If you plan to add many buildings, I suggest using josm instead of id. It's a much smoother experience.

Usually, you take care of the "imagery offset" in the beginning of a mapping session. For this you use gpx tracks to align the different map layers. You can also use other shapes you already aligned properly in the past. With newer or different imagery it can happen that the imagery is slightly off.

For a building, you usually use the roof and move the final shape to a corner which is visible on the map. You may have to adjust things if the roof has a different shape than the building.

Yes, you move all buildings if all of them are off. This is easier done in josm than in id. You can also align the map to all the other buildings and hope that some day someone else will align it. Hint: this will be you, but in the future.

Other commenters have left some good advice already, just wanted to chime and say the word you’re looking for is “tilt” (when the image is not straight down, causing the sides of buildings to be visible).

@Tippon Imagery often needs to be re-aligned to match up with local traces. The wiki has some guidance on the general "Using Aerial imagery" pages and on handling off axis imagery on the page for roof modelling.

https://wiki.openstreetmap.org/wiki/Using_aerial_imagery
https://wiki.openstreetmap.org/wiki/Roof_modelling

@Tippon you can use https://osmuk.org/cadastral-parcels/ as an overlay and work off of property boundaries. It's as close to perfect as you can really get using public layers, and it's becoming a bit of a standard round my patch for realignment jobs.

I use it for aligning Bing layers to cope with the tilt, and it's still handy as an overlay after that for jobs like chopping terrace rows into individual houses