They are substantially smaller in coverage, but have higher quality in my experience. Remarkably, they are also willing to correct their data if you notify them

They are substantially smaller in coverage, but have higher quality in my experience. Remarkably, they are also willing to correct their data if you notify them. This of course in is stark contrast to Google Scholar where the metadata of papers is frequently wildly inaccurate. On top of this, Semantic Scholar shares their underlying data (although you need to request an API key). Overall, they have been growing slowly and steadily over the years and I have a lot of respect for what their team is doing for researchers such as myself. Now for the less great. They are pushing the concept of "Highly Influential Citations" [1] as their default metric, which to the best of my knowledge is based on a singular workshop publication that produced a classifier trained on about 500 training samples to classify citations. I am a very harsh critic of any metrics for scientific impact. But this is just utter madness. Guaranteeing that this metric is not grossly misleading is nearly impossible and it feels like the only reason they picked it is because Etzioni (AI2 head) is the last author of the workshop paper. It should have been at best a novelty metric and certainly not the default one. [1]: https://webflow.semanticscholar.org/faq/influential-citations https://webflow.semanticscholar.org/faq/influential-citation... Recently, they introduced their Semantic Reader functionality and are now pushing it as a default way to access PDFs on the website. Forcing you to click on a drop down to access plain PDFs. It may or may not be a great tool, but it feels somewhat obvious that they are attempting to use shady patterns to push you in the direction they want. Lastly, they have started using Google Analytics. Which is not great, but I can understand why they go for the industry default. Overall, I use them nearly daily and they are the best offering out there for my area of research. Although, I at times feel tempted to grab the data and create an alternative (simpler) frontend with fewer distractions and "modern" web nonsense.