Based on previous discussions on this forum, most people’s preference is to attempt to capture the location of the organism observed, instead of the location of the observer. For one such discussion, read down from here
There are some gray areas, but in any case, I would say try to use the minimum accuracy distance that you are confident encompasses the location of the organism.
For example, If you believe your frog was about a kilometer away in a certain direction, you could map the observation there, and include an accuracy distance of 1000 m (or whatever seems right to you).
Using a bounding circle - whatever the size - means my locations are … 95 % wrong? I don’t have GPS. I know it was along the trail.
A rectangle that I could rotate to ‘follow the trail’ would be the 5% right.
But somewhere on the forum is an earlier thread. The circle remains.
That’s what I do. I do tend to walk much further than most people, so I sometimes have accuracies expressed in kms. For most observations, I don’t think it’s a big deal. For important observations of rare/localized species, I will sometimes narrow it down. If it was a localized/important species, there’s a good chance I will remember where on the trail I saw it - and I can usually narrow that down using the satellite view on google maps (by referencing landmarks). I have a hiking GPS, but I only bother to carry it when I want to plot a very precise route (for a difficult to access location), or for instances where there’s danger of getting lost. Mostly, I just use the GPS on my phone - the app I use will store waypoints regardless of whether I have cell reception. It’s not as accurate as the hiking GPS, but it’s good enough for marking where I saw something significant. YMMV.
For the database I manage (which contains data from iNat, eButterfly, museum collections, etc.), we plot observations on maps using the Breeding Bird Atlas square (roughly 10km x 10km). To some extent, this helps with the imprecision of the data. We determine the square based on the Lat/Long, but we don’t show the Atlas user a “dot” at a precise location. For old observations where we don’t have a precise location, we can flag observations as “County only”, and those observations don’t appear on the “Squares” map. Not a great solution, but it allows us to keep the older records in the database.
I use iNaturalist data for Environmental Niche Analysis and Modeling (ENM). For global modeling iNaturalist data is better than GBIF’s because georeferencing is generally more accurate and photographs help to check taxonomy. Of course, tens and even hundreds of thousands of records cannot be verified. However, our analysis method allows to select only key records for niche mapping. This makes checking easier. Usually we have to check tens or hundreds of records, so it’s realistic. Yes - every time we find a few errors in georeferencing and taxonomy, but that’s normal.