I’m doing a project to simulate various radioactive isotopes (with a HI-PROPE card and a BEAM card set as an isotope) placed 0.5 m away from a NaI detector and measuring DOSE-EQ using a USRBIN card while a RADDECAY card is set in semi-analogous mode. I’m aware that I cannot give activity to the FLUKA when Im in semi-analogous mode and I’m getting the output in pSv per primary/Bq. To find the dose rate I’m multiplying by the activity that the real scenario has. My question is, to get the most accurate results, should I give the value in START card equivalent to the activity?

For an example when my source has 95 MBq activity at start should I give 95E6 in START card to get most accurate results? (Which means should my simulation follow 95 million particles/decays)

The number of primaries used in a simulation has no physical meaning. You should use as many primaries as necessary to get a reasonable statistical error for your scoring.

So If I use 90 Million Primaries when my situation has 90 MBq activity, will I get a reasonably low statistical error? What if my situation has a very high activity source may be like 70000 MBq, simulating 70000 million primaries will take a lot of time so can I get a reasonable statistical error with like 100 Million primaries?

There is no correlation whatsoever between the number of simulated primaries and the radioisotope source activity, nor between the latter and the statistical error of your simulation. There is instead between the number of simulated primaries and the statistical error of your simulation, which depends on your geometry and the emitted radiation features.
Based on the resulting estimation of the statistical error, you should conclude whether you need (much) more primaries.