AI-Powered Search

AI-powered search is a cutting-edge approach to information retrieval that leverages artificial intelligence to enhance the accuracy, relevance, and thoroughness of search results. Basically, it’s how AI can find information from the internet and use it to give you better responses.

If you only used an LLM to get results, you’re limited to the knowledge cutoff date: the date that the model stopped receiving new data. If I were to ask GPT4o, which has a cutoff date of September 2023, who won the Superbowl in 2025, it would not have that information.

And unlike traditional search engines, like Google or Bing, which rely primarily on keywords and algorithms, AI-powered search uses advanced machine learning techniques to understand the context and intent behind a user's query. This means that AI-powered search can provide more personalized and insightful results, helping you find the information you need more efficiently.

Let’s look at an example. If I Google, “Superbowl LIX play by play” it will use the keywords and its algorithms to return the best results, which are links to other websites. I’ll only find a play by play if someone else has independently written one and published it online, and the keywords matched up with Google’s algorithms. I’d then have to read through the sites to find the relevant info.

On the other hand, if I give GPT4o the same prompt, it doesn’t even have info about who won the Superbowl, but it IS able to generate a play by play of a game if it had the info. That’s what LLMs are designed to do.

Putting the two together – LLMs and web searches, if I use the AI-powered search feature on top of GPT4o, and ask it to tell me about the Superbowl in 2025 in the form of a play by play recap, I can get a play by play, because it has access to the web, and it understands exactly what I'm asking – it isn’t just looking at keywords. It can also generate links that I can click on inside the chat – which includes email addresses.

Think of AI-powered search as a suped up hybrid between an LLM and a search engine, with an AI agent sitting in the middle, facilitating the web search and output generation.

One last important note: There is another way to search the internet in the chat : URL inputs. AI-powered search utilizes several steps to go out to the internet and find what you’re looking for based on the prompts you gave it, but the URL inputs give you the ability to ingest a specific page – or multiple specific pages.

For example, if I need to find out what the best platforms are for ecommerce, I would use the AI powered search to go out and search the internet for the products. It can click on links and explore new pages. But if I already knew two or three products I wanted to compare, I could grab those specific URLs and submit them to the chat.

However, the chat would only be able to ingest the information on those specific pages. It can’t click on any links or explore other pages.

In both cases, the LLM plays a role in generating results, but just keep in mind the limitations of URL inputs vs. AI-powered Search.

Now that we understand what AI-powered search is, let's talk about how it works. We’ll answer three more questions:

When you write a prompt in the chat and send it to the LLM with the AI-powered search enabled, the LLM gains this access to the internet. We send the prompt and part of your chat history to an AI Agent that has access to different search tools. Currently those tools are Brave Search, Firecrawl, and Perplexity.

These tools, when used by an AI Agent, provide a coordinated approach to searching, retrieving, and processing online information efficiently.

The actual tools may change over time – AI moves quickly – but the important part to know is that the AI Agent can access multiple tools at once, and then make a quick plan before it starts firing off operations.

The Agent can use a search engine, click through web pages, compile results, and then send data from the tools to the AI model you actually chose so it can generate a response based on your prompts. Think of the Agent as the mastermind behind the scenes, deciding which step to do and when, which tool to use for which step, and making autonomous decisions – all while you’re waiting for a response.

Let's look at it sequentially:

Try using the AI-powered search now and visualize all these steps happening as you watch the response stream in!

Module 3 focuses on real world use cases. Even though we just looked in-depth at how the AI-powered search works under the hood, it can be hard connect the dots to real life applications...to imagine exactly how you might use this feature at work. So in module 3 we’ll look at 3 use cases that you can try today:

AI as an Expert Teacher

In this use case, we're taking advantage of an LLM's ability to explain complex concepts in accessible ways, tailored to your level of understanding. This is great if you need to quickly learn about a subject or even a company, ahead of a presentation, or just to learn more about your profession.

