Sample size is independent of population size; this is a counter-intuitive result of statistics, but it is true. You can look up any polling from much larger countries like U.S. and you will see the samples for polls are between 1000-2000 people. To have a margin of error of ±3% with confidence level of 95%, you need 1067 samples, and this doesn't change with population size (technically it does, but for all intents and purposes, in elections it doesn't). 2400 will get you ±2%; to get to ±1%, you need 9604. As the difference between ±1 and ±2 is not big enough to warrant a huge increase of expenditure to increase the sample from 2400 to 9600, most go with something between 1000-2000, and get something close to ±2.5% margin of error.
This all assumes sampling is random. Sampling is never random (with people). That's why you get multiple polling companies; and they try different strategies to model a random sample. As a result, experienced polling companies will use information about past elections, demographic spread, rural voters, silent blocks, etc. to get as close to a random sampling as they can. This is why not all polls are created equal.
In Turkey, There are several well-known experienced non-partisan polling companies who have proven themselves in the past. There are also very transparently partisan ones that you can discard very easily as they are party polls (Avrasya and TAG for example are officially CHP partisan polls; on the other side you have AREDA (owned by Erdoğan's son-in-law's family) and also SONAR owned by Hakan Bayrakçı a known AKP partisan, and OPTIMAR, GENAR, TÜSİAR). These you should just discard if you're not after consoling yourself.
Another factor is polling companies that sprout up like mushrooms every election cycle, do one poll and then die out nowhere to be found again next election, giving their place to more brand new unknown entities coming up with anomalies. You discard these as well.
Now for a foreign observer, to be aware of all these nuances, different actors and their histories is almost impossible. This is why they say ignore polling in Turkey. There are also my compatriots who also say ignore polling, because they don't know any of these facts, statistical nuances or polling company details, have a cursory look at polls and when the results are not congruent, dismiss all polling as useless.
For someone who can read this situation, coming up with a reasonable prediction is very easy, since fortunately for a data obsessed person like me, Turkey is still dynamic enough of a society to have a slew of scientific independent polling companies with pedigree. And their polls are quite close to each other; and they have been close for the past elections as well, within the margin of error. This election, up until now, is outside the margin of error.
