Large language models have a feature called Deep Research. It’s meant to improve the accuracy and reliability of queries by allowing more processing time. 

The quality of the results however remain constrained by the content being searched. Often that material is coming from content marketing documents. 

To illustrate the problem, Pricing Culture built a library of content about venture capital and partnered with MindStudio to add an agent. We call it Smart Research. 

The difference between Deep Research and Smart Research is that we vet and curate the source material that will be used to answer the questions. 

Let’s compare and contrast asking Deep Research and Smart Research this question: 

“I’m interested in the top VC firms investing in ai startups. Please produce a report listing VC firms actively investing in ai startups over the past 12 months along with a brief on each investment they have made.”

Perplexity’s Deep Research answered that the top VC firms focused on AI are Sequoia, A16z, Khosla, Index Ventures, Lightspeed and Coatue. 

It seems like a reasonable answer. It’s worth drilling in, however, to note that the sources were documents published on websites including Affinity.co, Techcrunch, Gilion and R&D World. It’s a mix of content marketing and news sources. 

None of it is coming from primary sources. 

We used a MindStudio agent to run a Smart Research query across content about deal announcements we gathered directly from 125 venture capital firms. 

According to those results, Andreessen Horowitz (a16z) was the most active AI investor, followed by Madrona, Charles River Ventures, 8VC, and Bain Capital Ventures.

You can compare the two here:

Other LLMs like Gemini and Grok also produce varying answers to this query, depending on both the model’s architecture and the source content it relies on.

What I think the experiment really underscores is that the library of content you query matters as much if not more than the model you use. 

Models will keep getting better. But for enterprise applications—where the cost of a wrong answer is higher than in consumer use cases—clean, curated datasets are likely to remain critical.

If you work at a VC and want to try the agent dedicated to VC blogs, podcasts and funding announcements please reach out. We have a limited number of 30-Day Free MindStudio Pro promotional codes you can try.