What led me to try my hand at domain name speculation was curiousity about a particular angle on the market. Domaining had always seemed boring in practice, despite my theoretical interest in it as kind of a stock market for words. This angle that I'd thought of was to focus on domains composed of two three-letter words: names like Zipcar, GitHub, and HeyZap.
I liked these "3,3" domains because they're punchy, memorable, and balanced, but the space is big enough that good ones might still be available. By looking systematically at all the domains matching this pattern, I hoped to find and buy some for later use or resale. Of course, I didn't know if anyone had tried this same angle before me &mdash if they had, it would be much less effective. The evidence on this point turned out to be inconclusive. I ended up with some good but not great domains. Here's how.
I started with a list of all the three-letter words in English. There are about six hundred. After trimming out the ones that subjectively seemed low-quality ("aal", "cwm"), I generated a database of all the possible 3,3 domains. That's about 300,000 combinations, from aceace.com, aceact.com, and aceadd.com all the way to zoozip.com, zoozit.com and zoozoo.com.
Then I wrote a script to check their availability. When the script finished, the database showed about 100,000 taken names, and 200,000 available ones. Those numbers show that this space has both a large portion of good names, and a large portion of available names. The question now was whether those two subsets had any overlap. I had to sift through the database and see which domains, if any, I wanted to register for myself.
Evaluating the 200,000 available names by hand was out of the question, so I had to find a better way. My first thought was to subjectively evaluate each of the 500-or-so three-letter words for quality (box: good, nub:poor), then value each 3,3 domain by the the combined quality of its halves. If these quality ratings were numeric, a program could do the combining. I considered using people from Mechanical Turk to augment or replace my own ratings.
Then, I realized that a word might be better as a prefix than as a suffix (like "the"), or vice versa (like "ism"). This was twice the work, but might produce significantly better rankings.
Then, I had the aha moment that made this whole endeavor worthwhile for me — I could use the implicit data! Instead of manually rating the words (as prefixes or suffixes), I could see how many domains in my database with a given prefix or suffix were already registered. For example, if 33% of words starting with "ace" were registered, I could assign "ace" a rating of .33 as a prefix. This empirical evidence of what's valuable immediately seemed more reliable than my (or anyone else's) subjective appraisal. This was a doubly great insight because it was both right and saved me a lot of effort. In just a few lines of code, I had quality ratings for all 300,000 of the 3,3 domains.
So, what did it come up with? Here are the top ten:
|Top Ten "3,3" Domains
These are all domains that I would be very happy to own. I'm sure they aren't precisely the ten most valuable 3,3 domains on the market, but the methodology seemed good enough. Of, course, these names are all taken. What could I actually get? (At the time of this experiment.)
|Top Ten Unregistered
An unimpressive lot. The first thing I noticed was that the first available domain name was #691, and only a few were available in the top thousand. With numbers like those, it's unclear whether or not someone had used this same market angle before me. Either way, not thrilling. I did like dogcon, and bizlab was probably worth getting despite being a little sleazy. (Though it would turn out that bizlab.com — my program's best prospect — had been registered by someone else while my domain-checker script was still going through the list!)
Looking further down the list of unregistered domains, I was pretty ready to call this experiment a bust. But, I spent some time exploring the database further, focusing on prefixes and suffixes that I liked, and ended up buying a few, like dogcon.com. These domains aren't precious gems, but I could see them being useful to me or someone else.
And anyway, it was a fun experiment. Worth trying once, like a deep-fried candy bar at the county fair. If you're inspired, you could try looking at the 3,4 or 4,4 domains (Dogpile, Facebook) and see what you can find. Even if you're just looking for a name for your own new website, I think this could be an effective way to go about it.