Their elo is not reliable, but an estimate. It is very hard to make a strong player (as any boot from beginning) play bad (as most of the bots are supposted to). So it will typically mix moves to strong for its supposed elo with blunders to big for its supposed elo .
The elo system is also not reliable to measure two opponent playing each other again and again. They will of cource each get a precise elo, but it will not necessarily be the same elo, that they will get when facing other opponents. Some playingstyles match each other better or worse thant others. That is (part form the psychology) the reason why some players almost always beat certain other players of equel elo.
Premise: Obviously a bot is not comparable to a human player, I'm in fact talking about reliableness of the elo.
Said so, leaving your ego at the door: how much is reliable the elo of the bots compared to a player of similar elo?
Note that there is a reason why I told you to leave the ego away from this: because you have to consider that the Magnus Carlsen bot too could lose to the scholar's mate if the odds are in your favor (because of how the bots are programmed); beating the bot once doesn't mean you are better than it, beating it consistenly means you are better than it.
In other words: if you made an account playing with (dunno) the Willy bot, would he become hardstuck at 1800? Or would he become hardstuck way lower?