Genspark started as an agentic search product and has grown into a workspace that generates research pages, presentations, spreadsheets and even makes phone calls on your behalf. That breadth is the pitch and it is also the problem, because a product doing eight things is rarely doing all of them well, and the learning material is not organised to help you find out which ones to trust. The agentic search behaviour is the most defensible part. Rather than returning a chat answer, it assembles a structured page on the topic with sections, comparisons and cited sources.
For an unfamiliar subject where you want an orientation rather than a single answer, the format is better suited than a conversation, and the difference is noticeable when you compare the two directly on the same question. Slide generation is the feature most people will actually use repeatedly. Give it a topic or a document and it produces a deck with a sensible structure and reasonable content, which is a real saving on the blank page problem. The output is a first draft rather than a finished deck, the design is generic, and the structure is usually sound enough that editing beats starting over.
That is a fair description of the value and not one the marketing makes. Source handling is where I would urge caution. Research output arrives with citations attached, which creates an impression of verification that is not always warranted. Links sometimes support the claim loosely rather than directly, and occasionally a summarised point does not survive contact with the source.
Checking is not optional, and the material presents the cited output confidently enough that people skip that step. Anyone using this for work that matters needs to click through, and the documentation should say so and does not. The learning material itself is the weakest element. It reads as feature promotion rather than instruction, telling you that capabilities exist without teaching you how to get good results from them.
There is no guidance on prompting for better research structure, no discussion of where each generator is weak, and no honest positioning against the alternatives. You learn this product by using it, and the documentation contributes very little to that process. Quality variance between the features is the practical issue. Search and slides are decent, sheets are hit and miss on anything structurally complex, and the more experimental features feel like demonstrations rather than tools.
A user reading the material would have no way to know that ordering, and would waste time on the weak features before finding the strong ones. A page saying what the product is actually good at would be the single most useful addition. Pace of change makes any guide temporary. Features appear, change and disappear quickly, which is normal for a young product and means external tutorials are unreliable and internal ones lag.
Treat anything you read as provisional and verify against the current interface. Three point two. A broad product with several genuinely useful capabilities buried among several unfinished ones, supported by learning material that is closer to a brochure than a manual. The free credits are enough to work out what is good here, which is fortunate, because the documentation will not tell you.