Not all GEO datasets will necessarily have links to SRA with raw data. In some cases you will get just processed expression values (e.g., FPKMs, etc).
The many platforms that fall under 'Expression profiling by high throughput sequencing' should generally be for expression estimation of RNA and will therefore be variants of what is commonly called RNA-seq. However, some may be more specialized, incorrectly categorized, or represent datasets with mixed data types. For example, they might involve sample/library preparations designed to assay just small non-coding RNAs, or other special cases.