You record the amount of time it takes people to do a crossword puzzle or play a game of chess. After a while you'll be able to make a distribution graph of how long it takes. Then you can give an accurate estimate along with a probability.
The only reliable way to estimate is to find another relatively similar project and compare it to that. You can say stuff like this new project is roughly similar in scope to Project X but maybe it's about 20% more complicated due to more scope so it will probably take about 20% longer than what Project X took.
The key is to keep data on how long past projects actually took (which not a lot of organizations do). But once you have that real data, you can understand all the unknown unknowns that came up and assume that similar things will come up on the new project.
Do people writing this type of copy actually believe this?