Ask Perplexity this question: “Which VCs are investing in robotics?”

It’s an interesting, relevant and timely question that generative AI should be able to answer. 

The results I got included Intel Capital, Google Ventures, Toyota AI Ventures and DFJ Growth.

Those aren’t hallucinations. They aren’t made up. They aren’t factually wrong. 

But they are also not accurate in the sense that they are not the answers most VCs would consider correct. 

And the reason for the gap is that the answers are being sourced from the kind of content marketing documents that are all over the internet. 

In this case, the sources for the question about VCs investing in robotics are Waveup; BaseTemplates, Fprimecapital, Cybernetix and Seedtable. 

BaseTemplates builds pitch decks. Waveup is a consulting firm of 25 located in London and Kyiv. They do fundraising, M&A advisory and growth strategy. Seedtable is a startup database. 

All of those firms are probably amazing at what they do.

But with the exception of Cybernetix they aren’t specifically focused on robotics. 

The reason they come up as sources is because they publish reports to try and raise their visibility. The reports are written in an SEO-optimized way that gets them picked up by search engines and therefore LLMs.

Perplexity makes this easy to see because the results include a “sources” option. Sometimes the LLMs use independent sources such as TechCrunch or Sifted but it’s usually some sort of content marketing. 

All of this matters because it’s one of the things holding back the usage of LLMs by companies. Gen AI is fine for individuals. But enterprises usually require a much higher degree of confidence in the answers. 

Once you realize how often LLMs use content marketing as their “source of truth” you cannot unsee it. 

It should terrify anyone working in corporate. 

I think this is one of the reasons that more and more companies are looking at options to create and gatekeep content that they know is accurate. 

My startup, Pricing Culture, is working on this problem with several clients. What we’ve done is build large libraries of text documents derived from accurate sources. We then allow agentic systems to access the libraries.

We did a proof-of-concept by summarizing blog posts from major VCs into a machine-readable format.

When we asked that database the same question about robotics, the results were quite different.

DM me if this is interesting and you want to compare our answers with the Perplexity results.