I would like to use RNA-seq data from the NIH GEO database. How can I tell if library generation for the RNA-seq experiments are stranded vs not stranded?
If you already have a reference transcriptome assembly, then you may run the Salmon tool for quantification using your RNA-seq samples. It has an automated library type inference function (https://salmon.readthedocs.io/en/latest/salmon.html#what-s-this-libtype). It might help you to determine whether the libs are stranded or not.
Unfortunately, the stranded nature of cDNA sequencing is not typically included in the metadata attached to a project, which means that downloading a small subset of the data and checking for strand consistency (as suggested by the other answer) is often the quickest way to work that out (when that information isn't included in the methods section of the associated paper). This knowledge is not particularly useful for casually browsing projects to find a stranded protocol that fits a particular study specification.
Sometimes you get lucky, but not often. GEO RNASeq datasets usually have links to the SRA accessions, and those accessions usually include annotation about the library type. For example, consider this dataset on GEO:
It links to SRP265460 on SRA. Following the link for the first run for this project, I get to here:
This includes the following metadata about the run:
Instrument: NextSeq 500
Construction protocol: Replicates of oropharyngeal explants at different stages of development (37/38, 39, 41) were immersed in RNALater. Samples were homogenized with a Dounce homogenizer in RLT buffer containing ß-mercaptoethanol (Qiagen). The homogenate was then processed using an RNAEasy kit (Qiagen, Hilden, Germany). Replicate samples for each stage were analyzed for RNA quantity and quality on a Nanodrop 8000 spectrophotometer. RNA libraries were prepared for sequencing using standard Illumina protocols
But, as you can see, there's nothing about whether its stranded or not: Illumina has both stranded and unstranded "standard" protocols.