Including an analysis of the untypical DA profile of training sets created specifically for LLMs.
New data from @ziffdavis (owners of CNET, IGN, PCMag, Mashable and more) shows that AI companies are using more premium publisher data than they care to admit.
axios.com/2024/11/05/ai-trai… by @sarafischer via @axios
And naturally, I did my best to pick control periods for the brand:links ratio HCU results. Which produced the expected non-replication of HCU behaviour.
Full data: moz.com/blog/helpful-content…
But don't get too excited - a 12% gain in one update isn't worth much when you're 80% down YoY.
This is based on about 500 sites which are very clear HCU losers, btw.
Oh, and obviously, I took a look at how the losers from the last two HCU/combined updates faired in the most recent one, too.
I had a theory from the Google leaks: The HCU is not about content.
It's about brand:links ratio.
I've looked at the last 3 updates, ~2m keywords, and tens of thousands of sites, trying to prove it.
Results are a lot clearer than I expected.
Who is blocking AI bots, and why?
moz.com/blog/block-ai-bots xitter.cf/THCapper/status/181862…
