1) Trusts their well-paid chat operators and allows them to type freely.
2) Force-feeds their operators 'canned responses' (as are used in email support) and doesn't encourage them to move off script.
#2 is always what marketing departments want, but it was always worse for the customer. The operator's response was too quick (a paragraph of text hastily selected from a recommended response), there were no misspellings, and they used PR speak that normal humans don't use. Often people would respond, "are you a person?" and some small group would ask questions specifically designed to trip up a bot.
My advice is to allow the people talking to your customer to use a modest amount of slang (a smiley or an occasional lol is fine if the customer is in a happy mood) - let them make a few misspellings - and encourage them to really read the response and even empathize with a customer. Marketing departments hate this idea, but in the end the feeling of a connection to another human is worth the risk of offending a grammar nazi.