ChatGPT is NOT Google: Here's How to Actually Use AI for Search

One of the biggest myths I'm quick to dispel when using AI is that it's just like Google.

Traditional internet search is often the model that somebody applies during their first encounter with ChatGPT because we've become used to using Google as our primary gateway to the internet over the last 20 years. The application interface looks quite similar: a text input box that would invite a short sequence of space-delimited keywords.

However, you will have a disappointing experience if you treat tools like ChatGPT this way. You want to invert Google Search's best practices:?more is more. The more personal, precise, and voluminous context you can provide—especially in the context of a particular goal—the more valuable the initial output and subsequent interactions will be.

And, now that tools like ChatGPT and Claude have context windows of greater than 100,000 tokens in length (roughly equivalent to 10 or more hours of live spoken dialogue), they can become significantly more helpful when you arm them with relevant contextual data, sort of like onboarding a fractional expert with a massive onboarding document (except it's transmitted immediately).

All that said, while ChatGPT can't and likely won't replace traditional document-based internet search, there are interesting ways to use these AI tools to accomplish search tasks per se—but not in the rote information retrieval sense. Think of this type of search as, instead, venturing out into a physical area you've never been before, a dimly lit part of the map a few hours' detour from the well-trodden area where you've been doing your knowledge work.

In practice, what does this mean? Instead of searching for individual documents like you would with Google, I find it helpful to search for?individual thinkers?who have produced knowledge adjacent to the problem space in which I'm operating.

A practical way to do this is to work with AI as a thought partner, and once you've reached a point where you need help figuring out where to go next, you ask ChatGPT or Claude to imagine that it was a group dinner host. Direct it to invite five other credible (but extremely esoteric!) thinkers to discuss this very topic you're trying to make sense of at a private, off-the-record dinner where everyone could speak at length, challenge each other, and overall try to work with you in this problem space.

I recommend pushing the AI system to find the most esoteric thinkers it can because, initially, it will shortlist guests with more mainstream ideas. That's all well and good—and perhaps good enough, depending on your level of familiarity with this particular topic—but what's likely to be better is discovering someone with pertinent, original ideas that are underrated or undervalued in mainstream discourse because they have not received much attention (yet!).

At this point, you may seek out the seminal primary source documents they have produced (a Google Scholar search ranked by impact score or date may do the trick) and spend a little time chewing on those in the context of your problem. At that point, you're moving entirely outside the AI system, researching, reading, and "thinking naked."

Now, before doing that, if you want to try to make the most of the exercise, you can ask the AI system to model a vigorous debate between all the different simulated attendees by starting with the question, "What is one thing that you know to be true that everybody else at the table would disagree with you on?" (AKA the famous Peter Thiel question from?Zero to One).

What's valuable about this optional exercise is that it helps you grasp the diverse perspectives (and the gaps between them) that you and your dinner guests hold on the matter at hand). This context is valuable in understanding their primary source materials before delving deeper into their positions.

As I mentioned earlier, it's essential to understand that AI systems do not work like Google's search engine, which retrieves individual documents. However, they can be used to perform?a specific type of complementary search?that may help you know where to look for primary source documents. These documents can then be retrieved, analyzed, and made useful.

In summary, I recommend experimenting with AI systems to solve new problems and explore areas on the map of abstract information space (or "latent space" in AI terms) you haven't been before, starting with a simulated private dinner with esoteric thinkers who might make serendipitous contributions to your train of thought. Please keep an open mind when using these tools, understanding that you shouldn't expect to use them as a substitute for traditional search (or other media) but as complements or supplements.

At best, you might discover new ways of interacting with information that helps you accelerate progress toward your goals. And that's how innovation happens.