The answer to this depends an awful lot upon how far their 'interpersonal interaction' subroutines go.
For example: A robot that's very good at dealing with people in the short term may well be an excellent salesman, but the next week when that same customer comes back complaining that the thing they got doesn't work, the robot is going to have to refer them (very politely) to the human manager. If the robot is complex enough to realise that might be a problem and pre-empt it by offering a different product that better fits the customer's needs, then they will also likely be able to realise that once they start stealing everyone's jobs it's going to be much harder for them to interact with other people. This brings us to my first point:
If the robots are really, really good at this they'll be good enough to know when to be asshats.
Nobody really wants a 'perfect' social interaction. Heck, often people go out deliberately looking for a fight. If the robots are really, really good at interpersonal interaction they'll know when to be bad at interpersonal interaction in order to keep people happy overall for longer. In this scenario not much changes in the short term. People don't realise the robots are better at socialising than they are until the robots have already reached an equilibrium point with humanity (I'll come back to the 'equilibrium' later). This process will take so long that no humans will feel displaced or crowded out of their jobs, they'll just gradually come to accept that the robots are better than they are when it comes to bedside/telephone/wouldyoulikefireswiththat manner; because the robots are just that good at relating to everyone.
Of course, the robots aren't likely to be that good. They'll be better than us, but not good. If they're good but not perfect then they will know how to deal with complex interpersonal relationships with multiple people at once. This immediately lines up a much better set of work opportunities than simply being a call centre operator:
These robots will be amazing diplomats.
World peace will end up happening a lot faster. Once one diplomatic corps decides to use the very nice robots as their ambassadors all the others will too, which then leads to my comment about inter-robot communication. If the robots all know what their respective parties want and are willing to accept then a universal robotic communication protocol will cut otherwise arduous six month negotiations that get nowhere down to information exchanges measured in minutes that come up with elegant compromises, and then the respective robots will be able to convince their respective fleshy counterpart of why the ideas are good.
Some will also end up running call centres, much to the chagrin of the call centre employees, but I'm sure they'll quickly find far more fulfilling jobs, or the call centre robots will take pity on them and convince their bosses to hire some fleshy humans for old times sake.
People will still interact with people.
We won't interact with the robots all the time, because unless my first point is true they'll all be creepy as hell. Imagine that the only people you talk to all day are high class butlers. What would you give for just one of them to make a freudian slip and accidentally end up in bed with you five hours later?... OK, terrible example, but the point remains. We're social creatures, and unless the robots are good enough to realise that we're all doomed. Which leads me to my last point:
There will be a stable equilibrium point between 'all we do is talk to robots, then go extinct' and 'the robots are a terrible idea, burn them' This is the point where the desire to only talk to the robots (cos they're awesome) and the needs of the human race (like procreation) match up. The better the robots are at long term planning the faster and smoother we get there.
If the robots are so good (but lacking in planning) that we'd rather talk exclusively to them then human society will eventually fall due to everyone not meeting any actual fleshy people. Aside from the neo-hippies who refuse to have contact with machines (citing the apocalypse) and exist solely on home-grown soy products. Then their children will reactivate some of the robots and we'll go round the whole cycle again and again until we reach equilibrium (probably through legal ruling on robots or procreation).
If the robots aren't quite that good then we'll reach a point where people will have groups of friends that consist of both robots ('cos they're awesome) and humans ('cos they're friends of R-Bernard, who's a friend of R-Daneel, who's my friend, and actually they're not that bad comparatively, and I quite like the way that brunette laughs, and OMIGODTHEYJUSTWINKEDATMEDOESTHATMEANTHEYWANTSEX). The ratio of human/robot will correspond to just how many robots get made and how good they are at schmoozing.
I apologise for the above paragraph. Either way: Long term an equilibrium will be reached where robots do jobs that require good communication skills, but people will still put up with people because they either have to, the rest of the human race isn't actually all that bad, or the robots have realised they need to dial it back if they still want anyone to be interpersonal with.
Of course the equilibrium goes out of the window if the robots practice their skills on each other. Then you eventually end up with an entirely stagnant (but incredibly polite) race of automatons.