Sometimes I find images of comparitive study between CFD simulations and experimental results obtained in the CFD vendors websites.Do they do the CFd simulation and then perform experiment or just fit their numerical method to the experiment.?

If it is the latter case then how reliable can CFD simulation be in the design field?

Ah yes. The much misunderstood and under-appreciated craft of design with CFD .... It is of course very sound practice to check / calibrate / tune / train your design methods against known answers for a given class of problems. But then "DESIGN" does not consist of merely generating lots of answers for candidate designs and picking the best, whatever that may be. There is really a lot more to design than that.

I would say that the reliability of CFD in design has much more to do with the practitioner than with the code.