The "security questions" you choose your answers, in "knowledge-based authentication" the questions and answers are generated based on the information on your credit reports and other databases, such as insurance databases. You don't make up answers, they already know the answers.
Wikipedia refers to "security questions" as "Static KBA".
So, for example, if you always went by "Ron Weasley", having your driver's license , your utility bills, your apartment lease, your animagus registration, and so on all under that name, but your name in their records was "Ronald Weasley", you would be out of luck.
Correct is too easy to guess.
Obviously wrong will be recognised by the customer service representative, who is a kind person, and will help you out by just disabling the question (anecdotes galore for this, to the point we can call it data).
Lie, in a believable way. Put the answer in a password manager. Open your password manager up before calling any CS rep. Be a star in the security theatre. :)
I look forward to password managers including generators for your birth date, family tree, childhood friends, favorite color and maybe also social security number, name and address if those aren't actually required for anything. Then you can have a value for your mother's maiden name that isn't incredibly common, but nonetheless looks like an actual person's name.
Q: Where were you on New Year's Eve in 1999? A: demand-exact-answer-over-phone
Depending on the sources I scanned on the first two pages of results, the cutoff is somewhere between November 2 and November 21.