0. Let’s say the final check’s value would be 10, so assuming the conversation has 3 nodes with tagged lines, you’ll need to score at least 2 very favorable reactions and 1 favorable (or 3 very favorable ones) to pass the check. In longer dialogues you’d be able to fail a few times and still recover.
This system will maintain the importance of skills and encourage further investment but it will shift the focus to figuring out which lines would work best. Obviously, it might increase meta-gaming but that’s your choice and thus not our concern. Every time the player is offered to make a choice with different outcomes, 8 out of 10 people would want to know the outcomes in advance and the exact way to get to the outcome they want.
NPC text:
1. [skill 1] Dialogue option A
a. NPC very favorable reaction: +2
2. [skill 1] Dialogue option B
a. NPC negative reaction: -1
3. [skill 2] Dialogue option C
a. NPC favorable reaction: +1