Unlike static resources, an AI teacher can dynamically adjust explanations based on your questions and feedback. And with real-time access to the internet, it can give you much better answers – and with less chance of hallucinations (mistakes or fabricated data). It isn’t restricted to its training data; it can find the most up to date info in seconds.

Example prompt:

"You’re an expert in quantum computing and also teaching. Explain quantum computing as if I'm a software developer with no physics background, just a curious beginner in this field. Use the most current research and share links for further reading – but make sure they’re aimed at someone of my limited expertise. After your explanation, I'd like to ask follow-up questions."

Benefits to using AI as a teacher:

Things AI Can Teach You

In this use case, you leverage the AI's ability to gather, synthesize, and organize information from large datasets. Again, because the AI is performing live web searches, it can help you explore knowledge from beyond its training data, and also identify patterns and generate insights you might have missed. This is great if you need to scope out a prospect before a big meeting, or if you’re putting together a proposal or RFP response, and you need to gather data from multiple sources.

How to use it:

Example prompt:

"You’re a world-class researcher, a real digital detective. I'm researching the impact of remote work on organizational culture for a presentation on the big stage at a convention for HR professionals. Can you summarize the main findings from research on this topic, highlight any contradictory viewpoints, and suggest areas where the research might be limited?"

Benefits:

Things AI Can Research For You

AI as an Analyst

In this use case, the AI helps you interpret data, identify patterns, and draw meaningful conclusions. It can provide frameworks for analysis, suggest methodologies, and help you think through analytical approaches – while referencing data from live sources.

How to use it:

Example prompt:

"I've noticed our customer retention rates dropping over the past quarter. Can you help me think through possible causes, suggest metrics I should examine, and outline an analytical approach to determine the root causes? See if there are any similar trends being reported in our industry, and explore the possible causes."

Benefits:

What AI Can Analyze For You

By trying even one these three use cases in your AI Chats, you can immediately begin applying AI-powered search to enhance your learning, research, and analytical workflows, saving time and potentially uncovering insights you might otherwise miss.

So far, we’ve been focused on the Chat. But AI-powered search is also available in workflows! When you’re building a multi-step workflow, you can toggle the feature on or off for each individual step. For example, let’s say you have a 3-step workflow:

In step 2, the LLM is only able to give you a good quality output if it can go and investigate the current trends. Otherwise, it might resort to making up info (hallucinating) or giving a partial answer that isn’t very helpful.

In this case, you’d toggle the AI-powered search on for step 2, but it’s not needed for steps 1 and 3.

There are likely to be hundreds of workflow use cases that could benefit from access to the internet, guided by an AI Agent. Try coming up with some of your own!

Now you know what AI-powered search is, how it works, and what it can be used for. Here are some quick tips to get the most out of your searches.

Select the Right AI Model

Understand Model Strengths: Different AI models excel at various things. For example, smaller models might be faster and more efficient for straightforward tasks, while larger models can handle complex problems requiring deep understanding and can provide more nuanced responses.

Experiment and Iterate: Don’t hesitate to experiment with different models to see which one best suits your needs. Iterative testing can reveal which model provides the most accurate and contextually appropriate results.

You can find more info about LLMs in our Secure AI Chat course, or in this article.

Use Natural Language

This is no different than any other AI prompt engineering advice. Even when the AI has internet access, you can still use natural language – talk to it like you would talk to a person. This lets the AI interpret the meaning and context behind your question. Don’t worry about getting keywords exactly right, just ask the question.

Incorporate Current and Dynamic Terms

Include terms like “latest,” “current,” or “trending” to make sure the AI retrieves the most up-to-date information available online. For instance, "latest news on Mars rover missions."

Request Source Verification

Phrasing like "According to reliable sources..." or "Can you find peer-reviewed articles on..." ensures the AI prioritizes credible and authoritative information. You can also ask it to provide the clickable links of its sources, or even provide a works cited.