This topic is both complex and simple at the same time. It’s simple in the sense that having a finer mesh should, in theory, give you more accurate results. As your cells get smaller, the interpolations performed by the software are more consistent with the reality. As you refine your mesh more and more, at certain point the results should stop changing.
But this topic is also complex because, it’s not always easy to isolate the parameter number of cells. A lot of other factors can interfere: discretization schemes (order of schemes used, limiters?), quality of mesh (skewness, non orthogonality etc.), y+, local refinement versus evenly refining the whole domain…
If you would like to go more in depth on this topic, you can check out one of the most applied concepts when it comes to mesh independence studies. Kindly refer to Roache’s article where he proposed the concept of Grid Convergence Index (GCI): Perspective: A Method for Uniform Reporting of Grid Refinement Studies.