When you click into that dashboard to identify the biggest items, it has a bug where it combines the feature storage and photos. This may mean your '3 biggest' may be many GB but it's actually photos, and a lower-ranked item with <1gb of storage may be incurring 200x more credits in size
https://support.esri.com/en-us/bug/status-dashboard-for-credits-shows-combined-size-in-ite-bug-00014...
The credit consumption in your screenshot is correct but if you want to truly identify which items are consuming the most credits based on their size, you need to export the csv and break down both feature storage and attachments separately
https://esriaustraliatechblog.wordpress.com/2021/06/22/which-arcgis-online-items-are-consuming-the-m...
I have found in my experience it to be driven by:
- Large volume of complex geometry (polylines and polygons with lots of vertices). e.g. non-generalized state/country parcels. Rivers datasets for country.
- Editor tracking history. This grows substantially over time with large volume of editing.
- Replicas (but again, mostly tied to #1)