There are several things you could do. Unless you make your model an ordinal model, I think it is not easy to capture the peaks around the different numbers which indeed, most people will gravitate towards. You could simply bin your data so that you do end up with just an ordinal scale from 1-7 and run an ordinal regression - then you will have the different peaks although you will miss where some people did not really answer the exact number. How important do you think it is to capture that, realistically?
Other options could include beta regression or what another user on the forum here has been developing @saudiwin - ordinal beta regression. This matches the kind of data you have well: the foum post about it is here (New paper using Stan/brms to estimate feeling thermometers/VAS scales), which includes links to the paper and a vignette using the approach in brms.
The lack of your model capturing the peaks in the data is not something the prior will solve - if you are estimating with a gaussian model then the predicted values will follow a gaussian distribution.