At Tess AI, you can create custom agents for all sorts of tasks. But what if your agents could go beyond their initial training and search for up-to-date info straight from the internet? That’s exactly what the web data extraction AI Steps let you do.
This guide will show you how to turn your agents into smart researchers, able to find job openings, compare product prices on Google Shopping, get news, and much more, all in an automated and accurate way.
AI Steps are pre-configured steps you can add to your agent’s logic in AI Studio. They work like building blocks that give your agent specific skills, like transcribing audio, editing images, or, like we’ll see in this article, pulling data from the web.
The Web Data Extraction category groups together a set of steps designed to search different online platforms, like Google, Amazon, and LinkedIn. Instead of doing a generic internet search, these steps let you create super targeted and customized searches, making sure your agent finds exactly the info you’re after.
Step by Step: Setting Up an Agent with Web Search
Plugging a web search into your agent is easy. Just follow the steps below to get started:
Access Agent Studio: In Tess AI's side menu, navigate to Agent Studio. You can click "Add New" or "New” to start a new agent.
Add an AI Step: On the agent creation screen, find the AI Step option and click the button to add a new step.
Select the Category: A selection window will pop up. Choose the Web Data Extraction category, since our goal is to pull data from the internet.
Choose the Step Type: In this category, you'll find different search types. Options include, for example:
- Google Organic Search: For general searches, news, and info.
- Google Shopping Search: To search for products and compare prices.
- Google Jobs Search: To find job openings.
- Amazon Product Search: To look up products directly on Amazon.
And many more!
Set Up the Search: After you pick the step type, specific config fields will show up. Here’s where you tell the agent what to search for. For example, in a Google Shopping search, you’ll fill in the product name.
With these steps, the search feature will be ready. Now, the trick is to combine this step with a good prompt to guide the AI on how to analyze and show the results.
Practical Examples: Creating Smart Search Agents
The real magic happens when you combine a search step with a well-crafted prompt. Check out some real examples.
Example 1: Promotion Search Agent on Google Shopping
Imagine you want to create an agent that finds the best deals on a smartphone.
Step Configuration:
Step Type: Google Shopping Search
Product Search: iPhone 15 128GB
Domain and Language: Select Brazil and Portuguese.
Command Prompt:
The prompt tells the AI what to do with the search results. (Below, an example command. We recommend using more detailed prompts in your agents!)
"You are an expert in analyzing deals on electronics. Your task is to analyze the product search results (**search**). Identify the top three offers, considering price and store reputation. Give a clear summary of each offer, including the product link, and explain why they are the best choices."
When you run the agent, it will first search Google Shopping for "iPhone 15" (as in the step), and then use the prompt to analyze and format the answer the way you told it to.
Tips and Best Practices
To get the most out of your search agents, try out these tips:
The Prompt is the Brain of the Operation: The AI Step grabs the raw info from the internet. Your prompt is what turns that info into a smart, useful, and well-formatted answer. Be detailed with your instructions.
Use the Tess AI 5 Model: If you're not sure which LLM to use, go with Tess AI 5. It automatically picks the best model for the job—whether it's real-time searching or something more complex—so your answer will have the best quality.
Conclusion
Adding web data extraction AI Steps opens up a whole world of options for your agents. They stop being just static knowledge bases and become dynamic assistants, able to interact with the online world in real time.
By mastering the combination of research steps with detailed prompts, you'll be able to build powerful tools to automate market research, monitor news, find talent, and much more—saving time and getting more accurate, up-to-date insights.
