Since this still shows up in Google's top results for Golang random string generation, I wanted to share what I have been working with.
Here is the solution I am using:
package main
import (
  "fmt"
  "strings"
  "time"
)
var (
  opts  = strings.Split("option1,option2,option3", ",")
  start = time.Now()
)
func main() {
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
}
func getRandomOpt() string {
  len := len(opts)
  n := uint32(0)
  if len > 0 {
    n = getRandomUint32() % uint32(len)
  }
  return opts[n]
}
func getRandomUint32() uint32 {
  x := time.Now().UnixNano()
  return uint32((x >> 32) ^ x)
}
And results: 
option2 665ns
option1 41.406µs
option1 44.817µs
option3 47.329µs
option1 49.725µs
option3 52µs
option2 54.393µs
option2 56.798µs
option1 59.098µs
Source wise, I copied getRandomUint32() from fastrand: https://github.com/valyala/fastrand
And the solution proposed above. Performance isn't all that different, but I wanted to share results. 
package main
import (
  "fmt"
  "math/rand"
  "strings"
  "time"
)
var (
  opts  = strings.Split("option1,option2,option3", ",")
  start = time.Now()
)
func main() {
  rand.Seed(start.Unix())
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
  fmt.Println(getRandomOpt(), time.Since(start))
}
func getRandomOpt() string {
  return opts[rand.Intn(len(opts))]
}
And results: 
option3 11.865µs
option2 48.415µs
option3 52.809µs
option1 55.536µs
option3 58.191µs
option3 60.793µs
option1 63.391µs
option2 65.982µs
option2 68.601µs
These results were only run a few times locally and I grabbed what appeared to be the median result. There is certainly more work to be done in terms of iterations and whatnot, but I just wanted to share what I have.