My first recommendation is to consider the GEDmatch graphics display when you do a one-to-one comparison as interesting, but not informative of detail. Keep in mind that you have almost 3 billion base pairs of nucleotides in those 22 chromosomes displayed, that only about 700,000 (or some 0.03%) have been tested, and that GEDmatch is trying to squish that into a visual bar graph that fits on your screen. The bars can't even be presented in proportion horizontally with each other because their relative size differences are too great. For example, look below each bar and you'll see the image size reduction for each chromosome: at the default display of 1,000 pixels wide, chromosome 1 is at a 1/44 scale, and chromosome 22, the smallest of the lot, is at 1/8 scale...so Chr 22 is shown 5.5 times larger than it really is in relation to Chr 1.
If you want to see the graphics, watch for the blue sections. Small splotches of green are common and almost always mean a segment is what's referred to as IBS: Identical By State. Some people instead call this Identical By Chance--which is more descriptive and explanatory, but isn't a standard, recognized term. Segments of substantial length that show in green are atypical. GEDmatch is showing these as Fully Identical Regions. I've linked it so you can read more because it takes a bit to explain. Valid FIR segments are typically seen only in siblings, double-cousins, and in trees from endogamous populations or that have significant pedigree collapse. And, simply because of the way current testing works, segments that look like they're FIR may not be. So in the vast majority of instances, it's only GEDmatch's blue-colored segments that matter, the ones that show as Half-Identical Regions (HIR) shared by both kits.
The matter of segment length. This subject is like standing up in the middle of an automobile show and shouting, "Which manufacturer makes the best cars?" You'll get a cacophony of answers which range in supporting evidence from none to some. The truth is, there have as yet been no scientific, peer-reviewed studies published on the subject of segment size as produced by current DNA testing, the methods of determining same (reference genome maps, linear equations used, and the significant differences in the rate of recombination between females and males), and the validity or lack thereof in genealogy.
I'm on the side of small segments being highly suspect. One place to start exploring the issue is one of the most well-known genetic genealogists, Blaine Bettinger. Fact of the matter is that somehow 7 centiMorgans became some sort of baseline, but no one has ever been able to explain to me exactly how that was pulled out of the hat. It's what GEDmatch defaults to, and what WikiTree policy say is satisfactory for relationship triangulations. Some not-scientifically-tested but empirical data suggests that 6cM segments are false 74% of the time; 7cM segments are false 58% of the time; and 8cM segments are false 38% of the time. Not "difficult to validate"; false. Phasing helps improve how low you can go and hope a segment size is valid, but we never get to see both base-pair-level detail and phasing information (which, if done, is phasing by genotype, not actual trio phasing)...anywhere: not at GEDmatch or any of the testing providers.
What Jim Bartlett, noted genetic genealogist--who some point to as the source for saying 7cM segments are sufficient--actually wrote is this:
"However, my triangulation process involves a lot of work... I highly recommend starting with 15cM as a threshold (or even higher, if you don’t have the time or inclination). Setting a personal threshold is a good way control the amount of work you are willing to put in."
Jim's point is also that working to establish valid evidence using small segments is very complex and difficult...if it can even be done. Some think that "triangulation" with autosomal DNA simply means finding two cousins with whom you, all three of you, share a small segment and you're done. Conversely, Jim, as an example, has said that it's typical for him to work with triangulation groups that have upwards of 20 people in them before he's willing to accept a 7cM segment as valid: "If 1 or 2 of those triangulated shared segments turns out to be IBS, it's not harmful in the grand scheme."
In choosing what to examine, IMHO there are two starting points. Either, "Cousin Jane Doe and I show as both descended from our shared N great-grandparents and I want DNA substation for that," or "I've found a seeming DNA match and I want to see if it is valid and fits in the family tree."
For the latter, I recommend looking at the two charts I linked earlier: one for the mathematical average DNA sharing, and one using the data from the Shared cM Project. The farther removed the relationship, the more difficult and time consuming to validate. And, in fact, to even consider more distant cousin DNA validation, you have to have built a dataset of closer cousins as a benchmark to understand how the family DNA fits together. A free utility to help with that effort is DNA Painter; others, like Genome Mate Pro, are also available.
If you're simply looking at reported matches and want to start exploring how they might fit into your genealogy, I strongly recommend doing it in steps. Start with the matches that show the largest amount of in-common DNA. I'd set my sights on GEDmatch by limiting the display threshold to 75cM; at that level the SNP density really won't matter. That will encompass some 3rd cousins and closer. Some you'll know right away where they fit in the family tree, and others won't respond to emails; but keep trying. Work this group first and start mapping the chromosomes so that you'll better understand if and where a more distant match fits into the family.
When you get down to 3C2R and 4th cousins (shared 3g-grandparents) you're getting into rarefied air where it takes significant detail and effort to validate the match. At that point you'll be roughly at around 15cM to 30cM total DNA shared. Looking under 10cM is not for the meek, and adequately validating the atDNA evidence for 5th cousins is not easy or quick. Only 14.9% of your 5th cousins are going to share any detectable DNA with you at all, so it takes luck as well as work. That puts the odds at 17:3...or for every 20 5th cousins of yours who take a DNA test, you can hope to match only three of them.
We all can use some luck! :-)